<?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 Gladia.io on Medium]]></title>
        <description><![CDATA[Stories by Gladia.io on Medium]]></description>
        <link>https://medium.com/@gladia.io?source=rss-8c8a5f2d8dd8------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*Vpo1Q1hx-jq0w7_uYNzH_A.jpeg</url>
            <title>Stories by Gladia.io on Medium</title>
            <link>https://medium.com/@gladia.io?source=rss-8c8a5f2d8dd8------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Mon, 16 Nov 2026 02:49:46 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@gladia.io/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[How Prompt Engineering is Making AI Available for Everyone]]></title>
            <link>https://medium.com/@gladia.io/how-prompt-engineering-is-making-ai-available-for-everyone-c2e2512e9219?source=rss-8c8a5f2d8dd8------2</link>
            <guid isPermaLink="false">https://medium.com/p/c2e2512e9219</guid>
            <category><![CDATA[openai]]></category>
            <category><![CDATA[large-language-models]]></category>
            <category><![CDATA[prompt-engineering]]></category>
            <dc:creator><![CDATA[Gladia.io]]></dc:creator>
            <pubDate>Fri, 10 Feb 2023 13:32:37 GMT</pubDate>
            <atom:updated>2023-02-10T13:32:37.147Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/880/1*5-ceZ9xT_pTedGPSQQWjnw.png" /></figure><p>Imagine a world where you could tell an application what you want, and it would take care of the rest. The artificial intelligence (AI) industry has been edging toward this world for decades, but technology is finally catching up to the vision. The incredible advances in natural language processing (NLP) — a subfield of AI — have been credited mainly to the emergence of large language models (LLMs). Trained on massive text corpora, these models can understand and process natural language texts. This power isn’t limited to written words; image generation is also a significant focus for LLM developers. A “prompt” is an integral part of the NLP process, as it tells the system how to interpret and respond to specific texts. A well-written prompt will create a pattern the algorithm can match against, helping it understand and respond appropriately to the input provided. This article will discuss what fast engineering is and how you can use it to execute downstream, domain-specific NLP tasks.</p><h3>Prompting 101: The Basics of Prompt Engineering</h3><p>In contrast to the conventional method of using a carefully curated training set, a new technique called prompt engineering is at the heart of many NLP applications. It prompts work by providing a pre-trained LLM with a snippet of text as input and expecting it to be appropriately complete. For example, while writing queries against the Google search engine algorithm may seem like a black-box process, search engine optimization (SEO) guides people to consume the content most relevant to their search intent. In the same way, prompt engineering is a technique that can help anyone get the best out of an LLM’s architecture. They are both about ensuring you get results that fulfill your intent. However, getting the intuition to know what prompt will make the application work better can be difficult. Trying different prompts and fine-tuning the LLM’s parameters is an effective way to obtain the most relevant result. How you express what it needs to do can dramatically alter the output. For example, we asked the GPT3 engine to translate a text with different prompts.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/430/0*-iICn-oUR2wzKvIj.png" /></figure><p><em>Prompt 1 (GPT3 generated the highlighted text)</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/408/0*LxNDh2CT4tMibIlS.png" /></figure><p><em>Prompt 2 (GPT3 generated the highlighted text)</em></p><p>The output text signifies a firm reliance on the input prompt because altering the request changed the output to suit the new set of languages (Turkish, German, and Finnish). Therefore, to get the LLM to do anything we want, we need to “engineer” the prompt until it generates the most relevant result, hence the term prompt engineering.</p><h3>Approaches to Prompt Engineering</h3><p>Now that you’ve seen how an LLM like GPT-3 can be made to translate texts, look at the translation prompts we tried earlier. You’ll notice we didn’t provide examples to show the model what we want (i.e., sample translations). This is because GPT-3 was able to learn many valuable tasks when OpenAI originally trained it.</p><p>However, there are two approaches to prompt engineering: zero-shot learning and few-shot learning.</p><p>Zero-shot learning refers to a system’s ability to generate a relevant output from a prompt with no examples. It requires the model to draw inferences and reason about what is being asked of it only from instructions embedded in the prompt.</p><p>Few-shot learning teaches an application of the nuances of a prompt by showing some examples to determine the appropriate context. For our translation example, that would mean including a few English sentences and the corresponding translations in the prompt. It enables the model to perform better on tasks that require a deep understanding of the context.</p><p>In this example, we attempt to guide the model toward factual answers by showing it how to respond to questions outside its knowledge base.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/624/0*h3L8YcSYG1aLSZ6V.png" /></figure><p><em>Few-shot learning (GPT3 generated the highlighted text)</em></p><p>We used the question marks to ask the model not to respond to words, phrases, or topics it does not know. The few-shot learning technique enables this application only to provide an objective natural response or refrain from answering.</p><h3>How Prompt Engineering can Transform Technical Processes</h3><p>Prompt engineering is more than just a neat idea; there are already products and services in the market based on this technology — the popular GPT3 and Dall. E LLMs from OpenAI use prompts to execute several language and image generation tasks.</p><p>Many companies have extensive databases of their items and customer records. Still, when it comes time to share that data with a customer relationship management (CRM) tool, it can be arduous to sort through a spreadsheet and enter the information one record at a time.</p><p>You may also have to perform regression analysis on the records (for example, estimating the relationship between two variables), which requires updating the spreadsheet file every time there’s a change. This process can become even more tedious as you try to enter new items into the database.</p><p>To make your data compatible with your CRM or perform regression analysis, you can run your file through a script designed for those purposes. Python is a popular programming language that lets you do this quickly, but you must be adept at writing Python code. But now, there’s another way. You can write a text-based prompt to execute any analysis on your dataset using an LLM with the right prompt.</p><h3>The Way Forward</h3><p>There’s no doubt that the large language models we have today have a remarkable grasp on the nuances of human language. Indeed, they perform better than any other existing technique by far. Their performance is so impressive that it can be tempting to believe they can perform any task. However, there may be tasks for which these models are not well-suited, and we need to fine-tune our inputs to ensure they perform as expected.</p><p><em>The illustration’s prompt is “a robot playing with letter cubes.”</em></p><p><em>—</em></p><p><em>Article originally published on Gladia.io in 2022.</em></p><p><em>Author: Jean-Louis Queguiner, Founder &amp; CEO Gladia.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c2e2512e9219" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AlphaTensor Explained]]></title>
            <link>https://medium.com/@gladia.io/alphatensor-explained-1808265299a3?source=rss-8c8a5f2d8dd8------2</link>
            <guid isPermaLink="false">https://medium.com/p/1808265299a3</guid>
            <category><![CDATA[deepmind]]></category>
            <category><![CDATA[google]]></category>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[mathematics]]></category>
            <category><![CDATA[computer-science]]></category>
            <dc:creator><![CDATA[Gladia.io]]></dc:creator>
            <pubDate>Fri, 10 Feb 2023 13:28:35 GMT</pubDate>
            <atom:updated>2023-02-10T13:33:11.678Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/880/1*noNREtdHskJul87bg2vKig.png" /></figure><p>Despite the prevalence of algorithms in today’s society, the process of discovering new ones is incredibly difficult. This is because humans are limited in what we can imagine and how to invent complex algorithms with our current scientific knowledge. DeepMind’s latest breakthrough, <a href="https://www.nature.com/articles/s41586-022-05172-4">Alphatensor (published in Nature on October 5th)</a> is groundbreaking because it circumvents these limitations.</p><h3>Introducing AlphaTensor</h3><p>Although it is relatively easy to multiply two matrices, multiplying them quickly and correctly is not a trivial task. This is where matrix multiplication algorithms come in: they are sets of rules or equations that help computers perform matrix multiplication quickly.</p><p>There are <a href="https://en.wikipedia.org/wiki/Matrix_multiplication_algorithm">many such algorithms,</a> and some have been around for decades. However, researchers from DeepMind have recently shown that Artificial Intelligence (AI) techniques can discover more efficient methods for performing matrix multiplication than we might think of on our own.</p><h3>The Layman’s Curtain-Raiser to Matrix Multiplication and AlphaTensor</h3><p>Mathematicians have long held the view that the conventional matrix multiplication algorithm is the most efficient, but German mathematician Volker Strassen startled the academic community in 1969 by demonstrating that there are actually better algorithms. His new method uses one less scalar multiplication when compared with the standard algorithm.</p><p><a href="https://www.deepmind.com/blog/discovering-novel-algorithms-with-alphatensor">DeepMind</a> has now leveraged AI to solve this classic problem by using cutting-edge techniques to find new algorithms, starting with matrix multiplication. AlphaTensor, a system that combines human intuition with deep learning models, uncovered new methods of multiplying matrices.</p><p>The significance of their discovery could feel “ordinary”, but what is unknown to many is that matrix multiplication is critical to the functioning of computers and other digital devices. This elementary math problem provides the foundation for everyday digital tasks like image processing, speech recognition, data compression, and many other functions.</p><p>The community got really excited about this announcement, proving the AlphaTensor is a major milestone that has been achieved.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Fte8oFTqd6AiHvSN.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*PCxivOTpCrB2WhCg.png" /></figure><h3>And Now, The Technical Bit.</h3><p>In 2015, Deepmind created a computer program that defeated a three-time world Go champion. The computer program used an AI technique referred to as <em>reinforcement learning</em>. The AI-enhanced software, AlphaGo, was even improved on in 2017 with AlphaZero.</p><p>Are you looking to learn more about reinforcement learning? There is a great <a href="https://www.analyticsvidhya.com/blog/2017/01/introduction-to-reinforcement-learning-implementation/">analogy</a> here.</p><p>The DeepMind researchers had to frame the matrix decomposition issue in a way that a model like AlphaZero could resolve. They found that by framing matrix decomposition as a single-player game (which they refer to as TensorGame), they could make it significantly related to the issues that AlphaZero had addressed.</p><p>The Deepmind team converted the problem of finding efficient algorithms for matrix multiplication into a single-player game. At each step of the game, the player selects how to combine different entries of the matrices to multiply. A score is assigned based on the number of selected operations required to reach the correct multiplication result.</p><p>Reinforcement learning techniques such as a reward system and penalties limit the number of moves the agent can make.</p><p>Like AlphaZero, AlphaTensor utilizes a deep neural network to direct a Monte Carlo tree search (MCTS) planning process. Taking a matrix as input, the network outputs a value and a policy that provides a distribution over potential actions. The value, however, furnishes an estimate of how the returns are distributed. In the concept of reinforcement learning, this return is the cumulative reward, starting from the current state.</p><p>In this game, a three-dimensional tensor (array of numbers) captures the distance between an incorrect algorithm and the correct one. The player can perform a set of allowed moves corresponding to algorithm instructions, with each move modifying the tensor to reduce its entities to zero. When the network succeeds at this task, it has proven that the algorithm is provably correct, and the number of steps it takes to zero out the tensor indicates how efficient that algorithm is.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/936/0*UbtRMR2tX5vmcWSi.png" /></figure><p><em>Alphatensor methodology. Image by DeepMind</em></p><p>AlphaTensor starts from the target tensor and uses the MCTS planner to choose the next action after each step. Finished games are used as feedback to improve the network parameters.</p><h3>The Bigger Picture: What Promise does AlphaTensor Hold?</h3><p>Matrix multiplication algorithms are the bread and butter of the computing industry. In fact, many companies have invested huge amounts of resources into developing ways to optimize matrix multiplication. This is because any improvement to the efficiency of matrix multiplication is likely to have a widespread impact on the industry, potentially saving huge amounts of time and money for several companies.</p><p>Looking past matrix multiplication, the benefits for humanity is potentially endless. With such complex operations as the discovery of new algorithms made easy, there will be no limit to how much progress we can make in fields such as medicine, chemistry, and physics.</p><p>In addition to helping us solve problems faster or better than ever before, Alphatensor should also help guide further research in complexity theory, which aims to determine the fastest algorithms for solving computational problems.</p><h3>Takeaway</h3><p>In most cases, designing an algorithm is a time-consuming and challenging process that requires a great deal of intuition and ingenuity. But with what can be referred to as the most notable breakthrough in algorithm discovery, the use of AI-designed algorithms to solve complex computational problems holds great promise.</p><p>Although it may seem counterintuitive, AI allows humans to design much better algorithms than before, as researchers from DeepMind showed. Their deep reinforcement learning-based matrix multiplication algorithm was more efficient than human-designed ones, which is a significant step forward in the field of algorithmic discovery.</p><h3>Going further</h3><p>We recommend a few resources:</p><ul><li><a href="https://github.com/deepmind/alphatensor">Google AlphaTensor’s Github</a></li><li><a href="https://www.deepmind.com/blog/discovering-novel-algorithms-with-alphatensor">Deepmind blog announcement</a></li><li><a href="https://www.reddit.com/r/slatestarcodex/comments/xwenaw/deepmind_uses_alphazero_to_improve_matrix/">An interesting thread on reddit</a></li><li><a href="https://towardsai.net/p/l/deepminds-alphatensor-deepminds-alphatensor-the-ai-that-is-reinventing-math">Towards AI blog</a></li><li><a href="https://www.youtube.com/watch?v=3N3Bl5AA5QU">Yannic Kilcher’s deep explanation</a></li></ul><p>And to finish, here’s a very nice thread from Herum Shandilya.</p><h3>Herumb Shandilya on Twitter: &quot;AlphaTensor&#39;s release showed us a unique use case of RL for algorithm discovery and the community seems to be really thrilled about it. But how does it work? What exactly did they deliver? What does this mean?AlphaTensor, a 🧵 / Twitter&quot;</h3><p>AlphaTensor&#39;s release showed us a unique use case of RL for algorithm discovery and the community seems to be really thrilled about it. But how does it work? What exactly did they deliver? What does this mean?AlphaTensor, a 🧵</p><p><em>—</em></p><p><em>Article originally published on Gladia.io in 2022.</em></p><p><em>Author: Jean-Louis Queguiner, Founder &amp; CEO Gladia.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1808265299a3" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Google’s Imagen Video Generation Model]]></title>
            <link>https://medium.com/@gladia.io/googles-imagen-video-generation-model-30f71d32f70f?source=rss-8c8a5f2d8dd8------2</link>
            <guid isPermaLink="false">https://medium.com/p/30f71d32f70f</guid>
            <category><![CDATA[deepmind-ai]]></category>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[video-generation]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[generative-ai]]></category>
            <dc:creator><![CDATA[Gladia.io]]></dc:creator>
            <pubDate>Fri, 10 Feb 2023 13:24:25 GMT</pubDate>
            <atom:updated>2023-02-10T13:33:44.223Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*fdiB6wuPRoYo9BCTDcQ_6Q.jpeg" /></figure><p>Google released a new paper around image video generation using Diffusion models (<a href="https://imagen.research.google/video/paper.pdf">IMAGEN VIDEO: HIGH DEFINITION VIDEO GENERATION WITH DIFFUSION MODELS</a>).</p><p>The results are really impressive, especially considering that Facebook released a model almost at the same moment (see our <a href="https://www.gladia.io/blog/text-to-video-generation-how-it-works">previous blog</a>).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/642/1*uMc6MQvvpwaRBEhBF1i-lQ.png" /></figure><p><em>Source: </em><a href="https://imagen.research.google/video/"><em>Imagen video website</em></a></p><p>Here is a quick summary of the technics proposed:</p><p>1. Imagen Video, a text-conditional video generation system based on a cascade of video diffusion models.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/747/0*-WugTGY4kabTDIYx.png" /></figure><p>2. The Data gets into the Input Text Prompt and then gets into the T5-XXL model to perform the contextual encoding of the text prompt. This allows the system to have a deeper language understanding.</p><p>3. It’s then sent to the Base video model that generates the base structure of the video with a low resolution and a low frame rate</p><p>4. Then a series of Temporal Super-Resolution (TSR) and Spatial Super-Resolution (SSR). TSR is filling the gap between the frame to get a higher frame rate while SSR is used to improve the resolution guided by the Input text prompt.</p><p>3. The team describes how they scale up the system as a high-definition text-to-video model, including design decisions such as the choice of fully-convolutional temporal and spatial super-resolution models at specific resolutions and the selection of parameterization of the diffusion models.</p><p>5. Finally, the researchers applied progressive distillation to the video models with classifier-free guidance for fast, high-quality sampling.</p><p><em>—</em></p><p><em>Article originally published on Gladia.io in 2022.</em></p><p><em>Author: Jean-Louis Queguiner, Founder &amp; CEO Gladia.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=30f71d32f70f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[OpenAI’s Open Source Whisper is changing the speech recognition game]]></title>
            <link>https://medium.com/@gladia.io/openais-open-source-whisper-is-changing-the-speech-recognition-game-ebc95695bfe9?source=rss-8c8a5f2d8dd8------2</link>
            <guid isPermaLink="false">https://medium.com/p/ebc95695bfe9</guid>
            <category><![CDATA[video-generation]]></category>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[deepmind]]></category>
            <category><![CDATA[google]]></category>
            <dc:creator><![CDATA[Gladia.io]]></dc:creator>
            <pubDate>Fri, 10 Feb 2023 13:21:33 GMT</pubDate>
            <atom:updated>2023-02-10T13:53:29.608Z</atom:updated>
            <content:encoded><![CDATA[<p>Last week OpenAI released <a href="https://openai.com/blog/whisper/">Whisper</a> their first open-source model.</p><p>This model has two elements that are making it special:</p><ul><li>First, Whisper is the first <strong>multi-lingual</strong> open-source model.</li><li>Secondly, it&#39;s the first open-source model that is smartly combining <strong>multimodality</strong>: speech recognition and language understanding.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/514/1*hdGEHo5X7oUyApfurnLvwg.png" /></figure><p>From the inference standpoint this is making a huge difference:</p><ol><li>Reduce Latency: on the long run being multi-language reduces the latency of the Automated Speech Recognition System) as it allows the system to skip the speech-language classification step before passing the audio to the language-specific speech recognition system.</li><li>Less accent sensitivity: having only 1 model for multiple languages makes it potentially less sensitive to accents (this seems to be confirmed by the first qualitative analysis done by the community).</li><li>Context awareness: this topic has been a chimera for years in the speech recognition space. Combining speech recognition with the transformers’ mask technic allows for an increase dramatically the accuracy of the speech recognition system in the future.</li></ol><p>We wish for Whisper to create traction toward a multimodal model applied to speech recognition.</p><p>To conclude, here’s a great video about Whisper by Louis Bouchard.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FuFOkMme19Zs%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DuFOkMme19Zs&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FuFOkMme19Zs%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/7712e7b7b305e4ae6ef8cb297414f6c5/href">https://medium.com/media/7712e7b7b305e4ae6ef8cb297414f6c5/href</a></iframe><p>Prompt used for the illustration : <em>a photo of a robot writing with a microphone in a shape of a pen</em></p><p><em>—</em></p><p><em>Article originally published on Gladia.io in 2022. It’s a shortened version, with the original audio files and API embeds edited out due to Medium’s format limitations.</em></p><p><em>Author: Jean-Louis Queguiner, Founder &amp; CEO Gladia.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ebc95695bfe9" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Text-To-Video Generation: Here’s How It Works]]></title>
            <link>https://medium.com/@gladia.io/text-to-video-generation-heres-how-it-works-3c8e5f19628?source=rss-8c8a5f2d8dd8------2</link>
            <guid isPermaLink="false">https://medium.com/p/3c8e5f19628</guid>
            <category><![CDATA[video-generation]]></category>
            <category><![CDATA[text-to-video]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Gladia.io]]></dc:creator>
            <pubDate>Fri, 10 Feb 2023 13:16:15 GMT</pubDate>
            <atom:updated>2023-02-10T13:54:58.873Z</atom:updated>
            <content:encoded><![CDATA[<p>In a recent paper, <a href="https://arxiv.org/abs/2209.14792">“Make a Video: Text to Video Generation without Text-Video Data”</a>, Meta AI researchers propose a new approach to Text-to-Video generation that does not require text-video data.</p><p>The researchers’ intuition is that the model can understand how the world moves by learning from paired text-image data and unsupervised video footage</p><p>Their three-stage approach used involves training a text-to-image model, then training it on video data in an unsupervised fashion, and finally fine-tuning it with a frame interpolation network.</p><p>The results of the researchers’ experiments show that their approach outperforms previous baselines on two standard video generation benchmarks: MSR-VTT and cog video.</p><p>In addition, their approach generates videos that are more faithful to the input text of higher quality according to human evaluation.</p><p>The approach proposed in this paper represents a significant advance in the text-to-video generation and opens up new possibilities for applications of this technology.</p><p>Here is a great video made by <a href="https://www.youtube.com/c/TheAIEpiphany">Aleksa Gordić — The AI Epiphany</a> on the subject:</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FMmAJk2BD6WA%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DMmAJk2BD6WA&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FMmAJk2BD6WA%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/ab868ed033c6e28f4bd4f9dedb4cda7e/href">https://medium.com/media/ab868ed033c6e28f4bd4f9dedb4cda7e/href</a></iframe><p>Prompt used for the illustration:</p><p><em>a filmmaker robot recording a movie with a camera, futuristic style</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/636/1*QvJmSed4sJrfJsbQuumNYQ.png" /></figure><p><em>—</em></p><p><em>Article originally published on Gladia.io in 2022. It’s a shortened version, with the original audio files and API embeds edited out due to Medium’s format limitations.</em></p><p><em>Author: Jean-Louis Queguiner, Founder &amp; CEO Gladia.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3c8e5f19628" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Why you should expect a price drop for Nvidia P100 this year]]></title>
            <link>https://medium.com/@gladia.io/why-you-should-expect-a-price-drop-for-nvidia-p100-this-year-c4373328ba09?source=rss-8c8a5f2d8dd8------2</link>
            <guid isPermaLink="false">https://medium.com/p/c4373328ba09</guid>
            <category><![CDATA[computer-science]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[nvidia]]></category>
            <dc:creator><![CDATA[Gladia.io]]></dc:creator>
            <pubDate>Fri, 10 Feb 2023 13:12:27 GMT</pubDate>
            <atom:updated>2023-02-10T13:46:51.719Z</atom:updated>
            <content:encoded><![CDATA[<p>Back to the linear regression behind NVidia Business Model</p><p>As presented in our previous blog post, <a href="https://gladia.io/why-you-should-expect-a-price-drop-for-nvidia-p100-this-year/the-linear-regression-behind-nvidia-business-model">The linear regression behind Nvidia business model</a>, we will try in this blog post to predict the price drop of the Nvidia P100 on the Google Cloud Platform that should occur later in 2022.</p><p>Considering a price per TFlops (single precision) on GCP to be 0.17 $/TFlops (US EAST) and considering the price of the K80 that finished its amortization around 2017 (release date of the card was 2012 and <a href="https://developer.nvidia.com/blog/google-cloud-lowers-the-price-of-nvidia-tesla-gpus/">GCP price lowering was announced by Nvidia in 2017</a>) we are suspecting a 5 years amortization business model when it comes to GPUs.</p><h3>P100 is reaching its 5 years amortization anniversary</h3><p>The P100 model was released in 2016, which means that P100 should have reached its amortization already or will happen soon.</p><p>The price drop for K80 was said to be up to 36% the original price must have been …</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/200/1*ddtRdKINDsJFU_TXqV_JhA.gif" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*InZTSt4VV0NpDmJ_2Hdoqg.png" /></figure><p><strong><em>Note:</em></strong> If you want to know how to get the 0.17 $/TFlops/hour feel free to check the previous blog post, The linear regression behind Nvidia business model.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xk8ka0xWjHdzt1W-Az5dWw.png" /></figure><p>I love when numbers add up .. and YES, the new discounted price is 0.10$/h/TFlops meaning that the depreciation rate is close to 33%, which corresponds to the +30% discount advertised.</p><h3>Conclusions</h3><p>So P100 should drop between 30% and 36%, reaching a cost of 0.10$/h/TFlops</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/392/1*ap7GZhIjphkBF0Uc98q6rw.gif" /></figure><p>See you in a few months to see if my prediction was correct!</p><p><em>—</em></p><p><em>Article originally published on Gladia.io in 2022. It’s a shortened version, with the original audio files and API embeds edited out due to Medium’s format limitations.</em></p><p><em>Author: Jean-Louis Queguiner, Founder &amp; CEO Gladia.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c4373328ba09" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>