<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Davi on Medium]]></title>
        <description><![CDATA[Stories by Davi on Medium]]></description>
        <link>https://medium.com/@bluekoins?source=rss-98bb79fdfe35------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*XC6naWq7B-tGXjKBko8ACA.png</url>
            <title>Stories by Davi on Medium</title>
            <link>https://medium.com/@bluekoins?source=rss-98bb79fdfe35------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 03:51:13 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@bluekoins/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Putting Open Source AI in Your Pocket]]></title>
            <link>https://medium.com/@bluekoins/putting-open-source-ai-in-your-pocket-cc150c724d31?source=rss-98bb79fdfe35------2</link>
            <guid isPermaLink="false">https://medium.com/p/cc150c724d31</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[innovation]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[large-language-models]]></category>
            <dc:creator><![CDATA[Davi]]></dc:creator>
            <pubDate>Fri, 12 Jan 2024 11:12:18 GMT</pubDate>
            <atom:updated>2024-01-12T11:12:18.861Z</atom:updated>
            <content:encoded><![CDATA[<p>Ollama + Quantisation = AI That Can Chat on Your Laptop</p><figure><img alt="A group of llama data scientists teaming up to symbolize open source collaboration." src="https://cdn-images-1.medium.com/max/1024/1*LxQWcgfzeQSNe1Hd4ZGsdw.png" /><figcaption>Image Midjourney V6 | David Kolb</figcaption></figure><p>In my previous post, I walked through quantising the 7 billion parameter Llama-2 model down to 4-bit precision to reduce its size for local deployment on a consumer laptop. This compressed the model from 30GB down to 3.8GB while retaining high accuracy.</p><p><a href="https://medium.com/@kolbeuk/shrinking-giants-adapting-open-source-ai-for-everyday-devices-5dacbe63ced6">Shrinking Giants: Adapting Open Source AI for Everyday Devices</a></p><p>Now, I will demonstrate deploying this 4-bit quantised Llama model locally for chat applications using Ollama. By importing my optimised model into Ollama and running inferences, we can see first-hand the performance gains of quantisation.</p><p>As demonstrated in this post, Ollama provides flexibility to deploy optimised models from external sources by importing a custom quantised model. It’s also worth noting that Ollama offers several pre-optimised models available out of the box for even easier on-device deployment. However, I could fine-tune the precision and performance trade-offs for my specific hardware constraints by quantising and importing an external model myself.</p><p>How do you run a quantised model locally?</p><p>The answer is using ollama.ai. Ollama is a tool to get up and running locally with large language models, and it’s simple to use.</p><p>For MacOS Download Ollama from this link</p><pre>https://ollama.ai</pre><p>Create a local direction to store your quantised models</p><pre>mkdir models</pre><p>In this example, we will use the quantised model from the last blog post. <br>Llama-2–7b-chat-hf-Q4_K_M.gguf</p><p>Create a file called ModelFile with a From instruction.</p><pre>FROM Llama-2–7b-chat-hf-Q4_K_M.gguf</pre><p>There are other parameters which you can use. For more details, check “<a href="https://github.com/jmorganca/ollama">https://github.com/jmorganca/ollama</a>&quot;</p><p>Create the model in Ollama.</p><pre>ollama create Llama27bchat -f Modelfile</pre><p>Run the model</p><pre>ollama run Llama27bchat</pre><p>Here, you have a simple chat interface for the prompt and res</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*IsABKDuO0r8NwgkqDfv3AQ.png" /></figure><p>You can also access Ollama as a server call.</p><pre> r = requests.post(<br> &quot;http://0.0.0.0:11434/api/chat&quot;,<br> json={&quot;model&quot;: model, &quot;messages&quot;: messages, &quot;stream&quot;: True},<br> )</pre><p><strong>Conclusion</strong></p><p>By combining model quantisation techniques with user-friendly tools like Ollama, the latest AI conversational capabilities can now be run on standard consumer hardware like laptops.</p><p>This allows personal use cases to benefit from innovations previously only available through cloud APIs. Performance is more than adequate for applications like chatbots, question-answering, semantic search, etc.</p><p>Improvements around model optimisations and tools simplifying the experience would further increase the adoption of on-device AI. But this shows the potential for democratised access, as barriers around computing constraints and complexity get solved through quantisation and wrappers like Ollama.</p><p>Interested in the intersection of Generative AI and retail? Share your thoughts in the comments below,<a href="https://www.linkedin.com/in/david-kolb/"> or reach out for a deeper discussion.</a></p><p><a href="/service/mailto:david@davidkolbconsultancy.com">david@davidkolbconsultancy.com</a></p><p>More Links</p><p><a href="https://github.com/jmorganca/ollama">GitHub - jmorganca/ollama: Get up and running with Llama 2, Mistral, and other large language models locally.</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=cc150c724d31" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Shrinking Giants: Adapting Open Source AI for Everyday Devices]]></title>
            <link>https://medium.com/@bluekoins/shrinking-giants-adapting-open-source-ai-for-everyday-devices-5dacbe63ced6?source=rss-98bb79fdfe35------2</link>
            <guid isPermaLink="false">https://medium.com/p/5dacbe63ced6</guid>
            <category><![CDATA[large-language-models]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[innovation]]></category>
            <category><![CDATA[open-source]]></category>
            <category><![CDATA[technology]]></category>
            <dc:creator><![CDATA[Davi]]></dc:creator>
            <pubDate>Wed, 10 Jan 2024 14:38:41 GMT</pubDate>
            <atom:updated>2024-01-16T15:57:41.425Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="A llama stands atop GPU stacks, overseeing a tech environment, blending whimsy with modern computing in a surreal depiction." src="https://cdn-images-1.medium.com/max/1024/1*R_0l3WoRNQXofjZOYA1jdA.png" /><figcaption>Image Midjourney V6 | David Kolb</figcaption></figure><p>2023 saw an explosion of open-source AI with notable releases from Meta’s Llama2 and the French startup Mistral.Ai’s Mixtral 8x7B. While democratising AI access, these giant models still pose infrastructure and cost challenges for many due to their substantial computing and memory requirements.</p><p>To address this barrier, several tech firms like Google and Apple are finding ways to train or compress these models so they can run efficiently on consumer-grade hardware. For example, Google recently introduced Gemini Nano. At the same time, Apple released its MLX framework, which utilises quantisation to enable speech recognition on Macs via OpenAI’s Whisper model.</p><p>Quantisation compresses large language models by simplifying the detailed numerical values representing the model’s learned knowledge. For example, a number like 0.12345678 might be approximated as 0.125. This makes the overall model representation less precise but crucially allows it to take up far less computer memory and run faster with no internet connection required.</p><p>So, in essence, quantisation trades off some marginal accuracy to enable models that would typically require expensive specialised hardware to now run smoothly on consumer devices like laptops and phones instead. The models still perform remarkably well but can now readily fit into applications on ubiquitous personal devices and platforms.</p><p>It’s analogous to artists working with a small colour palette — while they may convey the essence of an image, the subtleties and realism are missing compared to if that artist had a full spectrum of colours available.</p><h4><strong>Use Cases</strong></h4><ul><li>AI on Consumer Devices: Enable real-time processing on phones/laptops without constant internet dependency</li><li>Startup Innovation: Develop cost-effective prototypes and minimum viable products powered by state-of-the-art AI</li><li>Research Tools: Support active learning and experimentation with quantised models on typical consumer devices</li></ul><p>To see how this works, I ran an experiment using the Open Source project called Llama C++ (Georgi Gerganov). I chose Meta’s Llama2 model because it represents a leading open source conversational AI now freely available for research under a permissive license, while pre-quantised Llama2 models are emerging, I quantised the model myself for greater transparency and control to tune for my MacBook Pro’s constraints.</p><p>The overall goal is to compress an Open Source AI model for feasible deployment on a MacBook Pro M1 16 GB while preserving accuracy as much as possible.</p><h4><strong>Prerequisites</strong></h4><p>A free Hugging Face account to download models.</p><pre>https://huggingface.co</pre><p>Accept Meta’s Llama license terms and get access approval before downloading the models.</p><pre>https://ai.meta.com/resources/models-and-libraries/llama-downloads/</pre><h4>Quantisation Steps</h4><ol><li>Set up GPU environment on Runpod for quantisation acceleration.</li><li>Download 7B parameter Llama 2 model from HuggingFace.</li><li>Configure Llama C++ toolkit.</li><li>Quantise Llama to intermediate format FP16.</li><li>Quantise to 4-bits for maximum compression.</li><li>Run inference.</li><li><strong>Set up GPU environment for quantisation acceleration</strong></li></ol><p>While the end goal is to run compressed models on a MacBook Pro, quantising large AI models requires significant one-time compute resources for acceleration. For this quantisation exercise, I leveraged GPUs rented from RunPod’s cloud service, as they provide necessary performance without involving personal data.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ml9jHn057fkUK76dW9RTfQ.png" /><figcaption>Screenshot of RunPod Instance Status</figcaption></figure><p>Note — the larger the model, the more GPU memory is needed to quantise it. For this experiment, the GPU setup with an NVIDIA RTX6000 (48GB VRAM, 40GB container volume, 900GB disk volume) cost approximately $5 on RunPod. The service enabled convenient data transfers between cloud storage providers to stage the models.</p><pre>https://www.runpod.io</pre><p>Once the GPU environment was deployed, I could access Jupyter notebooks or the SSH terminal to run the quantisation.</p><p><strong>2. Download 7B parameter Llama 2 model from HuggingFace</strong></p><p>It can take a while to download the models, so my tip is to start this before you set up the Llama C++ toolkit and let the download run in the background.</p><p>Install the huggingface_hub library</p><pre>pip install huggingface_hub</pre><p>Generating Filenames for Different Model Formats</p><pre># Define the model ID for Hugging Face<br>model_id = &#39;meta-llama/Llama-2-7b-chat-hf&#39;<br><br># Extract the model name from the model_id by splitting the string<br>model_name = model_id.split(&#39;/&#39;)[-1]<br><br># Append &#39;-f16.gguf&#39; to the model name for the FP16 formatted filename<br>filename_f16 = model_name + &#39;-f16.gguf&#39;<br><br># Append &#39;-Q4_K_M.gguf&#39; to the model name for the 4-bit model filename<br>filename_q4_0 = model_name + &#39;-Q4_K_M.gguf&#39;</pre><p>Run this code to download the model</p><pre>from huggingface_hub import snapshot_download<br><br># Retrieve your Hugging Face token from: Hugging Face <br># Profile -&gt; Settings -&gt; Access Tokens<br>token = &lt;HFTOKEN&gt;<br><br><br># Use snapshot_download to get the model<br># Specify the local directory, symlink usage, revision, cache directory<br># and token<br>snapshot_download(repo_id=model_id, <br>                  local_dir=model_name,<br>                  use_symlinks=False, <br>                  revision=&quot;main&quot;, <br>                  cache_dir=&quot;/workspace/hf/&quot;, <br>                  token=token)</pre><p>Note that use has the cache_dir=”/workspace/hf/” parameter to create the cache on your workspace volume, not the default container volume.</p><p>Once downloaded, you can manually delete the HF cache to save space.</p><pre>cd /workspace<br>rm -rf hf</pre><p><strong>3. Configure Llama C++ code</strong></p><pre># Clone the llama.cpp repository from GitHub<br>git clone https://github.com/ggerganov/llama.cpp<br><br># Navigate to the cloned repository directory<br>cd llama.cpp<br><br># Install required Python packages from the requirements.txt file<br>pip install -r requirements.txt<br><br># Compile the llama.cpp with GPU acceleration enabled<br>make LLAMA_CUBLAS=1</pre><p><strong>4. Quantise Llama to Floating Point 16-bit — intermediate format before 4-bit integer quantisation.</strong></p><p>This step is required as quantising gradually in multiple steps (e.g. 32-bit -&gt; 16-bit -&gt; 4-bit) allows for better accuracy and training stability compared to direct quantisation (32-bit -&gt; 4-bit).</p><pre>import os<br><br># Model file name as defined earlier, <br># In this example the model was download to &#39;/workspace/Llama-2-7b-chat-hf&#39;<br><br># Construct the command to convert the model to FP16 format using llama.cpp<br>command = f&quot;python llama.cpp/convert.py {model_name} \<br>  --outfile llama.cpp/models/{filename_f16} \<br>  --outtype f16&quot;<br><br># Execute the command using os.system<br>os.system(command)</pre><p>LLama quantisation to 16-bit floats completed in approximately 10 minutes. Once quantisation finished, the message appeared, indicating the quantised model file was written to the volume.</p><pre>Wrote llama.cpp/models/Llama-2–7b-chat-hf-f16.gguf</pre><p><strong>5. Quantise Floating Point 16-bit to 4 bits to maximise compression</strong></p><p>I am using the Q4_K_M quantisation scheme as experiments on HuggingFace have proven it strikes a good balance of compression and accuracy given my 16GB memory constraint.</p><pre># Construct the command for quantizing the model.<br># The &#39;./quantize&#39; script is used to convert the FP16 model file <br># to a 4-bit quantized version.<br>command = (<br>    f&quot;./quantize ./models/{filename_f16} &quot;<br>    f&quot;./models/{filename_q4_0} &quot;<br>    &quot;Q4_K_M&quot;<br>)<br><br># Execute the quantization command using os.system<br>os.system(command)</pre><p>Quantisation to 4-bit was completed in approximately 10 minutes. Once quantisation is finished, you get this message.</p><pre>llama_model_quantize_internal: model size  = 12853.02 MB<br>llama_model_quantize_internal: quant size  =  3891.24 MB</pre><p>Original 32-bit Model Size: Size: 30 GB<br>FP16 Model Size: Approximately 12.55 GB.<br>Quantised 4-bit Model: Size: Approximately 3.80 GB.</p><p><strong>6. Run inference</strong></p><p>To run inference on the quantised Llama model and test its performance, we utilise the ./main executable provided in the llama.cpp toolkit.</p><p>This tool allows passing in the compressed model file and input text prompts to generate language predictions. Some key parameters we can specify are:</p><pre>-m - Path to quantized model file<br>-p - Input prompt text<br>--color - Return colored text<br>-c - Number of token contexts<br>--temp - Temperature sampling<br>--repeat_penalty - Repetition penalty<br>--n </pre><p>For example, the following command runs the 4-bit compressed Llama model to generate a short intro based on the given blog title:</p><pre>./main -m ./models/Llama-2-7b-chat-hf-Q4_K_M.gguf \<br>       -p &quot;Write a short intro to this blog: Shrinking Giants&quot; \ <br>       --color -c 4096 --temp 0.7 --repeat_penalty 1.1</pre><p>I first tested inference performance on the Runpod GPU instance before transferring the quantised models to my local MacBook Pro to run. This ensured the toolkit and models were working properly before deployment.</p><p>I downloaded the model file to my MacBook using SCP. RunPod provides a cloud sync option to transfer between RunPod and AWS S3, Google Cloud, DropBox, Azure Blob or BackBlaze B2.</p><pre>scp -r username@runpod-hostname:transfer/ \<br>       models/Llama-2-7b-chat-hf-Q4_K_M.gguf \<br>       ~/Downloads</pre><p>On the MacBook Pro M1 16GB. The 4-bit quantised Llama model achieved fast inference performance on consumer hardware. Some highlight metrics include:</p><ul><li><strong>1.4 seconds</strong> to load the model</li><li><strong>73 ms</strong> average sampling time per prompt</li><li><strong>12.59 tokens/sec</strong> for text generation (faster than human read speed)</li><li><strong>11.9 seconds</strong> total runtime</li></ul><p>By quantising down to 4 bits, the model size was reduced enough to allow responsive inference on a personal laptop with output generated at about 13 tokens per second. <strong>This rate was actually faster than I could read the text</strong> — demonstrating good enough throughput for local applications.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*lYcFGXLCbc2vP5ve3CijpQ.png" /><figcaption>Screenshot from MacBook Terminal</figcaption></figure><p>In summary, through quantisation we attained a compressed model that runs efficiently on devices like a MacBook Pro with minimal accuracy loss.</p><p><strong>Conclusion</strong></p><p>In summary, this project demonstrated how quantisation can make large models accessible beyond the cloud, how complementary cloud and local device strategies maximise strengths, and the importance of responsible governance. These themes could drive much of the open source AI dialogue moving forward.</p><p>Here are three key takeaways</p><ol><li>Quantisation enables Open Source AI models to run efficiently on consumer devices, expanding access. By compressing models down from 32 bits to 4 bits, quantisation reduced the Llama model by 87% with minimal accuracy loss. This allowed fast beyond reading speed inference on a MacBook Pro.</li><li>Combining cloud and local devices is an impactful approach. Leveraging cloud GPUs for the resource-intensive one-time quantisation step and deploying the compressed model locally provides a best-of-both-worlds solution. This balances capabilities, cost and privacy.</li><li>Responsible governance is pivotal as AI proliferates. Initiatives like the EU’s AI Act, which promotes research while ensuring accountability, will help guide development. Structured oversight of open source projects enables innovation on top of an ethical foundation.</li><li>Quantisation shows promise to enhance AI efficiency yet remains in the early research stages. Start initiating small-scale pilot projects or experiments with quantisation for hands-on learning and understanding of how quantisation can be integrated and what benefits it may offer.</li></ol><p><a href="https://medium.com/@kolbeuk/putting-open-source-ai-in-your-pocket-cc150c724d31">In my next post,</a> I will showcase running the quantised Llama-2 model on a laptop for chat applications. This will demonstrate the real-world viability of deploying optimised open source AI on everyday consumer devices.</p><p>Interested in the intersection of Generative AI and retail? Share your thoughts in the comments below,<a href="https://www.linkedin.com/in/david-kolb/"> or reach out for a deeper discussion.</a></p><p><a href="/service/mailto:david@davidkolbconsultancy.com">david@davidkolbconsultancy.com</a></p><p><strong>More links to explore.</strong></p><ul><li><a href="https://www.davidkolbconsultancy.com/">David Kolb - Human-Centric Innovation Consultant</a></li><li><a href="https://github.com/ggerganov/llama.cpp">GitHub - ggerganov/llama.cpp: LLM inference in C/C++</a></li><li><a href="https://towardsdatascience.com/quantize-llama-models-with-ggml-and-llama-cpp-3612dfbcc172">Quantize Llama models with GGUF and llama.cpp</a></li><li><a href="https://huggingface.co/docs/optimum/concept_guides/quantization">Quantization · Hugging Face</a></li><li><a href="https://github.com/ml-explore/mlx">GitHub - ml-explore/mlx: MLX: An array framework for Apple silicon</a></li></ul><iframe src="https://drive.google.com/viewerng/viewer?url=https%3A//arxiv.org/pdf/2305.17888.pdf&amp;embedded=true" width="600" height="780" frameborder="0" scrolling="no"><a href="https://medium.com/media/1c8ed07a738f127cfcc422641a65fbef/href">https://medium.com/media/1c8ed07a738f127cfcc422641a65fbef/href</a></iframe><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5dacbe63ced6" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[An Experiment in Personalised Onboarding]]></title>
            <link>https://medium.com/@bluekoins/an-experiment-in-personalised-onboarding-512cc759e2f2?source=rss-98bb79fdfe35------2</link>
            <guid isPermaLink="false">https://medium.com/p/512cc759e2f2</guid>
            <category><![CDATA[generative-ai-tools]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[digital-transformation]]></category>
            <category><![CDATA[innovation]]></category>
            <dc:creator><![CDATA[Davi]]></dc:creator>
            <pubDate>Mon, 04 Dec 2023 07:01:55 GMT</pubDate>
            <atom:updated>2023-12-04T16:36:05.413Z</atom:updated>
            <content:encoded><![CDATA[<p>Personalised Onboarding for Today’s Diverse Global and Remote Teams</p><figure><img alt="Artificial Intelligence integrated in employee onboarding process depicted in a small, friendly and cozy office environment" src="https://cdn-images-1.medium.com/max/1024/1*12-Bj0_ss04kMbQzPYiNCQ.jpeg" /><figcaption>Image David Kolb DALL-E 3</figcaption></figure><p>The onboarding process represents a critical phase in welcoming new team members and integrating them into an organisation’s culture and values. With the increasing complexity and diversity of global and remote teams, the challenge lies in personalising this experience for each new hire while maintaining efficiency and consistency across the organisation.</p><p>This blog post explores an innovative solution to this challenge: leveraging OpenAI’s Assistants API to transform the onboarding experience.</p><p><a href="https://medium.com/@kolbeuk/quick-tip-demystifying-generative-ai-a-guide-to-text-and-image-generation-21c41609d8c6">Learn more about generative AI terminology and large language models.</a></p><p><strong>Enter Assistants</strong></p><p>The Assistants API Beta from OpenAI is a tool to craft custom AI helpers tailored to tasks akin to a personalised ChatGPT. It extends the dynamic capabilities of the recently announced OpenAI GPTs. The Assistants API Beta allows firms to integrate AI-driven tools into their existing platforms.</p><p><strong>Potential Use Cases</strong></p><ul><li>Customer Service: Assistants could handle high-volume customer queries across channels, providing fast, consistent responses by referencing knowledge bases.</li><li>Research and Analysis: Assistants could rapidly synthesise data into actionable intelligence, bringing efficiency and scale to document analysis.</li><li>Process Automation: Assistants could automate repetitive, rules-based tasks across operations, increasing efficiency and allowing employees to focus on high-value activities.</li></ul><p><a href="https://towardsdatascience.com/navigating-the-large-language-model-landscape-f56bd67e48d9">Learn more about large language model adoption.</a></p><p><strong>Purpose of this Experiment</strong></p><p><em>How might we leverage AI assistants to efficiently generate personalised, strategically-aligned onboarding communications while preserving authentic human connection?</em></p><p>This experiment aims to generate a customised onboarding letter for a new employee. This leverages existing company knowledge and utilises information extracted from internal documents to create a tailored, efficient, and informative introduction for new team members.</p><p><strong>Summary of how assistants work</strong></p><ul><li>Files provide assistants with the necessary contextual knowledge to generate accurate responses.</li><li>Assistants are AI agents who understand and respond to user queries by leveraging files and models.</li><li>Threads maintain continuity between user and assistant interactions.</li><li>Messages facilitate conversations, enabling users and assistants to exchange information.</li><li>Runs allow assistants additional processing time for complex queries before responding.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*SwBzkl8zAAniT8ADsAlEwg.png" /><figcaption>Possible Levels of Personalisation for Onboarding Assistants</figcaption></figure><p><strong>Pre-setup</strong></p><p>Developed a fictional company with its history, strategy, values, culture, and team leader personas using ChatGPT-4.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ytRoJLytD_FXiCuwjRbnTQ.png" /><figcaption>Fictional “Innovatech” Company Overview</figcaption></figure><p>Install the openai libraries</p><pre>pip install — upgrade openai</pre><p>Get the <a href="https://openai.com/blog/openai-api">OpenAI API key</a> and create the environment variable (MacOS).</p><pre>export OPENAI_API_KEY=’your-api-key-here’</pre><p>Import the libraries and create the client (Python)</p><pre>from openai import OpenAI<br>client = OpenAI()</pre><p><strong>Assistant Setup</strong></p><ul><li>File Creation: Prepared PDF files containing company strategy, values, culture, and leader personas.</li><li>Customisation and Attachment: Configured and personalised the assistant, attaching files. Included specific instructions for the assistant, such as generating onboarding letters.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/509/1*4XNxbXTos5TJgs3XMGo4Mw.png" /><figcaption>Basic contents of the assistant</figcaption></figure><p>This custom JSON structure in this experiment holds the file details.</p><pre>[<br> {<br> “file_name”: “teamleaderpersona.pdf”,<br> “file_purpose”: “assistants”<br> },<br> {<br> “file_name”: “companyhistory.pdf”,<br> “file_purpose”: “assistants”<br> },<br> {<br> “file_name”: “companyvalues.pdf”,<br> “file_purpose”: “assistants”<br> },<br> {<br> “file_name”: “companystrategy.pdf”,<br> “file_purpose”: “assistants”<br> }<br>]</pre><p>Read the files and upload them to OpenAI. Note if you repeat this step, you will have duplicate files, albeit with different file_ids. Possibly a beta version error.</p><pre>files = client.files.list()<br><br># Initialize an empty list to store the file IDs<br>file_ids = []<br><br># Loop through the files and extract the file IDs<br>for file in files.data:<br>    file_id = file.id<br>    file_ids.append(file_id)<br><br># Print the list of file IDs<br>print(&quot;List of File IDs:&quot;)<br>print(file_ids)</pre><p>List of files uploaded to OpenAI.</p><pre>File ID: file-npuP7NS0TmORKM7EShPbzVIi<br>File Name: companystrategy.pdf<br>File Purpose: assistants<br> — — — — — — — — — — — — — — — <br>File ID: file-OPFe3bUaX7luuCisiB69RO5Q<br>File Name: companyvalues.pdf<br>File Purpose: assistants<br> — — — — — — — — — — — — — — — <br>File ID: file-7R7ldtRFb7r0OHb2vZoSKbzV<br>File Name: companyhistory.pdf<br>File Purpose: assistants<br> — — — — — — — — — — — — — — — <br>File ID: file-cCQK87YWveOoMo8RnHacO2H9<br>File Name: teamleaderpersona.pdf<br>File Purpose: assistants<br> — — — — — — — — — — — — — — — <br>Total Number of Files: 4</pre><p>Create the assistant. You can choose an initial prompt (instruction) and the model type. I used the gpt-4–1106-preview.</p><pre>instruction = &quot;&quot;&quot;You are a helpful letter writer. Your job is to write <br>warm and personable onboarding letters welcoming a new employee to the <br>company. The letter briefly touches upon the company&#39;s history <br>and core values, emphasizing a sense of belonging and alignment <br>with the company&#39;s ethos.&quot;&quot;&quot;<br><br>assistant = client.beta.assistants.create( <br>    # Name of the Assistant.<br>    name=&quot;onboardingassistant&quot;,<br><br>    # Description of what the Assistant does.<br>    description=&quot;Tailors personalized onboarding experiences for new hires&quot;,<br><br>    # Here, version of the GPT-4 model.<br>    model=&quot;gpt-4-1106-preview&quot;,<br>    <br>    # Main prompt for all interactions with the asssitant<br>    instructions = instruction,<br><br>    # List of tools enabled for the Assistant.<br>    tools=[{&quot;type&quot;: &quot;retrieval&quot;}],<br>    file_ids=file_ids<br>)</pre><p><strong>Operating the assistant</strong></p><ul><li>Thread Attachment: Initiated a conversation thread with the assistant, establishing a continuous dialogue framework.</li><li>Message Creation: Composed messages to input into the thread, which can include additional files (like new hire details) and further instructions.</li><li>Prompt: Provided a prompt to the assistant specifying the desired response or action.</li><li>Run: Attach thread to the assistant to process and respond, effectively putting the system into operation.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/758/1*K4NXUHxQ1C4a9Ox4eiiWhQ.png" /><figcaption>Single message flow to the assistant</figcaption></figure><p><strong>Create a thread</strong></p><p>A thread is similar to a conversation.</p><pre>thread = client.beta.threads.create()</pre><p>Create a message. This is attached to the thread via the thread.id</p><p>Create a prompt ‘content’ to create a welcome letter for the new head of product.</p><p>Attached the persona of the new head of product as a file.</p><pre>message = client.beta.threads.messages.create(<br>    # Specify the ID of the thread to which this message belongs.<br>    thread_id=thread.id,<br>    # Define the role of the message sender. <br>    role=&quot;user&quot;,<br>    # Content of the message.<br>    content=content,<br>    file_ids=[fileid]<br>)</pre><p>Here is the output from the message before the run command. You can tie this back to the Message_ID and the Thread_ID.</p><p>The File IDs: [‘file-UcqlrjLgGzuQycbxgvOk5NpL’] is the new file with the head of content persona.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*RLRf1HJdAcPiS7i3SaG1FQ.png" /><figcaption>The message sent to the assistant</figcaption></figure><p>Submit the thread to run.</p><pre>run = client.beta.threads.runs.create(<br>    # Specify the ID of the thread that the run will process.<br>    thread_id=thread.id,<br>    # Provide the ID of the Assistant that will handle the thread.<br>    assistant_id=assistant,<br>    # Give specific instructions for this run.<br>)</pre><p>The run step is queued in OpenAI. To find out if your run has been completed, repeat this step.</p><pre>run = client.beta.threads.runs.retrieve(<br>  thread_id=thread.id,<br>  run_id=run.id<br>)<br>print(run)</pre><p>The code will return a status message, continuing several fields, including these. Once the ‘completed’ field has a time stamp, the run is completed.</p><pre>cancelled_at=None, <strong>completed_at=1700932243</strong>, created_at=1700932209, <br>expires_at=None, failed_at=None,</pre><p>This code extracts several details, including the main content (The letter).</p><pre>for message in messages.data:<br>    # Print the details of each message<br>    print(&quot;ID:&quot;, message.id)<br>    print(&quot;Object Type:&quot;, message.object)<br>    print(&quot;Created At:&quot;, message.created_at)<br>    print(&quot;Thread ID:&quot;, message.thread_id)<br>    print(&quot;Role:&quot;, message.role)<br><br>    # Print the content of the message<br>    for content in message.content:<br>        if content.type == &quot;text&quot;:<br>            print(&quot;Content:&quot;, content.text.value)<br><br>    # Print file IDs, if any<br>    print(&quot;File IDs:&quot;, message.file_ids)<br><br>    # Print the Assistant ID<br>    print(&quot;Assistant ID:&quot;, message.assistant_id)<br><br>    # Print the Run ID<br>    print(&quot;Run ID:&quot;, message.run_id)<br><br>    # Print any metadata<br>    print(&quot;Metadata:&quot;, message.metadata)<br><br>    print(&quot;\n----------------------------------------\n&quot;)</pre><p><strong>Results</strong></p><p>The assistant successfully produced a personalised, strategically aligned onboarding letter for the fictional new Head of Product, Sophia Martinez.</p><p><strong>Key highlights</strong></p><ul><li>Warmly welcomed Sophia while aligning her expertise with Innovatech’s innovation-focused values.</li><li>Blended company history and goals with details customised to Sophia’s background.</li><li>Achieved an enthusiastic yet professional tone that valued Sophia’s capabilities.</li><li>Incorporated customised information about Sophia’s role, interests, and location.</li><li>Accurately reflected company values around innovation, sustainability, and inclusivity.</li></ul><pre>ID: msg_oLYpr2W9sovf7wrjg5WrtVBW<br>Object Type: thread.message<br>Created At: 1700932225<br>Thread ID: thread_2ZT1U5nqIxVnDAle1BNjl8OU<br>Role: assistant<br>Content: Dear Sophia Martinez,<br><br>It is with great enthusiasm that I, Alexa Taylor, extend a heartfelt welcome<br>to you as the new Head of Product at Innovatech Solutions Ltd. As someone <br>who shares our passion for technological advancements and sustainability, <br>your expertise in product strategy and dedication to innovation will be <br>invaluable to our team.<br><br>Innovatech Solutions has a rich history that dates back to 2010, when our <br>visionary founders established a startup aimed at revolutionizing technology <br>applications across industries. Embracing innovative and user-friendly <br>designs, we quickly expanded our offerings into financial services, retail, <br>and hospitality, growing to become a tech leader known for pushing boundaries <br>to enhance business and consumer experiences worldwide【23†source】.<br><br>Our company values and culture are built around innovation, which echoes <br>your background in creating user-centric product designs and your proactive <br>role in women-in-tech initiatives. We believe in providing technology <br>solutions that are not only efficient and effective but also responsible <br>in their impact on society and the environment. This ethos aligns with <br>your own commitment to fostering educational and technological access <br>through community involvement.<br><br>Strategically, Innovatech aspires to be at the forefront of the tech <br>industry, transforming operations and customer interactions within our <br>target markets. Our focus on the development of tailored technology <br>offerings, sustainable practices, and continuous innovation is synergistic <br>with your experience in agile and lean methodologies, UX/UI design, and <br>data analytics. We are confident that together we can further our aim to <br>lead and redefine the role of technology in the fields we serve【27†source】.<br><br>I am excited about the fresh perspectives you will bring to our team, <br>especially given your capacity for leadership and strategic thinking. <br>Your proven ability to drive product vision while ensuring exceptional <br>user experiences aligns perfectly with our strategic goals, and I am eager <br>to see how your insights will drive our company forward.<br><br>Once again, welcome to Innovatech Solutions. I look forward to working <br>closely with you to achieve new milestones and foster a culture of innovation <br>and excellence.<br><br>Warm regards,<br><br>Alexa Taylor<br>Head of Operations<br>Innovatech Solutions Ltd.<br>File IDs: []</pre><p><strong>Environment clear down</strong></p><p>Since this is a beta version. To avoid unknown costs in OpenAI, delete the files and the assistants. Further tests are needed on the requirement to delete threads and messages.</p><p>Delete Files</p><pre>files = client.files.list()<br>user_confirmation = input(&quot;Do you want to delete all assistants? Type &#39;yes&#39; to confirm: &quot;)<br>if user_confirmation.lower() == &#39;yes&#39;:<br>    for file in files.data:<br>        file_id = file.id<br>        client.files.delete(file_id)<br>        print(f&quot;Deleted file with ID: {file_id}&quot;)<br>else:<br>    print(&quot;Deletion cancelled.&quot;)</pre><p>Delete Assistants</p><pre>my_assistants = client.beta.assistants.list(<br>    order=&quot;desc&quot;,<br>    limit=&quot;20&quot;,<br>)<br>print(&quot;List of Assistants:&quot;)<br>for assistant in my_assistants.data:<br>    print(f&quot;- {assistant.name} (ID: {assistant.id})&quot;)<br><br>user_confirmation = input(&quot;Do you want to delete all assistants? Type &#39;yes&#39; to confirm: &quot;)<br><br>if user_confirmation.lower() == &#39;yes&#39;:<br>    for assistant in my_assistants.data:<br>        response = client.beta.assistants.delete(assistant.id)<br>        print(f&quot;Deleted {assistant.name} (ID: {assistant.id})&quot;)<br>else:<br>    print(&quot;Deletion cancelled.&quot;)</pre><p><strong>Challenges during the experiment</strong></p><ul><li>Beta Stage Drawbacks: There were inconsistencies in how well the API worked.</li><li>File Buildup: Duplicate files accumulated over time and needed manual removal of unneeded ones.</li><li>Storage Expenses: Additional costs may come with storing files.</li><li>Inaccurate Answers: Sometimes, the responses were wrong. I needed to directly reference files like ‘companystrategy.pdf’ in the prompt to improve the responses.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*BRvYdCNh604D4MCstveUXQ.png" /><figcaption>Duplicate files but unique file IDs</figcaption></figure><p><strong>Conclusion.</strong></p><p>This experiment demonstrated capabilities in leveraging AI to provide the personalised onboarding experiences needed to welcome today’s diverse global and remote teams.</p><ul><li>Successful integration of customisation while maintaining organisational identity</li><li>Positioning aligned with onboarding best practices in setting context</li><li>Areas needing improvement, like file management and output inconsistencies</li><li>Initial confirmation that leveraging OpenAI assistants shows promise for this application</li></ul><p>Further testing could uncover more potential for personalisation and automation in HR communications. However, refinement is still required to enhance accuracy and efficiency sufficiently for enterprise viability.</p><p><strong>Striking the Balance</strong></p><p>Utilising AI for crafting onboarding letters presents a complex and nuanced challenge. While AI has strong capabilities to produce tailored content, realising the personalisation is machine-generated can detract from authenticity and resonance. This highlights balancing customisation with human connection in AI communications — striking the appropriate balance is vital, especially for sensitive applications like onboarding.</p><p>As we advance AI adoption, integrating it to complement human interactions is crucial — particularly for communications. Future iterations should emphasise gathering user feedback to guide the development of AI that drives efficiency while resonating authentically.</p><p><strong>Privacy Considerations</strong></p><p>In this experiment, all data, including company information and employee details, was synthesised and did not correspond to actual individuals or confidential company data.</p><p>For real-world applications utilising OpenAI’s services, it is crucial to consult OpenAI’s privacy policies to understand how your data will be handled and protected.</p><p><strong>Links</strong></p><p>OpenAI assistants overview: <a href="https://platform.openai.com/docs/assistants/overview">https://platform.openai.com/docs/assistants/overview</a></p><p>OpenAI assistants API: <a href="https://platform.openai.com/docs/api-reference/assistants/createAssistant">https://platform.openai.com/docs/api-reference/assistants/createAssistant</a></p><p>David Kolb: <a href="http://www.linkedin.com/in/david-kolb/">www.linkedin.com/in/david-kolb/</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=512cc759e2f2" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Decoding the Herb and Spice Aisle with Generative AI]]></title>
            <link>https://medium.com/@bluekoins/decoding-the-herb-and-spice-aisle-with-generative-ai-9603370adbc9?source=rss-98bb79fdfe35------2</link>
            <guid isPermaLink="false">https://medium.com/p/9603370adbc9</guid>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[python]]></category>
            <category><![CDATA[generative-ai-tools]]></category>
            <category><![CDATA[innovation]]></category>
            <category><![CDATA[computer-vision]]></category>
            <dc:creator><![CDATA[Davi]]></dc:creator>
            <pubDate>Sun, 03 Dec 2023 17:19:59 GMT</pubDate>
            <atom:updated>2023-12-04T16:17:49.119Z</atom:updated>
            <content:encoded><![CDATA[<p>How CV2 Segmentation and the GPT-4 Vision API Boosted Product Recognition</p><figure><img alt="An intricate supermarket aisle captured on an iPhone 12 Max, filled with a diverse array of products ranging from fresh produce to packaged goods." src="https://cdn-images-1.medium.com/max/1024/1*k_O7-1IHDSCjRTV0H28anw.jpeg" /><figcaption>Image David Kolb Dalle 3</figcaption></figure><p>Have you ever wandered the grocery store’s busy herb and spice aisle, searching in vain for a particular jar buried among the lookalike labels? I recently faced that mundane yet frustrating dilemma far too often. That prompted an idea — could ChatGPT4 simplify ingredient searches</p><p>ChatGPT4 iOS has demonstrated impressive capabilities in understanding images and identifying objects within them. However, when locating a specific product — Lemongrass — on a shelf, ChatGPT4 struggled, incorrectly placing the lemongrass and providing inaccurate descriptions.</p><p>To enhance the analysis, I created an experiment using openCV, an open-source computer vision library for Python. By segmenting the dense shelf image into separate sections, I could feed ChatGPT more targeted and simplified images to analyse.</p><h3><strong>Use Cases</strong></h3><p>The broader applications of this technology are vast and varied. From aiding visually impaired individuals in shopping to retail inventory management, the implications extend into various domains, including e-commerce and supply chain optimisation.</p><h3><strong>Approach</strong></h3><ul><li>Take a photo of a complex store setting on an iPhone 12 Max.</li><li>Perform image Segmentation using OpenCV.</li><li>Analyse the images with GPT-4 Vision Preview API.</li></ul><h3><strong>The setting</strong></h3><p>The original iPhone photo of the densely packed shelves with the visual complexity of jars, containers, and packages of all shapes crowded together.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*HHhGrmIBIhoOPSaRSoeE6w.jpeg" /></figure><h3><strong>Setup</strong></h3><p>Install the openai libraries.</p><pre>pip install openai</pre><p>Install the openCV libraries</p><pre>pip install opencv-python</pre><p><strong>OpenCV for Image Segmentation.</strong></p><p>Use OpenCV to split the image into six segmented images. Building in a 10% overlap between segments aimed to comprehensively cover all products without missing details that could undermine the vision analysis.</p><pre>    image = cv2.imread(image_name)<br><br>    # Calculate the dimensions of each square<br>    height, width, _ = image.shape<br>    square_height = height // rows<br>    square_width = width // cols<br><br>    # Calculate the overlap size in pixels<br>    overlap_height = square_height * overlap_percentage // 100<br>    overlap_width = square_width * overlap_percentage // 100</pre><p>The segmentation strategy, including the number of segments and the extent of overlap, can be adjusted based on the characteristics of different images.</p><pre>    for i in range(rows):<br>        for j in range(cols):<br>            y_start = i * (square_height - overlap_height)<br>            y_end = (i + 1) * square_height<br>            x_start = j * (square_width - overlap_width)<br>            x_end = (j + 1) * square_width<br><br>            sub_image = image[y_start:y_end, x_start:x_end]<br><br>            # Save the sub-image to disk<br>            sub_image_filename = f&#39;sub_images/sub_image_{i}_{j}.jpg&#39;<br>            cv2.imwrite(sub_image_filename, sub_image)</pre><p><strong>GPT-4 Vision API</strong></p><p>The GPT-4 Vision API handles multiple image inputs compatible with base64 encoded strings and image URLs. For this experiment, base64 encoding was utilised.</p><pre>  def encode_image(image_name):<br>    with open(image_name, &quot;rb&quot;) as image_file:<br>      return base64.b64encode(image_file.read()).decode(&#39;utf-8&#39;)</pre><p>Each image was sorted and base64 encoded to prepare them for processing. Images were stored locally to enable multiple accesses and processing at various stages of the experiment.</p><pre>    encoded_images = []<br>    # Get a list of filenames in the directory and sort them alphabetically<br>    image_files = sorted(os.listdir(image_directory))<br><br>    # Loop through the sorted filenames<br>    for filename in image_files:<br>        if filename.endswith(&quot;.jpg&quot;): <br>            image_name = os.path.join(image_directory, filename)<br><br>            # Encode the image using the encode_image function<br>            base64_image = encode_image(image_name)<br>            encoded_images.append(base64_image)</pre><p>Set up the connection with OpenAI’s API, with the type of data being sent and the unique access key.</p><pre>    messages = []<br><br>    headers = {<br>        &quot;Content-Type&quot;: &quot;application/json&quot;,<br>        &quot;Authorization&quot;: f&quot;Bearer {api_key}&quot;<br>    }</pre><p>A prompt is included in the API request. My prompt was:</p><p><em>“In these images, please specify in which image I can find the lemongrass and provide a detailed description of the product and its location.”</em></p><pre>    text_message = {<br>        &quot;role&quot;: &quot;user&quot;,<br>        &quot;content&quot;: [<br>            {<br>                &quot;type&quot;: &quot;text&quot;,<br>                &quot;text&quot;: custom_text_message,<br>            }<br>        ]<br>    }</pre><p>Add each of the six base64 encoded images to the API message.</p><pre>    # Iterate through encoded images and add image messages<br>    for base64_image in encoded_images:<br>        image_message = {<br>            &quot;role&quot;: &quot;user&quot;,<br>            &quot;content&quot;: [<br>                {<br>                    &quot;type&quot;: &quot;image_url&quot;,<br>                    &quot;image_url&quot;: {<br>                        &quot;url&quot;: f&quot;data:image/jpeg;base64,{base64_image}&quot;<br>                    }<br>                }<br>            ]<br>        }<br>        messages.append(image_message)</pre><p>Create the API payload, including the prompt and images.</p><pre>    # Create the payload with the list of messages<br>    payload = {<br>        &quot;model&quot;: &quot;gpt-4-vision-preview&quot;,<br>        &quot;messages&quot;: messages,<br>        &quot;max_tokens&quot;: 300<br>    }</pre><p>Send the request to OpenAI.</p><pre>response = requests.post(<br>    &quot;https://api.openai.com/v1/chat/completions&quot;, <br>    headers=headers, <br>    json=payload<br>)</pre><p>After the model processed the request, this code retrieves and displays the response.</p><pre>print(response.json())</pre><h3><strong>Results</strong></h3><p>GPT-4 Vision API’s response successfully details the location of lemongrass. It’s described as being in dried form, typically utilised as a culinary flavouring in Asian cuisine for its distinctive, citrus-like flavour. The lemongrass is stored on the second shelf from the bottom, housed in cylindrical containers with greenish-yellow labels.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/560/1*q_qwPfCdheOaZzaZGe9E7A.jpeg" /><figcaption>Zoom in on the Lemon Grass Location</figcaption></figure><p>The labels are marked with “LEMONGRASS” in bold black text against a light backdrop, next to a black-labeled product called “Palm Sugar.” The analysis also captures other elements, including various spices above and packages of dried mushrooms below the lemongrass.</p><p>The lemongrass’s price tag is also mentioned, further affirming the product’s identification. This detailed breakdown from the API showcases its capability to effectively recognise and describe products in a complex visual setup combined with CV2 image segmentation.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*lMAXXbKCxUSFwzVPgKVjiQ.png" /><figcaption>JSON from the GPT-4 Vision API’s response</figcaption></figure><pre>API request took 16.05 seconds<br>{&#39;id&#39;: &#39;&lt;removed&gt;&#39;, &#39;object&#39;: &#39;chat.completion&#39;,<br>&#39;created&#39;: 1700243878, &#39;model&#39;: &#39;gpt-4-1106-vision-preview&#39;, <br>&#39;usage&#39;: {&#39;prompt_tokens&#39;: 7030, &#39;completion_tokens&#39;: 217, <br>&#39;total_tokens&#39;: 7247}, &#39;choices&#39;: [{&#39;message&#39;: {&#39;role&#39;: &#39;assistant&#39;, <br>&#39;content&#39;: &#39;Lemongrass can be found in the last image you uploaded. <br>Here is a description and location:\n\nDescription:\nThe product in question <br>is lemongrass, which appears to be in dried form. Specifically, it is <br>&quot;Waitrose Cooks\&#39; Ingredients Lemongrass.&quot; This is commonly used as a <br>flavoring in Asian cuisine, providing a distinctive citrus flavor to dishes <br>without the tartness of actual lemon.\n\nLocation:\nThe lemongrass is <br>located on the second shelf from the bottom. There are two cylindrical <br>containers with greenish-yellow labels. Each container is labeled clearly <br>with &quot;LEMONGRASS&quot; in black text on a light background. The containers are <br>positioned next to a black-labeled product called &quot;Palm Sugar.&quot; Above the <br>lemongrass, you can see various spices and below, there are packages of <br>dried mushrooms. The price tag visible below the lemongrass indicates that <br>it costs £1.80. The shelf tag located directly under the lemongrass\&#39;s <br>shelf also appears to indicate the product\&#39;s name, reinforcing its <br>identification.&#39;}, &#39;finish_details&#39;: {&#39;type&#39;: &#39;stop&#39;, &#39;stop&#39;: <br>&#39;&lt;|fim_suffix|&gt;&#39;}, &#39;index&#39;: 0}]}<br>127.0.0.1 - - [17/Nov/2023 17:58:05] &quot;POST / HTTP/1.1&quot; 200 </pre><h3><strong>Key Takeaways</strong></h3><p>This experiment underscores the synergy between computer vision and large language models in practical AI applications. The successful identification of a specific item in a complex retail setting paves the way for more nuanced and sophisticated uses of Generative AI in retail and other settings.</p><ul><li>Combining CV2 image segmentation with GPT-4’s vision API enables more accurate identification of products in densely packed retail settings.</li><li>There is significant potential to transform retail experiences by integrating computer vision and large language models. Applications could range from improving customer service to optimizing inventory tracking.</li><li>Generative AI innovations often necessitate multidisciplinary approaches, like blending OpenCV computer vision libraries with neural network capabilities.</li></ul><p>Interested in the intersection of Generative AI and retail? Share your thoughts in the comments below, <a href="https://www.linkedin.com/in/david-kolb/">or reach out for a deeper discussion.</a></p><h3>Links to Code</h3><p>OpenCV2 <a href="https://pypi.org/project/opencv-python/">https://pypi.org/project/opencv-python/</a></p><p>OpenAI Vision <a href="https://pypi.org/project/opencv-python/">https://pypi.org/project/opencv-python/</a></p><p>David Kolb <a href="http://www.linkedin.com/in/david-kolb/">www.linkedin.com/in/david-kolb/</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9603370adbc9" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Prototyping Open Source Speech Recognition and Translation]]></title>
            <link>https://medium.com/@bluekoins/prototyping-open-source-speech-recognition-and-translation-ca571b3c4293?source=rss-98bb79fdfe35------2</link>
            <guid isPermaLink="false">https://medium.com/p/ca571b3c4293</guid>
            <category><![CDATA[llama-2]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[large-language-models]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[openai]]></category>
            <dc:creator><![CDATA[Davi]]></dc:creator>
            <pubDate>Sun, 03 Dec 2023 16:50:18 GMT</pubDate>
            <atom:updated>2023-12-04T11:03:27.165Z</atom:updated>
            <content:encoded><![CDATA[<p>Exploring OpenAI’s Whisper and Meta’s Llama2 7b integration on a MacBook Pro M1.</p><figure><img alt="Illustration depicting the transformation of sound waves into digital text, symbolizing the speech-to-text conversion process in advanced AI technology." src="https://cdn-images-1.medium.com/max/1024/1*v1E2iXPpyD0YPZ-8eeVj9Q.jpeg" /><figcaption>Image David Kolb Midjourney</figcaption></figure><p>Speech-to-text technology has become vital for accessibility and efficiency. However, privacy remains a rising concern as reliance on cloud services grows. This heightens the need for secure solutions that uphold confidentiality over sensitive data.</p><p>Open-source large language models present a viable, secure alternative to traditional cloud services, prioritising data privacy and user control at comparable, if not better, performance.</p><p>This experiment tests the feasibility of using large language models for translation on a MacBook Pro.</p><p>I used Openai Whisper ‘tiny’ and Llama2 7b to match the MacBook’s performance capabilities. However, this comes with translation accuracy and depth trade-offs compared to the larger Llama2 70b and Whisper’ large’ model, reflecting a necessary compromise in this experiment.</p><h3><strong>Use Cases</strong></h3><p><strong>Non-Profit Organisations</strong><br>Non-profits can benefit from models like Whisper and Llama2. By recognising speech and translating languages locally and privately, sensitive voice data remains under client control at a marginal cost. This allows non-profits to better serve diverse linguistic communities through enhanced, accessible tools.</p><p><strong>Privacy-Centric Applications</strong><br>For areas where privacy is paramount, like healthcare and legal, on-device large language models enable secure localisation of sensitive conversations — vital for compliance and safeguarding proprietary data. Firms maintain more granular control around data access policies, combining security for sensitive information with cost-effectiveness.</p><h3><strong>Approach</strong></h3><ul><li>Setup ollama.ai.</li><li>Set up the Whisper and Llama2 7b models on a MacBook Pro M1.</li><li>Use the Whisper model to convert speech from a webinar snippet into text.</li><li>Use the Llama2 7b model to translate the transcribed text.</li></ul><figure><img alt="A flowchart showing a process from ‘Webinar’ to ‘Whisper’ to ‘Llama2’ at ‘ollama.ai,’ with a language transition from English to English and then to Spanish." src="https://cdn-images-1.medium.com/max/1024/1*JXsD8F0LU7sqdZVhcYXppw.png" /><figcaption>Webinar Translation Workflow from English to Spanish</figcaption></figure><h3><strong>ollama.ai</strong></h3><p>ollama.ai can be downloaded at <a href="https://github.com/jmorganca/ollama">https://github.com/jmorganca/ollama</a><br>open the ollama app, and it will start the ollama server.</p><p>To download the Llama2 7b model enter the command.</p><pre>ollama pull llama2</pre><p>For other versions use llama2:13b or llama2:70b.</p><h3><strong>Whisper</strong></h3><p>To install OpenAI Whisper</p><pre>pip install -U openai-whisper</pre><p>OpenAI recommend to install rust</p><pre>pip install setuptools-rust</pre><h3>Whisper transcription</h3><p>Import the Whisper library</p><pre>import whisper</pre><p>Loading the Whisper model, transcribing audio content, and outputting the transcription.</p><p>For this example, I used the “tiny” Whisper model.</p><pre>  modelname = &quot;tiny&quot;<br>  model = whisper.load_model(modelname)<br>  result = model.transcribe(filename, fp16=False, language=&quot;en&quot;)<br>  transcription_text = result[&#39;text&#39;]</pre><h3>Llama2 7b Translation</h3><p>To initiate the translation append a prompt to the text to be transcribed.</p><pre>prompt = &quot;translate this English text into Spanish: &quot; + transcription_text</pre><p>Translate the text using Llama2 7b.</p><pre>llmmodel = &#39;llama2&#39;<br>try:<br>    r = requests.post(&#39;http://localhost:11434/api/generate&#39;,<br>                      json={<br>                          &#39;model&#39;: llmmodel,<br>                          &#39;prompt&#39;: prompt,<br>                      },<br>                      stream=True)<br>    r.raise_for_status()<br>    response_text = str(r.text)<br><br>    for line in r.iter_lines():<br>        body = json.loads(line)<br>        response_part = body.get(&#39;response&#39;, &#39;&#39;)<br>        # Process and print each token from the streamed response<br>        print(response_part, end=&#39;&#39;, flush=True)<br><br>except requests.exceptions.RequestException as e:<br>    print(f&quot;Error during API request: {e}&quot;)</pre><h3><strong>Results</strong></h3><p>The Whisper model’s performance in transcribing English text is impressive, showcasing high accuracy and contextual understanding. The provided example effectively captures the nuances of a strategic planning discussion, maintaining the technical specificity and coherence of the original speech. This level of precision in the transcription process illustrates Whisper’s robust capability in speech-to-text conversion, particularly in handling professional and context-rich content.<br>Whisper output.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xBYWsZwP_pff_wmr8wrvLA.jpeg" /><figcaption>Whisper output</figcaption></figure><p>Llama2 7b model’s translation into Spanish, while preserving the core essence and strategic concepts of the original content, lacks some of the linguistic fluidity and directness of the English version in places. This reflects trade-offs around precision due to using a smaller 7b parameter size rather than the more accurate but larger 70b model. However, it still reasonably conveys all the complex, technical ideas in Spanish well.<br>Llama2 output.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*vYvEZFotSAWK-bohHneCFg.jpeg" /><figcaption>Llama2 output</figcaption></figure><p>Integrating Whisper and Llama2 7b combines accurate speech-to-text with promising translation functionality. Whisper transcribes spoken words adeptly, while Llama2 7B conveys complex ideas reasonably well despite limitations in capturing linguistic nuance. The experiment offers a snapshot of large language models where precision, contextual comprehension, and adaptability approach human-level comprehension.</p><h3><strong>Key Takeaways</strong></h3><p>Our experiment with OpenAI’s Whisper and Meta’s Llama2 7b on a MacBook Pro M1 has successfully demonstrated the integration of open-source speech-to-text and translation technologies.</p><ul><li>Whisper exhibits exceptional accuracy in transcribing context-laden English speech. This highlights the viability for professional applications.</li><li>Llama2 7b model capably translates complex ideas but sacrifices linguistic fluidity and nuance, highlighting the need for refinement or a larger model.</li><li>On-device processing secures privacy-sensitive AI capabilities that otherwise rely on cloud solutions, underscoring rising user expectations of confidentiality.</li></ul><p>These insights reflect the current capabilities and future trajectory of open-source speech/language AI. They underscore the necessity of sustained invention in this fast-moving domain.</p><p>Interested in the evolving world of Generative AI? Join the conversation by sharing your thoughts in the comments below, or <a href="https://www.linkedin.com/in/david-kolb/">reach out to learn more.</a></p><h3><strong>Links to Code</strong></h3><p>ollama.ai <a href="https://github.com/jmorganca/ollama">https://github.com/jmorganca/ollama</a></p><p>Meta Llama2 <a href="https://ai.meta.com/llama/">https://ai.meta.com/llama/</a></p><p>OpenAI Whisper <a href="https://github.com/openai/whisper">https://github.com/openai/whisper</a></p><p>David Kolb <a href="http://www.linkedin.com/in/david-kolb/">www.linkedin.com/in/david-kolb/</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ca571b3c4293" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[From Innovation to Advantage]]></title>
            <link>https://medium.com/@bluekoins/from-innovation-to-advantage-6c1d000d35a9?source=rss-98bb79fdfe35------2</link>
            <guid isPermaLink="false">https://medium.com/p/6c1d000d35a9</guid>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[innovation]]></category>
            <category><![CDATA[business-strategy]]></category>
            <category><![CDATA[generative-ai-tools]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Davi]]></dc:creator>
            <pubDate>Mon, 27 Nov 2023 14:12:19 GMT</pubDate>
            <atom:updated>2023-11-27T14:12:19.622Z</atom:updated>
            <content:encoded><![CDATA[<p>Harnessing the Power of Generative AI in Your Business</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*nrHyRAvOOPk4fZLFyyjnrQ.png" /><figcaption>Image : <a href="https://www.linkedin.com/in/david-kolb/">David Kolb</a></figcaption></figure><p>Generative AI is a paradigm-shifting tool in the realm of business, offering immense opportunities for innovation, efficiency, and customer engagement. However, its introduction into the business landscape is not without challenges. In this blog post, I will delve into the transformative potential of generative AI, while also addressing the critical considerations and complexities it brings to the fore.</p><p><strong>Generative AI vs AI — What’s the Difference?</strong></p><p>Artificial Intelligence (AI) is the broader concept of machines being able to carry out tasks in a way that we consider akin to human intelligence. Technologies include simple rule-based systems to complex deep-learning models. Examples are predictive analysis in Amazon or Netflix, Siri and Google Assistant or autonomous vehicles, and medical image analysis.</p><p>Generative AI takes it a step further. It generates fresh content based on its training. Applications could include creating written text (Large language models) like ChatGPT and Bard, visual images (Diffusion models) like Midjourney, or music and speech.</p><p><a href="https://medium.com/p/21c41609d8c6">Learn more on artificial intelligence and generative AI terminology</a></p><p><strong>Two Strategic Paths with Generative AI</strong></p><p>In strategy, generative AI offers two paths. The first offers immediate applications by enriching the strategic development process by aiding in ideation, scenario planning, prototyping and storytelling. Generative AI can enhance creativity, enhance strategy development, provide fresh perspectives, identify patterns and generate diverse strategic scenarios for Strategists to consider.</p><p>The second path sees generative AI as a cornerstone in your strategy, underpinning products, services and experiences and transforming your value proposition. This path requires careful assessment of your market, competitors and capabilities. Whether as a cost leader or differentiator, sustainable use of generative AI in your strategy could be the key to standing out in your industry.</p><p><strong>Innovation meets authenticity</strong></p><p>Generative AI opens new horizons for innovation. Adobe’s use of generative AI in Adobe Firefly is a compelling example of balancing human creativity and AI-powered innovation.</p><p>Adobe Firefly is trained on images from Adobe Stock and openly licensed content without copyright restrictions to develop creative generative AI responsibly, with creators’ interests in mind. This approach helps avoid copyright issues, respects intellectual property rights, and prioritises user safety when using AI-generated content.</p><p>However, the use of generative AI poses a challenge. AI-generated content may not resonate with audiences seeking authentic, human-centric experiences, and careful evaluation is necessary to ensure the responsible and ethical integration of generative AI aligns with your organisation’s purpose and values.</p><p><strong>Unlocking Potential with Generative AI</strong></p><p>Integrating generative AI involves identifying strategic opportunities. For instance, using broader AI solutions in customer interactions, such as chatbots, has already been successful in customer service.</p><p>While the adoption rate varies, businesses should experiment with various use cases, prototype and test extensively, and stay receptive to feedback. This exploration and willingness to adapt are crucial to unlocking the true potential of generative AI in your business.</p><p><strong>Differentiation and Cost Leadership with Generative AI</strong></p><p>Generative AI can drive differentiation or cost leadership in your market. It gives organisations the capability to create unique offerings and operational efficiencies. These benefits do not come without challenges. Quality control is essential; while generative AI can generate vast amounts of content or solutions, the consistency and standard of these outputs can vary.</p><p>Overcoming hurdles related to adoption and implementation is crucial to unlock generative AI’s potential.</p><p><a href="https://medium.com/@kolbeuk/an-overview-of-porters-generic-competitive-strategies-453342d1a6ae">Learn more on competitive strategy</a></p><p><strong>From Adoption to Advantage, Making Generative AI Work for You</strong></p><p>While Generative AI offers potential advantages, simply having it may not provide a competitive edge as it becomes more widely adopted. The true advantage comes from using generative AI in a way that aligns with your business strategy, core competencies, and market positioning.</p><p>This might involve a unique application of the technology, a novel combination with other technologies, or a particular focus on leveraging generative AI for customer experience, efficiency, or innovation. As with any tool, the benefits of generative AI depend on how well you use it.</p><p>It’s not just about jumping on the bandwagon but about deeply integrating generative AI into your processes, systems, and culture in a way that creates distinctive value.</p><p><strong>Ethical and Authenticity Concerns</strong></p><p>One of the significant challenges when deploying generative AI is addressing potential biases in generated content, intellectual property concerns and maintaining authenticity. Organisations must integrate proper safeguards and ethical guidelines to ensure that AI-generated content remains fair, unbiased, and transparent.</p><p>Customers should be informed when interacting with AI-generated content to maintain authenticity and prevent misinformation or deception. By adhering to these ethical guidelines, businesses can responsibly harness the power of generative AI, mitigating adverse effects and building trust with their customers.</p><p><strong>Building Generative AI Capabilities</strong></p><p>Creating generative AI solutions requires expertise, in some cases computational power, effective data management, and high-quality training data. The power of this technology truly lies in the quality of the data it is trained on.</p><p>Equally important is the human factor — training and upskilling teams to work seamlessly with AI. Should your organisation lack these resources, strategic partnerships can fill the gap. Be mindful, however, that resource limitations could pose challenges to fully realising generative AI’s benefits.</p><p><strong>Conclusion</strong></p><p>Generative AI presents several strategic business opportunities, from boosting creativity and strategy development to creating operational efficiencies. It’s a powerful tool capable of driving differentiation or cost leadership. Key takeaways include:</p><ul><li>Generative AI provides an opportunity for strategy development or as part of your differentiation and value transformation.</li><li>Generative AI unlocks innovation but needs ethical integration for authenticity, avoiding copyright issues, and respecting user safety.</li><li>Businesses that want to use generative AI successfully must be aware of these challenges and have the right capabilities, including upskilling teams.</li></ul><p>Integrating it into your strategy demands thoughtful deliberation that balances innovation with authenticity, address ethical considerations, and ensures the technology aligns with your organisation’s purpose and values.</p><p>For more information check out <a href="https://www.davidkolbconsultancy.com/">David Kolb Consultancy</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6c1d000d35a9" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[ Quick Tip : Exploring the Future with Design Fiction]]></title>
            <link>https://medium.com/@bluekoins/quick-tip-exploring-the-future-with-design-fiction-755a9dcd7207?source=rss-98bb79fdfe35------2</link>
            <guid isPermaLink="false">https://medium.com/p/755a9dcd7207</guid>
            <category><![CDATA[prototyping]]></category>
            <category><![CDATA[innovation]]></category>
            <category><![CDATA[storytelling]]></category>
            <category><![CDATA[startup]]></category>
            <category><![CDATA[design-thinking]]></category>
            <dc:creator><![CDATA[Davi]]></dc:creator>
            <pubDate>Mon, 27 Nov 2023 14:08:35 GMT</pubDate>
            <atom:updated>2023-11-27T14:08:35.491Z</atom:updated>
            <content:encoded><![CDATA[<h3>🚀 Quick Tip : Exploring the Future with Design Fiction</h3><figure><img alt="Conceptual illustration of Design Fiction as an innovative tool for strategic transformation in business and creative sectors" src="https://cdn-images-1.medium.com/max/1024/1*RMmB5YklrSSppqMucyLCLg.png" /><figcaption>Image David Kolb Midjourney</figcaption></figure><p>In the fast-paced business world, companies are well-versed in tackling the challenges of today. However, the real potential lies in influencing and preparing for the future, and that’s where Design Fiction comes into play.</p><p><strong>What is Design Fiction?</strong></p><p>Design Fiction is an innovative tool that allows individuals and businesses to imagine and engage with futures that may or may not happen. Through this, one can create impactful narratives and prototypes to predict the future and critically explore and assess the implications of new technologies and ideas.</p><p><strong>Understanding the Core Components:</strong></p><ul><li>Narratives: Begin by crafting possible scenarios about upcoming technologies, products, or systems. It provides a foundational blueprint to understand the implications of each.</li><li>Storytelling Tools: Modern tools such as visual elements, AI-driven language models, and image generators can be instrumental in breathing life into your envisioned future.</li><li>Innovation: This isn’t about the routine, incremental improvements. It’s about radical, groundbreaking ideas that can potentially redefine industries.</li><li>Ethical Considerations: It’s essential to ponder novel tech’s advantages and pitfalls. Aim for a future that’s in harmony with collective human values.</li><li>Prototyping: Visions should never be static. As the world evolves, so should your envisaged future. Involve your stakeholders in refining and honing your visions.</li><li>Practicality: Unlike belief, design fiction isn’t purely academic. Entities, ranging from startups to conglomerates, can benefit immensely by preemptively exploring their role in the future.</li></ul><p>In conclusion, Design Fiction isn’t about passive prediction. It’s about active, proactive engagement with what lies ahead. By harnessing its potential, businesses can ensure they’re not just reacting to the world but actively shaping it.</p><p>For more on Design Fiction, check out the excllent book <a href="https://www.nearfuturelaboratory.com/the-manual-of-design-fiction/"><strong>“The Manual of Design Fiction” </strong></a>by Julian Bleecker, Nick Foster, Fabien Girardin and Nicolas Nova, provides a comprehensive, perspective on the approach, mindset, and potential of design fiction</p><p>For more information, checkout <a href="https://www.linkedin.com/in/david-kolb/">David Kolb</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=755a9dcd7207" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Navigating the AI Revolution]]></title>
            <link>https://medium.com/@bluekoins/navigating-the-ai-revolution-1f3b736ba745?source=rss-98bb79fdfe35------2</link>
            <guid isPermaLink="false">https://medium.com/p/1f3b736ba745</guid>
            <category><![CDATA[generative-ai-market]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[innovation-culture]]></category>
            <category><![CDATA[business-strategy]]></category>
            <category><![CDATA[technology-strategy]]></category>
            <dc:creator><![CDATA[Davi]]></dc:creator>
            <pubDate>Tue, 19 Sep 2023 12:16:03 GMT</pubDate>
            <atom:updated>2023-09-19T12:16:03.112Z</atom:updated>
            <content:encoded><![CDATA[<h3>How Generative AI is Transforming Business</h3><figure><img alt="An intricate pattern showcasing a vibrant spectrum of interconnected rainbow hues, artistically merging and intertwining in a harmonious display" src="https://cdn-images-1.medium.com/max/1024/1*GDCeOx37YlxVkzkaqmwHLQ.png" /><figcaption>Image : David Kolb : Midjourney</figcaption></figure><p>What does it really mean to be at the cusp of an AI revolution? Are we boldly embracing the future, or are we simply sleepwalking into a technologically advanced world, one we don’t yet fully understand?</p><p>Just as the printing press brought knowledge to the masses, the rise of radio waves linked continents and the internet? It became the bridge between human curiosity and established knowledge. And now, enter Generative AI. Suddenly, our computers and phones are not just providing information, they are joining in, socialising and brainstorming with us, they are leading to a-ha and wow, really moments and even throwing in a plot twist or two. It’s more than just automation; it’s about co-creation.</p><p>Historically, every big tech leap has made us pause, reflect, and sometimes second-guess our trajectory. And here we are, with Generative AI taking centre stage, captivating professionals around the globe. I mean, 80%** of professionals are already considering incorporating it into their work lives.</p><blockquote>While debates rage over the granular details, the overarching truth stands undisputed. Generative AI is advancing at a speed unparalleled by any preceding technology.</blockquote><p>Yet, it prompts us to ponder: Are we truly equipped for the profound implications and the manifold responsibilities that come with it?</p><p>Having navigated the intricate paths of digital and business transformation, I’ve seen the successes and failures of countless innovations. Each one brought its own set of challenges and opportunities. Generative AI, however, feels different, more profound. Building on my experience, let’s delve into how this technological marvel intersects with the well-trodden paths of transformation</p><p><a href="https://medium.com/@kolbeuk/quick-tip-demystifying-generative-ai-a-guide-to-text-and-image-generation-21c41609d8c6">Learn more about Artificial Intelligence and Generative AI Terminology</a></p><p><strong>The Age-old Quest for Clarity</strong><br>Have you ever bought a hammer before knowing what you needed to fix? Of course not. Yet, we often rush to adopt new technologies without clearly understanding what problems we’re trying to solve. Could the allure of the ‘new’ sometimes blinds us to the nuances of the ‘need’?</p><p>Before we rush, let’s pinpoint the exact problems we want this tech to address. Let our strategies, not just the allure of the tech, guide our steps.</p><p><strong>The Liberation of Trial and Error</strong><br>Should we keep generative AI in a silo, safely guarded by an exclusive group of experts? We’re living in an age that thrives on collaboration and diverse input. Remember when the printing press was invented? It didn’t just stay with the scholars; it became a tool for the masses. History shows us that widespread adoption often brings about the most profound changes. Allow departments to experiment, iterate, and find the sweet spots of its utility to boost your innovation pipeline.</p><p><strong>Beyond the Classroom</strong><br>While formal education has its merits, experience remains a formidable teacher. When it comes to generative AI, let’s not put ourselves in boxes. Get a mix of people on board — those with formal education in AI and those who are instinctively brilliant. You’ll end up with a veritable melting pot of ideas that could lead to solutions you’d never have thought of otherwise.</p><p><a href="https://medium.com/@kolbeuk/the-power-of-micro-experiments-by-embracing-small-scale-innovation-for-big-results-bb16eda38f1b">Learn more about the power of micro experiements</a></p><p><strong>Double Down on Domain Expertise</strong><br>Does AI possess the delicate nuance of human judgment? Can it truly understand the history of your industry or the culture of your market? We’ve always needed human intervention, haven’t we? From the Industrial Revolution to the digital age, the human element brings irreplaceable quality. Embrace the synergy of a co-creative relationship between AI and domain experts.</p><p><strong>The Tiny Triumphs that Make the Journey</strong><br>Small wins! Why are they often overlooked? Your AI-infused enterprise won’t be built in a day. So why not celebrate every milestone, big or small? How about an “AI-Deas” of the month? A little encouragement could go a long way in fostering innovation and uplifting morale.</p><p><strong>Stress Critical Thinking More Than Just a Buzzword</strong><br>While AI can suggest, should we always concur? Encourage a culture of questioning, a culture that values ethics and scrutinises the algorithms before integrating them fully into the business. Question to understand, not undermine.</p><p><a href="https://medium.com/@kolbeuk/quick-tip-the-power-of-thought-provoking-questions-169fbc612f25">Learn more about The Power of Thought-Provoking Questions</a></p><p><strong>Safety First</strong><br>Generative AI’s potential is vast, but it’s not without its pitfalls, notably biases — prioritise transparency and feedback loops. When setting up your generative AI system, the guiding principle should be trustworthiness, ensuring its usage benefits all.</p><p><strong>Conclusion</strong></p><p>So here we are, at this fascinating juncture where AI isn’t just a tool; it’s becoming a collaborator. It’s a winding road of discovery, paved with setbacks and triumphs. Rather than a mere instrument, generative AI is a tireless co-creator — a digital cofounder who never sleeps. By working hand-in-hand, blending AI capabilities with human judgment and values, we can shape solutions that uplift our collective potential.</p><p>But as we navigate these uncharted waters, I’m curious: what are your experiences and thoughts on Generative AI? Have you already embraced it, or are you still on the fence? Share your stories, insights, or concerns in the comments below.</p><p>If you found this intriguing and want to delve deeper into the world of technological advancements and business innovation, I invite you to explore more of my writings. Check out my other blogs in this series. “<a href="https://medium.com/@kolbeuk/from-digital-banking-to-ais-new-dawn-20f1e1692a37">From Digital Banking to AI’s New Dawn: A Personal Journey into the Future of Business Innovation</a>” and “<a href="https://towardsdatascience.com/navigating-the-large-language-model-landscape-f56bd67e48d9">Navigating the Large Language Model Landscape: Choices and considerations for Visionary Leaders and Entrepreneurs.</a>” These pieces further expand on the transformative potential of technology in our modern age and offer insights for forward-thinking individuals eager to harness the promise of the digital futur</p><p>**Acknowledgements:McKinsey’s report on “The state of AI in 2023” and OpenAI’s introduction to “ChatGPT Enterprise,” providing a foundation for our understanding of AI’s evolving landscape.</p><p>For more information check out <a href="https://www.davidkolbconsultancy.com/">David Kolb Consultancy</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1f3b736ba745" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Navigating AI Adoption]]></title>
            <link>https://medium.com/@bluekoins/navigating-ai-adoption-ed4914c11137?source=rss-98bb79fdfe35------2</link>
            <guid isPermaLink="false">https://medium.com/p/ed4914c11137</guid>
            <category><![CDATA[governance]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[leadership]]></category>
            <category><![CDATA[board-of-directors]]></category>
            <category><![CDATA[technology]]></category>
            <dc:creator><![CDATA[Davi]]></dc:creator>
            <pubDate>Mon, 11 Sep 2023 09:58:40 GMT</pubDate>
            <atom:updated>2023-09-11T10:13:16.783Z</atom:updated>
            <content:encoded><![CDATA[<h3>The Essential Perspectives for Board Members</h3><figure><img alt="visionary leader americana cinematic fullshot" src="https://cdn-images-1.medium.com/max/1024/1*lXW28WHvBXLzk5SsTRjJ9A.jpeg" /><figcaption>Image : David Kolb</figcaption></figure><p>As Artificial Intelligence (AI) rapidly evolves from a speculative technology of the future into a game-changer in the present, board members must comprehend and navigate this landscape. This post dives into a multifaceted perspective for a balanced approach to AI adoption, including an essential understanding of generative AI.</p><p><strong>The AI vs Generative AI Perspective</strong></p><p>At its core, AI technology allows machines to mimic human intelligence, typically involving learning, reasoning, creative problem solving, perception, and language understanding. For instance, your phone’s virtual assistant uses AI to understand and respond to your voice commands.</p><p>On the other hand, Generative AI takes this a step further, creating new content or predictions based on its understanding. Generative AI models can generate new ideas and designs and even write text. A prime example is ChatGPT by OpenAI, a language prediction model that generates human-like text based on the input.</p><p>Understanding the distinction between these two is crucial for board members as it affects investment, strategy, and the organisation’s implementation scope.</p><p><a href="https://medium.com/@kolbeuk/quick-tip-demystifying-generative-ai-a-guide-to-text-and-image-generation-21c41609d8c6">Learn more about Generative AI terminology</a></p><p><strong>The Business Perspective</strong></p><p>AI’s transformative potential for businesses lies in its ability to enhance efficiency, profitability, and innovation. However, successful AI adoption isn’t universal. Consider the example of Amazon, which has employed AI to streamline its warehouse operations, dramatically improving efficiency. Every organisation needs a comprehensive, strategy-driven approach to AI that considers its unique business model, industry, operational processes, and corporate culture. Moreover, as board members explore AI adoption and embrace the power of analytical, creative, and systems thinking within the organisation becomes crucial.</p><p><a href="https://medium.com/@kolbeuk/quick-tip-unlocking-the-power-of-analytical-creative-and-systems-thinking-2a0ecdeacbc0">Learn more about analytical, creative, and systems thinking.</a></p><p><strong>The Customer Perspective</strong></p><p>AI’s role in enhancing customer experience is evident. For instance, Netflix uses AI algorithms to personalise movie recommendations, significantly improving user satisfaction. But it’s essential to strike a balance. Let’s take the banking sector as an example. Many customers appreciate the convenience of AI-driven chatbots for basic queries. However, they often prefer speaking to a human representative who can understand nuances and provide empathy when it comes to complex financial decisions or issues. So, businesses must maintain the human aspect of customer interactions, ensuring a blend of AI efficiencies and genuine human touch for optimal customer experience.</p><p><strong>The Investor/Stakeholder Perspective</strong></p><p>Investors are attracted to companies effectively leveraging AI technologies as they signify a willingness to innovate. Transparency in AI utilisation is essential, as demonstrated by Google, which publishes AI principles and practices to ensure investor confidence. Clear communication aligns stakeholder expectations, fostering a shared understanding of the company’s AI strategy.</p><p><strong>The Regulatory/Compliance Perspective</strong></p><p>Navigating the regulatory landscape of AI is complex as laws and regulations evolve continually. The EU’s General Data Protection Regulation (GDPR), which imposes strict rules on data usage, is a case in point. Companies must proactively ensure their AI strategies align with current laws and anticipate future changes.</p><p><strong>The Ethical/Social Impact Perspective</strong></p><p>AI adoption holds broader societal implications. For example, concerns over bias in AI algorithms and their impact on job automation necessitate an ethical framework integral to any AI strategy. Businesses like IBM, with its ‘Principles for Trust and Transparency’, must proactively address these issues to enhance their reputation and build trust.</p><p><strong>The Competitive Perspective</strong></p><p>AI adoption is a significant competitive differentiator in today’s dynamic business environment. Companies like Tesla have significantly altered the automotive industry’s landscape with their AI innovations. Staying updated on competitors’ AI strategies is not just about the present but also about predicting future trends. As AI continues to evolve, we can anticipate more personalised user experiences, advanced real-time data analytics, and increased integration of AI in areas we haven’t yet imagined. Businesses that can forecast these trajectories and adapt will undoubtedly lead in their respective industries.</p><p><strong>The Risk Management Perspective</strong></p><p>AI brings new risks like cybersecurity threats, data privacy issues, and challenges of integrating AI into existing operations. A proactive risk management strategy that includes these AI-specific risks in the broader enterprise risk management plan is critical. This can mitigate potential risks and build resilience, as companies like Cisco exemplify with their robust AI risk management frameworks.</p><p><strong>Conclusion</strong></p><p>AI is reshaping the business landscape. For board members, understanding this change requires viewing AI adoption from multiple perspectives, including the nuances between AI and Generative AI. As board members explore integrating AI, focus on these actionable takeaways:</p><ul><li>Conduct an AI readiness assessment of your operations and processes</li><li>Develop a step-by-step roadmap for AI adoption aligned with business strategy</li><li>Survey customers to understand AI preferences and concerns</li><li>Set guidelines to balance AI efficiencies with human expertise</li><li>Communicate AI adoption plans and progress proactively</li><li>Publish AI principles to ensure transparency</li></ul><p>We invite you to reach out for an in-depth discussion as you consider these perspectives. Navigating the AI landscape is a shared journey, best undertaken together.</p><p>Given the rapid evolution of AI and its multifaceted impact, how are board members in your organisation approaching its adoption? Share your experiences below! 💡👇</p><p>For more information check out <a href="https://www.davidkolbconsultancy.com/">David Kolb Consultancy</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ed4914c11137" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Quick Tip — The Power of Thought-Provoking Questions!]]></title>
            <link>https://medium.com/@bluekoins/quick-tip-the-power-of-thought-provoking-questions-169fbc612f25?source=rss-98bb79fdfe35------2</link>
            <guid isPermaLink="false">https://medium.com/p/169fbc612f25</guid>
            <category><![CDATA[innovation]]></category>
            <category><![CDATA[growth-mindset]]></category>
            <category><![CDATA[questions]]></category>
            <category><![CDATA[stroytelling]]></category>
            <category><![CDATA[design-thinking]]></category>
            <dc:creator><![CDATA[Davi]]></dc:creator>
            <pubDate>Sun, 27 Aug 2023 15:23:06 GMT</pubDate>
            <atom:updated>2023-08-27T15:23:06.822Z</atom:updated>
            <content:encoded><![CDATA[<h3>Quick Tip — The Power of Thought-Provoking Questions!</h3><figure><img alt="People sitting in a cozy coffee shop, with a steaming cup beside them, deeply engrossed in sketching out a preliminary concept for a business experiment on a notebook." src="https://cdn-images-1.medium.com/max/1024/1*EFsa2Bt32E0wupuvhUrCHw.png" /><figcaption>Image : David Kolb</figcaption></figure><p>🚀 Quick Tip : The Power of Thought-Provoking Questions for Business Innovation 🧠💡</p><p>Harnessing the power of thought-provoking questions can be a game-changer for your business. These questions can unveil pivotal insights, spotlight overlooked challenges, and reveal hidden growth avenues. They’re catalysts for innovation.</p><p>Consider the following when crafting your questions:</p><p>🎯 Opt for precision. Avoid general inquiries like “Is this a good idea?” Instead, probe more profound: “What hurdles did you face when tackling this issue in the past?”</p><p>🧐 Approach with genuine curiosity. Stay open to diverse viewpoints and avoid presumptions.</p><p>❓ Foster innovation with open-ended prompts. Rather than “What are our main obstacles?” ask, “How might we navigate our challenges?”</p><p>😓 Embrace the hard-hitting questions. Facing issues head-on, like inquiring, “Where are we missing the mark?” can spark pivotal shifts.</p><p>👂 Once you’ve inquired, attentively absorb the feedback. This is where invaluable insights about your business lie.</p><p>🤔 Continually seek and reflect on feedback. Use it as a compass for refining your strategies.</p><p>Which transformative questions will you bring to the table? Engage below and share your insights! 🚀</p><p>For more information check out <a href="https://www.davidkolbconsultancy.com/">David Kolb Consultancy</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=169fbc612f25" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>