<?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 Sab Severino on Medium]]></title>
        <description><![CDATA[Stories by Sab Severino on Medium]]></description>
        <link>https://medium.com/@thesab?source=rss-e35bcd4a52e6------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*H970WbXA6KfK5fr2NGK80w.jpeg</url>
            <title>Stories by Sab Severino on Medium</title>
            <link>https://medium.com/@thesab?source=rss-e35bcd4a52e6------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 19:03:42 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@thesab/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[GPT-5 & GPT-OSS: A Focused Look at OpenAI’s Next Chapter]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@thesab/gpt-5-gpt-oss-a-focused-look-at-openais-next-chapter-2a23711aba7c?source=rss-e35bcd4a52e6------2"><img src="https://cdn-images-1.medium.com/max/618/1*z_FJtrKkQei0RtCmG8fnUw.png" width="618"></a></p><p class="medium-feed-snippet">Introduction</p><p class="medium-feed-link"><a href="https://medium.com/@thesab/gpt-5-gpt-oss-a-focused-look-at-openais-next-chapter-2a23711aba7c?source=rss-e35bcd4a52e6------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://medium.com/@thesab/gpt-5-gpt-oss-a-focused-look-at-openais-next-chapter-2a23711aba7c?source=rss-e35bcd4a52e6------2</link>
            <guid isPermaLink="false">https://medium.com/p/2a23711aba7c</guid>
            <category><![CDATA[generative-ai-tools]]></category>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[gpt-5]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[openai]]></category>
            <dc:creator><![CDATA[Sab Severino]]></dc:creator>
            <pubDate>Tue, 19 Aug 2025 11:05:05 GMT</pubDate>
            <atom:updated>2025-08-19T11:05:05.496Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[DeepSeek: L’alternativa economica a ChatGPT con Python — Guida completa]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@thesab/deepseek-lalternativa-economica-a-chatgpt-con-python-guida-completa-a9bfa0c6a8dd?source=rss-e35bcd4a52e6------2"><img src="https://cdn-images-1.medium.com/max/1704/1*zcKl5BYo3UPoW3rKLLyIgg.png" width="1704"></a></p><p class="medium-feed-snippet">Hai mai desiderato un&#x2019;API di intelligenza artificiale potente come ChatGPT ma con costi notevolmente ridotti? Allora dovresti conoscere&#x2026;</p><p class="medium-feed-link"><a href="https://medium.com/@thesab/deepseek-lalternativa-economica-a-chatgpt-con-python-guida-completa-a9bfa0c6a8dd?source=rss-e35bcd4a52e6------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://medium.com/@thesab/deepseek-lalternativa-economica-a-chatgpt-con-python-guida-completa-a9bfa0c6a8dd?source=rss-e35bcd4a52e6------2</link>
            <guid isPermaLink="false">https://medium.com/p/a9bfa0c6a8dd</guid>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[tutorial]]></category>
            <category><![CDATA[development]]></category>
            <category><![CDATA[deepseek]]></category>
            <category><![CDATA[python]]></category>
            <dc:creator><![CDATA[Sab Severino]]></dc:creator>
            <pubDate>Sun, 02 Feb 2025 18:32:57 GMT</pubDate>
            <atom:updated>2025-02-02T18:32:57.887Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[A Green Dawn for AI: How DeepSeek’s Breakthrough Challenges Climate Pessimism]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@thesab/a-green-dawn-for-ai-how-deepseeks-breakthrough-challenges-climate-pessimism-464974702c0f?source=rss-e35bcd4a52e6------2"><img src="https://cdn-images-1.medium.com/max/1280/1*uW3uDVFHgXiw0HarLitdyw.jpeg" width="1280"></a></p><p class="medium-feed-snippet">The rapid growth of artificial intelligence has long been shadowed by an uncomfortable truth: its voracious appetite for energy. Headlines&#x2026;</p><p class="medium-feed-link"><a href="https://medium.com/@thesab/a-green-dawn-for-ai-how-deepseeks-breakthrough-challenges-climate-pessimism-464974702c0f?source=rss-e35bcd4a52e6------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://medium.com/@thesab/a-green-dawn-for-ai-how-deepseeks-breakthrough-challenges-climate-pessimism-464974702c0f?source=rss-e35bcd4a52e6------2</link>
            <guid isPermaLink="false">https://medium.com/p/464974702c0f</guid>
            <category><![CDATA[environment]]></category>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[deepseek]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Sab Severino]]></dc:creator>
            <pubDate>Thu, 30 Jan 2025 19:04:09 GMT</pubDate>
            <atom:updated>2025-01-30T19:05:47.318Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[How to Install DeepSeek in Your Browser for Offline AI — Featuring a Live Hugging Face Demo]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@thesab/how-to-install-deepseek-in-your-browser-for-offline-ai-featuring-a-live-hugging-face-demo-08ee53b8e572?source=rss-e35bcd4a52e6------2"><img src="https://cdn-images-1.medium.com/max/1908/1*UMLAhQ0L4HrQrUslA-JqvA.png" width="1908"></a></p><p class="medium-feed-snippet">How-To Guide: Run DeepSeek-R1 Locally in Your Browser with WebGPU Acceleration</p><p class="medium-feed-link"><a href="https://medium.com/@thesab/how-to-install-deepseek-in-your-browser-for-offline-ai-featuring-a-live-hugging-face-demo-08ee53b8e572?source=rss-e35bcd4a52e6------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://medium.com/@thesab/how-to-install-deepseek-in-your-browser-for-offline-ai-featuring-a-live-hugging-face-demo-08ee53b8e572?source=rss-e35bcd4a52e6------2</link>
            <guid isPermaLink="false">https://medium.com/p/08ee53b8e572</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[deepseek]]></category>
            <category><![CDATA[openai]]></category>
            <category><![CDATA[hugging-face]]></category>
            <dc:creator><![CDATA[Sab Severino]]></dc:creator>
            <pubDate>Wed, 29 Jan 2025 14:28:29 GMT</pubDate>
            <atom:updated>2025-01-29T16:33:50.818Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[How DeepSeek Overtook ChatGPT: The Rise of China’s AI Powerhouse]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@thesab/how-deepseek-overtook-chatgpt-the-rise-of-chinas-ai-powerhouse-bfd1a0062161?source=rss-e35bcd4a52e6------2"><img src="https://cdn-images-1.medium.com/max/684/1*Pm5-outZiR27G5O_BYoZKA.png" width="684"></a></p><p class="medium-feed-snippet">ChatGPT vs DeepSeek</p><p class="medium-feed-link"><a href="https://medium.com/@thesab/how-deepseek-overtook-chatgpt-the-rise-of-chinas-ai-powerhouse-bfd1a0062161?source=rss-e35bcd4a52e6------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://medium.com/@thesab/how-deepseek-overtook-chatgpt-the-rise-of-chinas-ai-powerhouse-bfd1a0062161?source=rss-e35bcd4a52e6------2</link>
            <guid isPermaLink="false">https://medium.com/p/bfd1a0062161</guid>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[deepseek]]></category>
            <category><![CDATA[generative-ai-tools]]></category>
            <category><![CDATA[openai]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Sab Severino]]></dc:creator>
            <pubDate>Mon, 27 Jan 2025 16:34:00 GMT</pubDate>
            <atom:updated>2025-01-28T17:02:51.327Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[My Comeback: Notes (and Prompts) from a Journey with AI and Jump Rope]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@thesab/my-comeback-notes-and-prompts-from-a-journey-with-ai-and-jump-rope-956281b01bf4?source=rss-e35bcd4a52e6------2"><img src="https://cdn-images-1.medium.com/max/600/1*ZlVVrjySMkxUbFPpGDvCAw.gif" width="600"></a></p><p class="medium-feed-snippet">How I Lost 8 kg Thanks to GPT4-o, Apple Watch, and a Dash of Determination</p><p class="medium-feed-link"><a href="https://medium.com/@thesab/my-comeback-notes-and-prompts-from-a-journey-with-ai-and-jump-rope-956281b01bf4?source=rss-e35bcd4a52e6------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://medium.com/@thesab/my-comeback-notes-and-prompts-from-a-journey-with-ai-and-jump-rope-956281b01bf4?source=rss-e35bcd4a52e6------2</link>
            <guid isPermaLink="false">https://medium.com/p/956281b01bf4</guid>
            <category><![CDATA[apple-watch]]></category>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[openai]]></category>
            <category><![CDATA[workout]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Sab Severino]]></dc:creator>
            <pubDate>Thu, 23 Jan 2025 15:19:33 GMT</pubDate>
            <atom:updated>2025-01-23T15:21:57.340Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Exploring smolagents: Simplifying Agentic AI for LLM]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@thesab/exploring-smolagents-simplifying-agentic-ai-for-llm-6c9c695d0e23?source=rss-e35bcd4a52e6------2"><img src="https://cdn-images-1.medium.com/max/617/1*v3f_J-Xzvvujp1xAhdH5vw.png" width="617"></a></p><p class="medium-feed-snippet">Recently, I came across an exciting new library called smolagents, which aims to simplify the implementation of Agentic AI in LLM. This&#x2026;</p><p class="medium-feed-link"><a href="https://medium.com/@thesab/exploring-smolagents-simplifying-agentic-ai-for-llm-6c9c695d0e23?source=rss-e35bcd4a52e6------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://medium.com/@thesab/exploring-smolagents-simplifying-agentic-ai-for-llm-6c9c695d0e23?source=rss-e35bcd4a52e6------2</link>
            <guid isPermaLink="false">https://medium.com/p/6c9c695d0e23</guid>
            <category><![CDATA[generative-ai-tools]]></category>
            <category><![CDATA[python]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[genai]]></category>
            <category><![CDATA[agentic-ai]]></category>
            <dc:creator><![CDATA[Sab Severino]]></dc:creator>
            <pubDate>Tue, 14 Jan 2025 09:24:35 GMT</pubDate>
            <atom:updated>2025-01-14T09:29:56.995Z</atom:updated>
            <cc:license>http://creativecommons.org/publicdomain/zero/1.0/</cc:license>
        </item>
        <item>
            <title><![CDATA[Erase unwanted objects from a photo with AI]]></title>
            <link>https://medium.com/@thesab/erase-unwanted-objects-from-photo-with-ai-7d450181bde7?source=rss-e35bcd4a52e6------2</link>
            <guid isPermaLink="false">https://medium.com/p/7d450181bde7</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[computer-vision]]></category>
            <category><![CDATA[flux]]></category>
            <category><![CDATA[hugging-face]]></category>
            <category><![CDATA[inpainting]]></category>
            <dc:creator><![CDATA[Sab Severino]]></dc:creator>
            <pubDate>Tue, 07 Jan 2025 20:31:43 GMT</pubDate>
            <atom:updated>2025-01-08T10:28:35.766Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*l62lT_lgrKn3SPKjZCfZew.png" /></figure><p><em>Photography has always been more than just a hobby for me; it’s a way to capture the fleeting moments of life. I still remember the first time I held my grandfather’s old Polaroid camera — the excitement of waiting for the image to develop, and the magic of seeing a moment frozen in time. But sometimes, despite our best efforts, unwanted elements sneak into our perfect shots — be it a stray tourist, an unexpected photobomber, or a distracting background object.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/990/1*oC10qf5AGTf8pt2YJQ9nyQ.jpeg" /><figcaption>My vintage Polaroid Supercolor 1000</figcaption></figure><p>In the age of digital photography and advanced technology, we’re no longer bound by these imperfections. Today, I’d like to share how you can harness the power of AI to remove unwanted elements from your photos effortlessly, without being a Photoshop expert.</p><h3>The Power of AI in Image Editing</h3><p>Artificial Intelligence has revolutionized many aspects of our lives, and <strong>image editing </strong>is no exception. By leveraging advanced models and algorithms, AI can now understand and manipulate images in ways previously thought impossible.</p><h3>Introducing AI-Powered Inpainting</h3><p><em>Inpainting</em> is the process of reconstructing lost or deteriorated parts of images.</p><p>With AI, this process has become more accessible and efficient. Advanced models can analyze the surrounding pixels of the area to be filled and generate content that seamlessly blends with the rest of the image.</p><h3>Getting Started with AI Inpainting</h3><p>Below, I’ll guide you through the core components needed to set up an AI-powered inpainting tool. We’ll focus on key code snippets and explanations to help you understand how the system works.</p><h3>Core Technologies Used</h3><ul><li><strong>PyTorch</strong>: An open-source machine learning library for Python.</li><li><strong>Hugging Face Transformers &amp; Diffusers</strong>: Libraries providing pre-trained models for natural language processing and computer vision.</li><li><strong>Pillow</strong>: A Python Imaging Library that adds image processing capabilities.</li></ul><h3>Setting Up the Environment</h3><p>Ensure you have (at least) Python 3.9 installed on your system. Then, install the required libraries:</p><pre>pip install diffusers==0.30.2 torch transformers accelerate huggingface_hub sentencepiece pillow protobuf</pre><h3>Loading the Models</h3><p>To perform inpainting, we’ll use specialized AI models from <strong>Hugging Face</strong>. These models are designed to understand and manipulate images intelligently.</p><h4>Importing Necessary Libraries</h4><pre>import torch  <br>from PIL import Image  <br>from io import BytesIO  <br>import base64  <br>import config  <br>  <br># Import custom modules from &#39;flux&#39; directory  <br>from flux.controlnet_flux import FluxControlNetModel  <br>from flux.transformer_flux import FluxTransformer2DModel  <br>from flux.pipeline_flux_controlnet_inpaint import FluxControlNetInpaintingPipeline  <br>  <br># Authenticate with Hugging Face if needed  <br>import huggingface_hub  <br>huggingface_hub.login(&quot;your_huggingface_token&quot;)  <br>  <br>from diffusers.utils import check_min_version  <br>check_min_version(&quot;0.30.2&quot;</pre><h4>Loading Pre-trained Models</h4><pre># Load the transformer model  <br>transformer = FluxTransformer2DModel.from_pretrained(  <br>    &quot;black-forest-labs/FLUX.1-dev&quot;,  <br>    subfolder=&#39;transformer&#39;,  <br>    torch_dtype=torch.bfloat16  <br>)  <br>  <br># Load the controlnet model  <br>controlnet = FluxControlNetModel.from_pretrained(  <br>    &quot;alimama-creative/FLUX.1-dev-Controlnet-Inpainting-Beta&quot;,  <br>    torch_dtype=torch.bfloat16  <br>)  <br>  <br># Build the inpainting pipeline  <br>pipe = FluxControlNetInpaintingPipeline.from_pretrained(  <br>    &quot;black-forest-labs/FLUX.1-dev&quot;,  <br>    controlnet=controlnet,  <br>    transformer=transformer,  <br>    torch_dtype=torch.bfloat16  <br>)  <br>  <br># Move the pipeline to the desired device (GPU recommended)  <br>pipe.to(&quot;cuda&quot;)  # Use &quot;cpu&quot; if GPU is not available</pre><h3>Preparing the Input Images</h3><p>Before processing, we need to load and prepare the input image and the corresponding mask that indicates the areas to be inpainted.</p><pre># Load the input image and mask  <br>def load_images(image_base64, mask_base64):  <br>    image = Image.open(BytesIO(base64.b64decode(image_base64)))  <br>    mask = Image.open(BytesIO(base64.b64decode(mask_base64)))  <br>    return image, mask  <br>  <br># Example usage  <br># image_base64 and mask_base64 are base64 encoded strings of the images  <br>image, mask = load_images(input_image_base64, input_mask_base64)</pre><h3>Resizing Images for the Model</h3><p>The models require the images to be of specific dimensions, typically multiples of 8.</p><pre>def resize_images(image, mask):  <br>    _h, _w = image.size  <br>    h = (_h // 8) * 8 + (8 if _h % 8 != 0 else 0)  <br>    w = (_w // 8) * 8 + (8 if _w % 8 != 0 else 0)  <br>    size = (h, w)  <br>  <br>    image_resized = image.convert(&quot;RGB&quot;).resize(size)  <br>    mask_resized = mask.convert(&quot;RGB&quot;).resize(size)  <br>    return image_resized, mask_resized, image.size  <br>  <br># Resize images  <br>image_resized, mask_resized, original_size = resize_images(image, mask)</pre><h3>Running the Inpainting Pipeline</h3><p>Now, we can run the pipeline to perform the inpainting.</p><pre># Set the generator for reproducibility  <br>generator = torch.Generator(device=&quot;cuda&quot;).manual_seed(config.DEFAULT_SEED)  <br>  <br># Run the pipeline  <br>result = pipe(  <br>    prompt=&quot;&quot;,  # Empty prompt since we are not adding new objects  <br>    negative_prompt=config.DEFAULT_NEGATIVE_PROMPT,  <br>    height=image_resized.height,  <br>    width=image_resized.width,  <br>    control_image=image_resized,  <br>    control_mask=mask_resized,  <br>    num_inference_steps=config.DEFAULT_NUM_INFERENCE_STEPS,  <br>    generator=generator,  <br>    controlnet_conditioning_scale=config.DEFAULT_CONTROLNET_CONDITIONING_SCALE,  <br>    guidance_scale=config.DEFAULT_GUIDANCE_SCALE,  <br>    true_guidance_scale=config.DEFAULT_TRUE_GUIDANCE_SCALE  <br>).images[0]</pre><h3>Post-processing the Result</h3><p>Resize the output image back to its original dimensions.</p><pre># Resize the result to the original image size  <br>processed_image = result.resize(original_size)</pre><h3>Saving the Edited Image</h3><p>Save the resulting image to a file or encode it as needed.</p><pre>import uuid  <br>from pathlib import Path  <br>  <br># Save the image to a file  <br>output_folder = Path(&quot;output&quot;)  <br>output_folder.mkdir(parents=True, exist_ok=True)  <br>image_path = output_folder / f&quot;edited_image_{uuid.uuid4().hex}.png&quot;  <br>processed_image.save(image_path)</pre><h3>Understanding the Key Components</h3><h3>The Inpainting Pipeline</h3><p>The FluxControlNetInpaintingPipeline combines the <strong>transformer</strong> and <strong>controlnet</strong> models to process the image and mask inputs. It uses advanced AI techniques to analyze the image context and generate a realistic fill for the masked area.</p><h4>The Transformer Model</h4><p>The transformer model, FluxTransformer2DModel, is responsible for understanding the high-level features of the image. It processes the image data and aids in generating coherent content that matches the surrounding areas.</p><h4>The ControlNet Model</h4><p>FluxControlNetModel guides the transformer model by providing additional conditioning based on the input mask. It helps focus the inpainting process precisely where it&#39;s needed.</p><h3>Image and Mask Handling</h3><p>Proper image preprocessing ensures compatibility with the models and optimal performance. Resizing images to the correct dimensions and converting them to the appropriate formats are crucial steps.</p><h3>Parameters and Configuration</h3><p>Adjusting parameters like guidance_scale, num_inference_steps, and controlnet_conditioning_scale can influence the inpainting results. Experimenting with these values allows you to fine-tune the output.</p><h3>Before and After: Transforming Your Photos</h3><p>Let’s look at some examples of how this AI-powered tool can enhance your photos.</p><h3>Example 1: Removing Unwanted Objects</h3><p><em>Imagine capturing a perfet day but there’s a distracting signpost in the foreground.</em></p><p><strong>Before:</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/728/1*4FscT4lg_KpkAqOzlLOdcQ.png" /></figure><p><strong>After:</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/728/1*6l9DztyypR7ixF7x7KE8Rg.png" /></figure><h3>Example 2: Say Goodbye to Photobombers</h3><p><em>You took the perfect photo, but someone walked by right at that moment.</em></p><p><strong>Before:</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/728/1*MbHC_1xsZT6ko33jBFJT0g.png" /></figure><p><strong>After:</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/728/1*NIIJOymzRNFdWkZbh3LkTw.png" /></figure><h3>Example 3: Cleaning Up the Background</h3><p><em>For the rest, a romantic portrait ruined by a strange presence.</em></p><p><strong>Before:</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/728/1*qIHs9iiThqFOFnPC97AiqQ.png" /></figure><p><strong>After:</strong></p><p><em>The portrait with a clean and discreet background, which allows the lovers to stand out.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/728/1*D2-157paFPj_sUEpowWO8w.png" /></figure><p><em>Note: The ‘After’ descriptions are representations of what the tool can achieve.</em></p><h3>Example 4: Where is the Bridge?</h3><p><em>Me and the Golden Gate: An Iconic Moment!</em></p><h4>Before:</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/1*4uC2MIfgmclfiUGwc-tmfQ.png" /></figure><h4>After (…the Apocalypse):</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/1*NF7Uja2RKJwFV9hr94uDAA.png" /></figure><h3>Conclusion</h3><p>Photography is all about capturing the essence of a moment. With the help of AI, we can now ensure that nothing detracts from the memories we cherish.</p><p>This tool empowers anyone to enhance their photos effortlessly, without the need for advanced editing skills.</p><p>I hope this guide inspires you to explore the possibilities of AI in image editing.</p><p>Whether you’re a professional photographer or just someone who loves taking pictures, this technology opens up new avenues to perfect your images.</p><p>Happy shooting!</p><blockquote>Disclaimer: This article is for educational purposes. The code snippets provided are simplified and may require adjustments to work in a production environment. Always test and validate your implementations.</blockquote><blockquote>Disclaimer (2): The views and opinions expressed in this analysis are solely my own and do not reflect the official policy or position of my employer or any other organization. This analysis is purely for informational purposes and should not be considered as professional advice.</blockquote><p><em>Enjoyed this article? Support my work by </em><a href="http://buymeacoffee.com/thesab"><em>buying me a coffee</em></a><em>. Thank you!</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7d450181bde7" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Ho Potenziato Alexa con GPT4-o di OpenAI]]></title>
            <link>https://medium.com/@thesab/ho-potenziato-alexa-con-gpt4-o-di-openai-194636a1a333?source=rss-e35bcd4a52e6------2</link>
            <guid isPermaLink="false">https://medium.com/p/194636a1a333</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[openai]]></category>
            <category><![CDATA[alexa]]></category>
            <category><![CDATA[gpt-4]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Sab Severino]]></dc:creator>
            <pubDate>Mon, 14 Oct 2024 20:43:07 GMT</pubDate>
            <atom:updated>2024-11-12T10:14:51.694Z</atom:updated>
            <content:encoded><![CDATA[<h4>La Mia Esperienza con Lumi</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7QQ60TpLw24Zj2f3Jd6foA.png" /></figure><p>Nel mondo frenetico di oggi, la gestione delle comunicazioni condominiali, delle assemblee e delle scadenze può diventare un compito arduo.</p><p>Per risolvere questo problema, ho deciso di sviluppare una skill privata per Alexa che sfrutta GPT4-o di OpenAI come assistente personale privato.</p><p>In questo articolo, voglio condividere la mia esperienza nello sviluppo di questa skill, chiamata “ALexa genAI”, utilizzando Python e l’infrastruttura Alexa-Hosted Skills.</p><h3>L’Idea di Partenza</h3><p>L’idea di creare Lumi (è il nome che si è scelto quando ha detto di non chiamarsi Alexa) nasce dalla mia insoddisfazione per le limitazioni della modalità di interazione con Alexa. Nonostante il modello AlexaTM 20B fino al 2022 performasse meglio di GPT-3, oggi è sotto gli occhi di tutti la differenza con i modelli avanzati di OpenAI.</p><p>Volevo un assistente capace di:</p><ul><li>Analizzare le comunicazioni di condominio (assemblee, spese, etc.)</li><li>Ricordarmi le scadenze</li><li>Riassumere i punti principali discussi durante le riunioni</li></ul><p>Questi compiti possono sembrare semplici, ma richiedono un notevole sforzo organizzativo e di memoria …che ultimamente peggiora.</p><p>Qui entra in gioco Lumi, il mio assistente personale basato su GPT4-o.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9-A9Vb0N6XksxXKihpGG3Q.png" /></figure><h3>Perché Alexa-Hosted Skills?</h3><p>Quando si crea una skill per Alexa, è possibile gestire il codice e le risorse di backend da soli oppure utilizzando le Alexa-Hosted Skills.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*P8Wsovvmva9RXoFAEP5hlw.png" /></figure><p>Ho optato per la seconda opzione per diversi motivi:</p><ol><li><strong>Rapidità di Sviluppo</strong>: Utilizzando una delle skill template disponibili nel developer console, ho potuto iniziare rapidamente il mio progetto.</li><li><strong>Gestione Semplificata delle Risorse</strong>: Alexa si occupa di provisionare e gestire le risorse AWS necessarie, come gli endpoint AWS Lambda, un bucket Amazon S3 per lo storage dei media, una tabella Amazon DynamoDB per la persistenza dei dati e un repository AWS CodeCommit.</li><li><strong>Accessibilità</strong>: Non è necessario un account AWS separato per creare una skill buildata su Alexa. Alexa crea e gestisce le risorse direttamente dal developer console.</li></ol><h3>Lo Sviluppo di Lumi</h3><h4>Creazione della Skill</h4><p>Per iniziare, ho utilizzato la developer console di Alexa per creare una nuova hosted-skill . Il processo è stato facilitato dalla presenza di un editor di codice online che mi ha permesso di scrivere, modificare e distribuire il codice direttamente agli endpoint AWS Lambda.</p><h4>Utilizzo del GPT4-o</h4><p>L’elemento chiave della skill è l’integrazione con GPT4-o, il più potente modello di intelligenza artificiale (ad oggi Ottobre 2024) sviluppato da OpenAI. Questo modello è in grado di comprendere e generare testo in linguaggio naturale, rendendolo ideale per analizzare le comunicazioni condominiali e fornire riassunti accurati.</p><h3>Funzionalità Implementate</h3><ol><li><strong>Analisi delle Comunicazioni</strong>: Lumi analizza automaticamente le comunicazioni condominiali, identificando le informazioni chiave come le spese e le decisioni prese durante le assemblee.</li><li><strong>Promemoria delle Scadenze</strong>: Lumi invia promemoria per le scadenze importanti, come le date delle prossime riunioni o i pagamenti delle spese condominiali.</li><li><strong>Riassunti delle Riunioni</strong>: Dopo ogni assemblea, Lumi genera un riassunto dei punti principali discussi e delle decisioni prese, permettendomi di rimanere sempre aggiornato.</li></ol><h3>Video Dimostrativo</h3><p>Per mostrare l’efficacia e la fluidità dell’interazione con Lumi ho realizzato un <a href="https://vm.tiktok.com/ZGdew6psM">video</a> dimostrativo (dal video si evince perchè non ho intrapreso la carriera da regista)</p><p>Questo video evidenzia come, grazie a questa integrazione, Alexa non richieda più comandi anticipati dalla parola chiave “Alexa” (…finalmente!)rendenddo l’interazione molto più naturale e simile a una conversazione con un assistente umano.</p><h3>Conclusioni e Riflessioni</h3><h4>Vantaggi dell’Integrazione</h4><p>L’integrazione di GPT4-o di OpenAI con Alexa ha portato numerosi vantaggi, tra cui:</p><ol><li><strong>Automazione e Efficienza</strong>: Lumi automatizza compiti ripetitivi e dispendiosi in termini di tempo, liberandomi da noiose attività di gestione delle comunicazioni condominiali.</li><li><strong>Precisione e Affidabilità</strong>: Grazie all’intelligenza artificiale di GPT-4, Lumi è in grado di analizzare e interpretare le informazioni in modo accurato, riducendo al minimo gli errori umani.</li><li><strong>Esperienza Utente Migliorata</strong>: L’interazione senza la necessità di comandi anticipati dalla parola chiave “Alexa” rende l’esperienza utente molto più fluida e naturale.</li></ol><h4>Sfide Affrontate</h4><p>Nonostante i numerosi vantaggi, lo sviluppo di Lumi ha comportato alcune sfide:</p><ul><li><strong>Integrazione del Modello AI</strong>: L’integrazione di GPT4-o con Alexa ha richiesto una comprensione approfondita delle API e dei servizi offerti da entrambe le piattaforme e ho scoperto che la developer console ha molti aspetti da migliorare…</li><li><strong>Gestione della Sicurezza e della Privacy</strong>: Trattandosi di una skill privata che gestisce informazioni sensibili, è stato fondamentale garantire la sicurezza e la privacy dei dati trattati, anche per questo motivo ho preferito non pubblicarla.</li></ul><h4>Gestione del Codice e Limiti di GitHub</h4><p>Una delle limitazioni principali che ho incontrato riguarda la gestione del codice. Poiché l’ambiente di sviluppo per le Alexa-Hosted Skills non permette una sincronizzazione diretta con GitHub, non è possibile condividere tutto il codice in un repository pubblico. Questo significa che una parte significativa dello sviluppo è stata effettuata direttamente nel developer console di Alexa (eh già..).</p><p>Tuttavia, ho creato una baseline su GitHub che può essere utilizzato come base per sviluppare ulteriormente la skill. Questo skeleton include le configurazioni iniziali e alcune funzionalità di base per aiutare altri sviluppatori a partire con il piede giusto.</p><h4>Futuri Sviluppi</h4><p>Il successo di Lumi mi ha ispirato a considerare ulteriori sviluppi e miglioramenti, in primis, l’integrazione con altre piattaforme di gestione delle comunicazioni (outlook, slack,..) e dei calendari per una gestione ancora più centralizzata e efficiente.</p><h3>Conclusione</h3><p>Sviluppare Lumi è stata un’esperienza incredibilmente gratificante. Utilizzando le Alexa-Hosted Skills e la potenza di GPT4-o di OpenAI, sono riuscito a creare un assistente personale che semplifica la gestione delle comunicazioni condominiali, ricordandomi le scadenze e riassumendo i punti principali delle riunioni.</p><p>Lumi è diventato un alleato indispensabile nella mia vita quotidiana, dimostrando come la tecnologia possa realmente migliorare l’efficienza e la qualità della nostra vita.</p><h3>Contattami per Ulteriori Informazioni</h3><p>Se sei interessato a sviluppare una skill simile o hai domande riguardo il processo di sviluppo, non esitare a contattarmi.</p><p>Se sei qui, molto probabilemente ti interessa il codice, ecco il link al repository:</p><ul><li><strong>GitHub Repository</strong>: <a href="https://github.com/bsab/Alexa-AI-Skill">https://github.com/bsab/Alexa-AI-Skill</a></li></ul><p>Questo skeleton include le configurazioni iniziali e alcune funzionalità di base per aiutarti a partire con il piede giusto. Anche se non posso condividere tutto il codice a causa delle limitazioni dell’ambiente di sviluppo Alexa, spero che questa base ti sia utile per iniziare il tuo progetto.</p><h3>La tecnologia è in continua evoluzione, e le opportunità per migliorare la nostra vita quotidiana sono infinite. Sarebbe uno spreco non sfruttarle.</h3><p><em>Enjoyed this article? Support my work by </em><a href="http://buymeacoffee.com/thesab"><em>buying me a coffee</em></a><em>. Thank you!</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=194636a1a333" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[PandasAI: The Innovative Data Analysis Project Born from an Italian Engineer]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@thesab/pandasai-the-innovative-data-analysis-project-born-from-an-italian-engineer-f313d63e53a7?source=rss-e35bcd4a52e6------2"><img src="https://cdn-images-1.medium.com/max/704/0*cyWWxtxdD8imKhc2" width="704"></a></p><p class="medium-feed-snippet">Generative AI has emerged as a powerful technology with the ability to revolutionize how we interact with data and code. By incorporating&#x2026;</p><p class="medium-feed-link"><a href="https://medium.com/@thesab/pandasai-the-innovative-data-analysis-project-born-from-an-italian-engineer-f313d63e53a7?source=rss-e35bcd4a52e6------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://medium.com/@thesab/pandasai-the-innovative-data-analysis-project-born-from-an-italian-engineer-f313d63e53a7?source=rss-e35bcd4a52e6------2</link>
            <guid isPermaLink="false">https://medium.com/p/f313d63e53a7</guid>
            <dc:creator><![CDATA[Sab Severino]]></dc:creator>
            <pubDate>Mon, 30 Oct 2023 13:50:05 GMT</pubDate>
            <atom:updated>2024-11-04T12:55:39.848Z</atom:updated>
        </item>
    </channel>
</rss>