Patrick John

How to get AI to make discoveries on open scientific problems? Most methods just improve the prompt with more attempts. But the AI itself doesn't improve. With test-time training, AI can continue to learn on the problem it’s trying to solve: test-time-training.github.io/discover.pdf

@HarmonicMath , @axiommathai, @logic_int (et al) want to go from vibe coding to vibe proving. How is that even possible? LLMs hallucinate all the time, so why should we trust them with math?

That’s a lot of @WCHTokyo25 medallists 👀 #WorldAthleticsChamps

Jake gets his silver 🥈 Announced by his dad Geoff Wightman and delivered by Seb Coe 👏 #WCHTokyo25 #NovunaGBNI #WorldAthleticsChamps

Today I'm excited to share that @StudioAtelico has raised $5M in seed funding. We'll use these funds to build games that couldn't exist before with AI tools that empower developers, all while following a clear ethical line on AI. The other piece of news is that the amazing Chris Olson is joining the team! He already started supercharging our game production. Thank you to @nathanbenaich of @airstreet, with support from @MinifigVC and @talfanevans from Clay, @HazyResearch and @samjackson from Factory, @TommasoMoraca11 and Isabella Timossi at @vento_ventures , angels including @Thom_Wolf , @doppenhe, @ajratner, @Tkaraletsos, @msciglio, @jacopotagliabue, @pancrazioauteri and @alexfo and @MarcusZe at @prediidata for partnering with us in this journey. Read more about it the blog post, atelico.studio/blog/funding-a…. By the way, we are hiring for an Art Director role! Feel free to reach out.

AI in games can flood the world with forgettable slop, or it can make the impossible playable. We are not here to cut corners or replace artists. We are using AI to unlock new play styles, living worlds, and evolving art. But only if: ✔️ Artists explicitly approve how their work is used for training Ai models ✔️ Revenue is shared from games where AI draws on their style ✔️ Labor and creativity are respected throughout Read @StudioAtelico 's latest blog post to see how to use AI in video games without losing your soul: atelico.studio/blog/game-ai-a… #GameDev #AIethics #GenerativeAI

Ciao Roma! Tomorrow I'll be in Rome for the second leg of the Studio Atelico Europe Summer Tour 2025 (Malmö you are next) to give a talk at /dev/games about our Generative Agents Realtime Playground and the game tech we are developing at @StudioAtelico for running AI on-device. 🗓 Thursday, June 5th 🕒 11:10–12:10 Looking forward feedback, in particular from the amazing @MarkSkyzoid and Andrea Tucci (thanks for inviting me). Come say hi if you are around the conference, and want to chat about ethical AI in games!

🚀 Introducing Bauplan A serverless, code-native platform for building data and AI pipelines — directly on your object store. No clusters. No notebooks. No GUI based workflows. Just Python + SQL + S3. 👉 bauplanlabs.com/blog/hello-bau…

“You could attach prices to thoughts. And how does one pay for thoughts? The answer, I think, is: with courage.” Can we re-imagine the data lifecycle so that it works for humans and machines alike? Yes, we can. Thanks @TreybigDavis (and team) and @adityaag (and team)!

🤗 @HuggingFace’s most impactful contributors you should follow [August 2024] : huggingface.co/spaces/mvaloat… 🏛️ Top 10 Model downloads: 👏 @jonatasgrosman, @patrickjohncyh, #avichr, @eugene_siow, @limsanity23, #MahmoudAshraf, @Emily_Alsentzer, @perezjotaeme, #intfloat, @TheBlokeAI 📊 Top 10 Dataset downloads: 👏 @haileysch__, @lukaemon, @moonares, @princeton_nlp, @nathanhabib1011, @algo_diver, #ChristophSchuhmann, #haonanli, #hendrycks, @jxmnop 🚀 Top 10 Space likes: 👏 @fffiloni, @multimodalart, @flngr, @_akhaliq, @angrypenguinpng, @yvrjsharma, #KingNish, @hysts12321, @xenovacom, @radamar 🤗 Top 10 Profile followers: 👏 @TheBlokeAI, @mervenoyann, @_akhaliq, @fffiloni, #lllyasviel, @teknium, @xenovacom, #Undi95, @erhartford, @maximelabonne

🤗 Here are the top 100 of @HuggingFace’s most impactful users of July 2024 (models, datasets, spaces, followers): 🏆 Top Contributors To Follow: huggingface.co/spaces/mvaloat… - 🏛️ Top 10 Model Downloads: 👏 @jonatasgrosman, #PatrickJohnChia, @Lin_Chen_98, @Emily_Alsentzer, #LiangWang, @limsanity23, @eugene_siow, @mrm8488, #avichr, @TheBlokeAI 📊 Top 10 Dataset Downloads: 👏 @haileysch__, @nathanhabib1011, @lukaemon, #HaonanLi, @algo_diver, #YebHavinga, @princeton_nlp, @jonbtow, @moonares, #DanHendrycks 🚀 Top 10 Space Likes: 👏 @fffiloni, @flngr, @multimodalart, @_akhaliq, @yvrjsharma, @angrypenguinpng, @hysts12321, #NishithJain, @xenovacom, @radamar 🤗 Top 10 Profile Followers: 👏 @TheBlokeAI, @mervenoyann, @fffiloni, @_akhaliq, #LvminZhang, @teknium, @xenovacom, #Undi, @erhartford, #WizardLM

This is the most interesting paper I've read in a while. A calculus of language instead of number. infinitesimal = a criticism integration = editing over criticisms you optimize LLM programs via backprop where each update is expressed linguistically

⚡️This is the most fun project! We built PyTorch-for-text! 🔥 #TextGrad: automated "differentiation" via text to optimize AI systems by backpropagating LLM text feedback. TextGrad + GPT4o: 💻LeetCodeHard best score ❓GPQA sota 🧬Designs new molecules 🩺Improves treatments 🧵

Tired of the same static benchmarks for #LLMs? Have them negotiate against each other to see how they do! 📢Our #chatbot #NegotiationArena is accepted by #ICML2024. Updated paper: openreview.net/pdf?id=CmOmaxk… Code: github.com/vinid/Negotiat… Great job @federicobianchy Patrick Chia @mertyuksekgonul @jacopotagliabue @jurafsky! 👏

@woo_gerald @PatrickKidger 😇

I'm particularly proud of this project because I've started to write tests comprehensively for the first time 😇. And overall, just implementing the tips from @PatrickKidger's guide (kidger.site/thoughts/just-…) - loving the jaxtyping library. Only thing left is documentation 😅

🚀 OPEN SOURCE ALERT 🚀 I'm so excited to share that we have open sourced uni2ts and Moirai!!! Training pipeline and all the good stuff are also available! So excited to see what's in store for the future of foundation models for time series.

⚡️Excited to introduce #LLM Negotiation Arena! We pair #GPT4, #Claude2 etc to negotiate against each other to buy/sell, share resources + other scenarios. Paper: arxiv.org/abs/2402.05863 Code: github.com/vinid/Negotiat… So which #AI🤖do you want to negotiate for you? 👇1/3

I'm hosting a student researcher at Google DeepMind in 2024! If you or some you know is interested in robotic manipulation, multi-modal learning, and want to work at London GDM then apply by Dec 15 (note it is tight!). Link: shorturl.at/fivV6 and lmk if you applied!

Probably the best reasonable scale #MLOps template. You don’t need a bigger boat: bit.ly/3gFXIUl End-to-end working MLOps system you can build on.

It's #opensource @github repo where @jacopotagliabue, @patrickjohncyh, Luca Bigon, Andrew Sutcliffe, Leopoldo Garcia Vargas put together an e2e (@MetaflowOSS-based) implementation of intent prediction & session reco. Great starting point when building (1st or not) MLOps system.

Come out and play: still 9 days to submit to our @kdd_news workshop on evaluating #recsys for real-world impact. We have a tiny window before #AGI makes all #recsys useless: make it count!