Thomas Hufener

We are still very early in machine learning for protein engineering. A core belief we have held at Cradle is that the only way to build a high quality product with best in class generative models is to invest in a low-latency, high throughput lab. This lab allows us to build 'foundational datasets', a/b test new releases of our models and demonstrate generalisation across many protein modalities (i.e. enzymes, antibodies, car-t's, peptides etc). Two years ago we set a high bar: we wanted our lab to be able to complete an experimental round in 2 weeks, at 384 throughput for many different assays. This meant trying to reduce latency in every step of our DBTL cycle. In this blog series we share the methods we developed. This month: express and purify 384 variants in just 3 days of relatively light lab work and not breaking the bank. The protocol is added to the blog-post. Give it a try and feel feel to ask us questions! Follow Cradle for more methods being published to speed up your research and development. The blogpost with protocol link is in the comments!

222

10 commentaren

Are open LLM models truly open? 🤔 It seems they might not be. Researchers at Radboud University assessed over 40 large language models and six text-to-image generators, including OpenAI’s #ChatGPT and #DALL·E2. They evaluated 14 characteristics concerning Availability (source code, pretraining data, base weights, fine-tuning data, etc.), Documentation (code, architecture, preprint paper, etc.), and Access (downloadable package and open API). Results: Of the language models, OLMo7B Instruct and LLM360/AmberChat were the only fully available. BigScience’s BLOOMZ was the only model that was fully open in documentation. Some prominent “open” models scored less well. Alibaba Group’s Qwen 1.5, Cohere’s Command R+, and Google’s Gemma-7B Instruct were judged closed or partially open for most characteristics. Neither Meta’s Llama 2 Chat nor Llama 3 Instruct achieved any open marks. Larger corporations often claim openness while providing limited transparency compared to smaller entities that adhere more closely to open-source principles. The study highlights the risks of relying solely on licensing to determine openness, as it can be manipulated to meet minimal standards without genuine transparency. It's time regulatory bodies adopt composite and gradient approaches to assess openness. Companies/Businesses adopting LLM models need to scrutinize and check for detailed documentation to ensure that AI systems labelled as open-source genuinely adhere to open-source principles. How can we ensure the AI systems we rely on are truly transparent and accountable? Share your thoughts in the comments. 👇

10

1 commentaar

At PyData Amsterdam 2024, join Vitalie Spinu for his talk, "Drift Detection on Irregular Time Series with Multiple Non-Uniform Seasonal Patterns Using MIST and DTW Algorithms." In this session, Vitalie will explore the complex challenges of detecting drifts in irregular time series with non-uniform seasonal patterns, like end-of-month or holiday effects. Vitalie will introduce a powerful four-step approach to drift detection, combining MIST (Multiple Irregular Seasonalities and Trend decomposition) and DTW (Dynamic Time Warping) algorithms. He’ll demonstrate how this method can efficiently handle issues like missing data, irregular time points, and complex seasonal patterns, making it adaptable across various domains. Don’t miss this opportunity to learn how to manage long-term drifts in large-scale datasets with millions of time series, and discover how to apply these techniques directly to raw data with minimal pre-processing. 🎟 https://lnkd.in/eEtC9ESU #PyDataAmsterdam2024 #MachineLearning #AI #LLM #PyData2024 #data #datascience #python

28

𝗡𝗘𝗪 𝗕𝗟𝗢𝗚 𝗣𝗢𝗦𝗧: 𝗘𝘅𝗽𝗹𝗼𝗿𝗶𝗻𝗴 - 𝗗𝗮𝘁𝗮-𝗔𝘄𝗮𝗿𝗲 𝗢𝗿𝗰𝗵𝗲𝘀𝘁𝗿𝗮𝘁𝗶𝗼𝗻 𝘄𝗶𝘁𝗵 𝗔𝗽𝗮𝗰𝗵𝗲 𝗔𝗶𝗿𝗳𝗹𝗼𝘄 𝗗𝗮𝘁𝗮𝘀𝗲𝘁𝘀 Welcome to “Exploring: Dataset-aware Orchestration”, where we delve into the power of “Airflow Datasets”. In this post, our colleague Kenny Peeters invites you to his journey throughout a development-need, and the expansive scope, that led to the following article. Read more about it via the link below: https://ow.ly/7KoR50RqFQr #AirflowDatasets #Orchestration #Cubis

4

I've talked about this before: Frontier Models > Older Model Methodology/ALgos with (years of data/years of research). See Erik Meijer from Microsoft talking about how frontier LLM doing much better than older research and algos. Other examples: latest and the greatest traditional OCR (Object Character Recognition) vs LLM-based visual recognition in OpenAI's 4o-mini. I've also anecdotally heard of Spotify's existing recommendations ML algo with access to years of music-listening user data giving lower quality recommendations than uploading a screenshot of favourites in Spotify app to an LLM and asking for recommendations. Insightful talk from Erik Meijer (MSFT VB, C#, LINQ): link in comments. The best part title of his talk: "Hinton's Nightmare is My Dream" 🤣

10

5 commentaren

Got inspired by a dutch podcast that creates stories from wikipedia pages. I used gpt-4o and their text-to-speech model to build something that sounded human. Of course it's not as great as the original podcast but this tech works way better than expected! Try it yourself 👇 Huggingface Space: https://lnkd.in/eKdT3XBY How did I do it? - Use <break time="1s"/> tag on strategic places to create breaks. - The TTS model is smart enough to understand these tags! - Prompt to use colloquialism english. - Use filler words like "um", "well", ... Bonus: I also generated an intro song using facebooks "MusicGen" and tweaked it a bit with with online tool called "AudioMass" Tools I used: AudioMass: https://audiomass.co/ MusicGen: https://lnkd.in/eAfjHZzA If your dutch speaking; the original podcast is called "Nooit Geweten", I can highly recommend: https://lnkd.in/ex66X6_a #gpt-4o #text-to-speech #tts #huggingface #spaces #podcast

14

2 commentaren

In PyData Eindhoven definitely the building of trusted Generative AI applications is the main theme. Not only the building of Retrieval Augmented Generation pipelines is being discussed, but also, how to make them performant and traceable. A lot of the discussions made me remember some talks by Dr. Chris Hillman (Who you should definitely follow), and this article by our colleague Chetan Hirapara (Who you should also definitely follow). https://lnkd.in/g5JYrhHX

32

2 commentaren

🚀 Exciting developments in Large Language Models! Our latest blog post, written by Jeroen Overschie, discusses Retrieval Augmented Generation (RAG), a technique that transforms data access and reduces hallucinations. 🗝️ Key Insights: Level 1️⃣ - Basic RAG - Fundamentals of document retrieval and answer generation. Level 2️⃣ - Hybrid Search - Combining vector and keyword search for better accuracy. Level 3️⃣ - Advanced Data Formats - Handling complex formats like HTML, Word, and PDF. Level 4️⃣ - Multimodal Models - Using models that understand text, audio, and images. RAG is rapidly evolving in the industry. Whether you're a seasoned professional or just embarking on your journey, this post provides actionable steps for implementing RAG. ⛓️ Read more here: https://hubs.ly/Q02LptM40. #ai #machinelearning #rag #xebia

68

3 commentaren

✅️ #Reference van Rooij, I., Guest, O., Adolfi, F. et al. Reclaiming AI as a Theoretical Tool for Cognitive Science. Comput Brain Behav (2024). https://lnkd.in/d8typUgS A few days ago, the #SamAltman prediction of an #ASI "in a few thousand days" was made, but what about the #AGI? The Artificial General Intelligence, such as #AI as smart as the human brain, seems instead in each AI Company announcement to be just around the corner, closer than ever, but in reality, that's questionable. Let me clarify from the beginning why #ThisPaperMatters and you can't miss it: 🗣 "Despite the current hype surrounding "impending" AGI, this practical infeasibility actually fits very well with what we observe (for example, running out of quality training data and the non-human-like performance of AI systems when tested rigorously)." The latest publication from Computational Brain & Behavior by the paper's lead author, Prof. Van Rooij (Radboud University ), pours cold water on the hype and marketing promoting the relentless pursuit of AI. 💧 Frequently in my posts, you have found how that challenge is #overshadowing human inalienable principles and polarizing people against Institutions where regulations are protecting privacy and these principles simply because their work is "slowing down" the diffusion or adoption of the "latest features," outlined for achieving human-like performance in "reasoning" or breaking down concepts to emulate the underlying mechanisms of the #CognitiveScience of our brain. 💡 #What do we really want to achieve? And #why? 🤷 That's not at all a "wow" effect, as with any new paper I share, but to point out and underline how researchers are well aware of the #complexity of creating human-like or human-level artificial intelligence through current ML approaches is practically #infeasible, despite the current hype surrounding AGI. 💡The risk, in my opinion, is to limit or simply reframe human thinking as a #downgrade simply because an AI, according to some #benchmarks, can be "smarter" and "faster" at solving quizzes. So, is our life just a #puzzle? 🧩 ✅️ And more: ➡️ The #Makeism which is the idea that cognition can be understood as a form of computation: known as "#SyntheticMethodology" or "understanding by design and building" it refers to the quote: 🗣"What I cannot create, I do not understand" by physicist Richard Feynman. ➡️ #Intractability ✔️AI-by-Learning is NP-hard ✔️Resource requirements grow exponentially with input size ✔️Even for moderate inputs (e.g., 15-min!), required samples exceed atoms in universe ✔️"Big Data" is insufficient to cover the astronomical space of possible situations ✔️AI companies are running out of usable data ✔️Datasets becoming more homogeneous, not more complex 💡Isn't it perhaps the case to reconsider the role of AI in cognitive science as a tool to understand the natural cognition rather than replicate it? Let's continue in the comments.👇 #Psychology #Cognitive #AI

44

6 commentaren

Your contribution for the evening 🌖: sodadata/soda-core🥤 Co-founded by Maarten Masschelein and Tom Baeyens, Soda is a data quality platform built for the modern data stack. Here’s why you should contribute to this repo: 1️⃣ With a growing repo now at ~1,700 stars, contributing to the soda-core repo allows you to refine your skills in data quality management, gain hands-on experience with the Soda Checks Language (SodaCL) and learn how to implement practical data quality tests across various data sources. You also have the opportunity to connect with other data professionals, expanding your network and opening doors to working opportunities. 2️⃣ Maintainers ensure a response rate of 100%, with typically a median PR merge time of around 3 hours. This guarantees that your contributions receive the attention they deserve. 🩶 3️⃣ The repo is more accessible. At Quira, we rank Developers by their DevRank (more on this here: https://lnkd.in/eNtrs97F). We found that the median developer working on the repo belongs to the 90th percentile of GitHub developers. For reference, many of the open-source companies we reported on have a median contributor level upwards of the 95th percentile. This means that you have a real opportunity to make an impact. ⚡️ If you want to help improve a pivotal tool for many organisations in ensuring data integrity, you can check more of their open-source stats on Quira (https://lnkd.in/egxYTC4N) or access the GitHub page: https://lnkd.in/e2r3iazu 📖

6

🌟 Our MVP Marc Lelijveld is currently at #MSIgnite and these are his highlights of day 1! Marc attended the keynote of Satya Nadella, who shared some very exciting announcements, including: ▪️ New AI Agents in Copilot: Autonomous AI-driven agents to handle tasks like customer returns and shipping invoices. ▪️ Microsoft's new AI Chips: Proprietary data center infrastructure chips aimed at accelerating AI applications and improving data security ▪️ Windows 365 Link: A Cloud-based mini PC designed for seamless remote working. Marc's favourite announcement? The launch of SQL Databases in Fabric! What's yours? Let us know in the comments below 👇 #MSIgnite2024 #TechInnovation #AI #MicrosoftFabric #Azure #Cloud

21

1 commentaar

🚨 Free ticket for MLCon 🚨 👉 Comment to this post with your name and we add you to the raffle. We will pick one person to join us by Thursday 20/06. Want to have the latest insights into #MachineLearning & #AI ? Join Structize AI at MLCon where our CTO Pieter Buteneers together with many other experts give updates on the latest trends.

“If data is the new gold, then you could say reliable and authentic data is the new platinum.” – Frederik Rosseel, CEO at Docbyte Maxime Vermeir and Frederik Rosseel unpack the importance of quality, integrity and authenticity in the data used to train AI models in this episode of the AI Pulse Podcast by ABBYY. Want to hear these AI experts explore the role of business data in ensuring trustworthy AI? Check out their conversation here: https://hubs.li/Q02w2NrK0 #AI #ArtificialIntelligence #IntelligentAutomation #Data #Privacy #TrustworthyAI #GenerativeAI #LLMs #Podcast

25

Mere weeks after Devin (devin.ai) was launched as the 'First AI Software Engineer', several open source equivalents have emerged, such as OpenDevin. What makes OpenDevin extremely interesting is that you can achieve the same functionality as Devin using whatever local LLMs you want - it's also free. The rapid emergence of open source equivalents to a well funded commercial #AI offerings is happening at a blistering pace. There's even an open source perplexity.ai equivalent that you can run completely local to your machine with private data. As a VC, how would you react to seeing an open source project emerge that completely mimics the functionality of a venture you just backed, but with the added benefit of its users having complete control over their institutional data? https://lnkd.in/gyYkNJaz

2

⛓️ Last week we introduced Chains, a framework for building and orchestrating compound AI workflows. 👀 Interested in how we built Chains, or what makes it so powerful? Learn more in our new technical deep-dive from Marius Killinger and Rachel Rapp: https://lnkd.in/eBshFqiH You’ll learn: 🔗 The philosophy and design principles behind Chains 🔗 The technical details about how Chains is built 🔗 How our customers are leveraging Chains to increase performance and lower costs In practice, we’ve already seen our customers’ processing times halved and GPU utilization improve 6x due to the efficient resource allocation and optimal scaling that Chains enables. 🚀 Have a question about Baseten Chains? Reach out! And don’t miss Marius’ live webinar on July 18th covering all of the above and more: https://lnkd.in/eRHYsJXP

23

𝗗𝗮𝘁𝗮𝗡𝗼𝗿𝘁𝗵 𝗚𝗼𝗲𝘀 𝘁𝗼 𝘁𝗵𝗲 𝗔𝗜𝗚𝗿𝘂𝗻𝗻 𝗖𝗼𝗻𝗳𝗲𝗿𝗲𝗻𝗰𝗲! 🚀 On Friday, November 29th, Chris van Riemsdijk, AI Consultant at DataNorth AI, will deliver an insightful session on “𝗚𝗼𝗶𝗻𝗴 𝗕𝗲𝘆𝗼𝗻𝗱 𝗧𝗮𝗯𝘂𝗹𝗮𝗿 𝗗𝗮𝘁𝗮 𝘄𝗶𝘁𝗵 𝗚𝗿𝗮𝗽𝗵𝘀.” In this session, Chris will explore how Graph Neural Networks (GNNs) revolutionize AI by going beyond traditional tabular data formats. If your work involves complex, interconnected systems like social networks, healthcare, or infrastructure, this talk is a must-attend! 🌟 𝗪𝗵𝗮𝘁 𝗬𝗼𝘂’𝗹𝗹 𝗟𝗲𝗮𝗿𝗻: • The fundamentals of GNN architectures and how they differ from conventional deep learning models. • How GNNs can tackle challenges like node classification, link prediction, and graph generation. • Practical insights to leverage GNNs for real-world projects involving graph-structured data. 🔍  Find out more here: https://aigrunn.org/ 𝗘𝘃𝗲𝗻𝘁 𝗗𝗲𝘁𝗮𝗶𝗹𝘀: 📅 𝗪𝗵𝗲𝗻: November 29th, 10:15 - 10:45 📍 𝗪𝗵𝗲𝗿𝗲: Forum Groningen 🎟️ 𝗧𝗶𝗰𝗸𝗲𝘁𝘀: Sold out! 💡 Looking forward to seeing you for this informative session into graph-based data structures and AI! Organised by: Stekz, (Berco Beute) Sponsored by: Gemeente Groningen, TKP Pensioen, TVM verzekeringen, See Tickets Benelux, Weaviate, Voys Nederland n8n, Tardis Research, CropX Sopra Steria, HackerOne. #AIGrunn #GraphNeuralNetworks #DataNorth #AI #Innovation

23

2 commentaren

Explore the transformative impact of structured data on Generative AI (GenAI) with our comprehensive study led by Kurt Janssens, AI Product Owner, and Jonas De Keuster, VP of Product Marketing at VaultSpeed. 🚀 🤖 Discover how optimized data organization enhances GenAI performance, reducing errors and improving response accuracy. Our research, featuring ChatGPT-4 experiments and a Retrieval-Augmented Generation (RAG) setup, reveals critical insights for businesses navigating AI-driven decision-making. Gain actionable strategies to leverage structured data effectively and unlock GenAI's full potential. 🔑 🔓 Click to read more: https://lnkd.in/eqAJxEMZ

32

1 commentaar