David Cardozo

My second official Microsoft x Mila - Quebec Artificial Intelligence Institute research collaboration during my PhD led to this fresh work, 𝐋𝐞𝐚𝐫𝐧𝐢𝐧𝐠 𝐭𝐨 𝐄𝐱𝐭𝐫𝐚𝐜𝐭 𝐒𝐭𝐫𝐮𝐜𝐭𝐮𝐫𝐞𝐝 𝐄𝐧𝐭𝐢𝐭𝐢𝐞𝐬 𝐔𝐬𝐢𝐧𝐠 𝐋𝐚𝐧𝐠𝐮𝐚𝐠𝐞 𝐌𝐨𝐝𝐞𝐥𝐬, being accepted to the 𝐄𝐌𝐍𝐋𝐏 𝐌𝐚𝐢𝐧 𝐂𝐨𝐧𝐟𝐞𝐫𝐞𝐧𝐜𝐞 2024. This work wouldn’t have been possible without the teamwork (Liana Mikaelyan, Alexander Meulemans, Steve Liu, James Hensman), especially the efforts of my co-first author Ye Yuan and the project mentor Bhaskar Mitra! Hope to see old and new friends in Miami 🏖 this November! Entity extraction is a key challenge in NLP, but the traditional triplet-centric approach limited broader evaluations. In our work, we redefined the task with an entity-centric perspective, better catering to downstream applications and allowing for more comprehensive evaluations. We also introduced the Approximate Entity Set OverlaP (AESOP) metric, which aligns more closely with human evaluations. On top of that, we developed a multi-stage entity extraction model leveraging language models, enhancing both its effectiveness and efficiency. For more details, please check our paper: https://lnkd.in/ekmwSvfp

108

4 Comments

I just got my own copy of “Building LLMs for Production” by Louis-François Bouchard and Louie Peters. As I am getting more into applications of GenAI I expect this book to become one of my go-to resources in the upcoming months. https://amzn.to/3Yv5FmL #AI #GenAI #LLMs

2,251

51 Comments

I agree with Yann LeCun, accedemia is not the only place for computer science PhDs, industry offers more opportunities in AI/ML reseach, and is more rewarding. #ai #machinelearning #datascience #entrepreneurship #startups

8

Optimizing Response Times for GenAI Applications. The LLM latency guidebook is here ! https://lnkd.in/eviexaiN

9

Vancouver AI people: I'm hosting an event! I'm founding the Vancouver chapter of The Ai Salon and will be hosting our inaugural session on Thursday, November 28. The Ai Salon is a community founded in San Francisco that brings together scientists, founders, builders, and the generally curious to hold small-sized group conversations on the sociological, economic, cultural, and philosophical impacts of AI. Our inaugural event: Autonomous Scientific Discovery Will we ever be able to build an AI scientist? What would that look like? Is the genius and creativity of scientific breakthroughs something AI could ever replicate? What is the significance of the recent Nobel Prizes in physics and chemistry awarded for work on AI? Let's discuss. This will be an in-person event. Details and registration information at the link below. https://lu.ma/u1px71f0

38

3 Comments

Very interesting to see examples of how LLMs seem to encode specific features and react to subtle changes in those features. Also interesting how much this reads like a paper reverse engineering a biological process.

4

⚡ Just a chill (very intense) guy in Vancouver for #NeurIPS... 🎉 🚨 Presenting not one, not two, but THREE papers that the Enkrypt AI team and I authored this weekend at the NeurIPS workshops! 🧠💥 Here’s what’s on deck from the Enkrypt AI team: 1️⃣ Red Teaming Gen AI: The SAGE-RT Synthetic Data Generation Process 2️⃣ Pluralistic Alignment: SAGE-RT Dataset's Impact on Model Safety Alignment 3️⃣ Safe Gen AI: Bias uncovered in over 50 LLMs (yes, it’s wild!) If you're here and want to chat about AI, AI safety, or just share some workshop wisdom, let’s connect! 📩 Let’s make some brilliant conversations happen. 🚀 Reach out if you're here! Let's talk all things AI and AI Safety! 🚀 #NeurIPS #NIPS #AISafety #AISecurity #Vancouver #Enkrypt AI

51

Can you be someone as honest as the harsh reality, politically correct, fun at the same time, and also one of the most knowledgeable in artificial intelligence? Only if you are French and your name is Yann LeCun ! Take a look to this master piece from Yann 👇 <sarcasm> Come work at ClosedAI. With AGI just around the corner, your shares will be worth 42 sextillionnollars. We can claw back your vested shares if you quit, unless you sign a non-disparagement agreement. Oh wait, sorry, we didn't realize our contract was this harsh until someone published an article on it. You can keep them. We're worth 42 octillionnollars, but since we're a private company, you can't really sell your shares. Also, we can actually stop you from selling them. And you can't say a word about what it is we're working on. If you do, we'll fire you within the hour, claw back your shares, and you won't be able to say *anything*. Don't take it personally but, you know, we're soooo far ahead of everyone else and AI is soooo dangerous in the hands of the unwashed masses. It's safe only if *we* do it. </sarcasm>

35

10 Comments

I encourage you all to try the demo (link at bottom) - mind blown!

1

Are you at NeurIPS 2024? Check out Google Research & Google DeepMind booths to see the latest AI developments. The coolest demos and results will be shown!

15

2 Comments

My new favourite book 🤩 A must-have for anyone who want to build LLMs Very practical! Thanks Louis-François Bouchard and Louie Peters P.S: but please, make a kindle version 🥹 #dataengineering #llm #ai #dataengineer

143

6 Comments

Teasing the notion of Artificial General Intelligence (AGI) by referring to new code simulating human reasoning is not helpful. Yann Lecun, several days back, presented the process simulating human reasoning in terminology not likely to evoke the AGI question and quagmire. Lecun, also several months back alluded to the level of human intelligence computer software and hardware can deliver, today, as less than the intelligence of a cat. I, for one would like to read more accurate copy on topics like "reasoning". Do other readers agree? #agi #llama3 #llms #humanlikethinking #meta

1

A few facts about AGI from Leopold Aschenbrenner's paper on AI's future. As technology evolves faster, it becomes more interesting to discuss its influence 🧐

7

I’ve always been fascinated by graphs. In my computer science classes, my favorite subjects were graphs and automata. I loved how we could use graph structures to represent all kinds of connections. Now, at vestiaire.io, we are using Neo4j, and I love it. So, I decided to write a post that explains why graph databases are cool. Check it out here: https://lnkd.in/ewXAjJX8 #neo4j #graph #graphdatabase

6

HOW TAKING A BREAK SPARKED THE COOLEST IDEA IN DEEP LEARNING In 2014, Ian Goodfellow, pursuing his PhD at the University of Montreal, was at a bar with friends celebrating a lab-mate’s new job at the company named DeepMind. As the drinks flowed and the conversation turned to AI, a debate arose about how to create a machine capable of generating photorealistic images (Remember this was 2014, DALL-E, Midjourney didnt exist publicly). Goodfellow’s colleagues proposed various complex solutions, but he, fueled by craft beer, suggested a radically different approach: a neural network that learned from another neural network. Here's how he thought: - One network would generate an image and attempt to fool the second network into believing it was real. - The second network would identify flaws in the generated image, providing feedback to the first. - This adversarial process would continue until the first network created images indistinguishable from real photographs. Goodfellow’s friends, convinced this approach was overly complicated, dismissed the idea. A sober Goodfellow might have agreed. However, driven by his beer-inspired conviction, he went home and immediately started coding. That night, he proved his friends wrong, creating the first Generative Adversarial Network (GAN). GAN became notoriously famous as underlying technology behind DeepFakes. But it had many positive applications too- for image enhancement, generating realistic art, restoring old photos, medical imaging, augmenting datasets for AI training etc. In 2016, Yann LeCun called GANs “the coolest idea in deep learning in the last twenty years. Goodfellow's story emphasizes a valuable lesson: deliberate time away from work, allowing the mind to relax and explore unconventional ideas, can lead to breakthroughs that ultimately benefit our work. In my book- "What’s Your Human Edge? : 53 Timeless Ideas for Thriving in the Age", I make a mention of the following (inspired from David Perell)- Creatives oscillate between two primary work modes – Beer mode and Coffee mode. (Included a sketchnote from my book) Beer mode embodies a relaxed, unfocused state of play—a sandbox where novel ideas germinate and flourish.  (that's Goodfellow at pub where the inspiration striked him) Coffee mode represents a concentrated phase, characterized by deliberate efforts to achieve specific results (that's Goodfellow at night going deep and trying the idea he got at pub) This balance between allowing the mind to wander and harnessing its focus is crucial for fostering innovation and realizing visionary ideas. What do you think? #DeepLearning #WhatsYourHumanEdge

26

1 Comment

If you want to understand why LLMs are not the solution for AGI (Artificial General Intelligence), I highly recommend listening to Lex Fridman podcast with Yann LeCun : Meta AI, Open Source, Limits of LLMs, AGI & the Future of AI T͟L͟D͟R͟: 1. LLMs do not understand the physical world 2. LLMs have not the ability to remember 3. LLMs have no persistent memory 4. LLMs are not able to reason and to plan

27

5 Comments

Discussing size of models and benchmarks is one aspect. But knowing whether a particular model is good in a particular context and another model in a different use case is key for adoption. „The focus on SLMs represents a maturation of the AI field, shifting from a preoccupation with raw capabilities to a more nuanced understanding of real-world applicability.“ „Moreover, while smaller models offer advantages in terms of efficiency and accessibility, they may not match the raw capabilities of their larger counterparts in all tasks. This suggests a future AI landscape characterized by a diversity of model sizes and specializations, rather than a one-size-fits-all approach.“ #llms #slms #adoption https://lnkd.in/e4gVaD2p

4