Repo: github.com/codelion/optil…

This podcast is created with Lettercast lettercast.ai/home --- 📚 arxiv.org/pdf/2309.06275 👇 Also all my Paper Podcasts are also available on my YouTube channel playlist 👇 youtube.com/playlist?list=…

Learned a lot from this thread! Let me try to summarize the insights below: [Clarifications] No doubt we can iteratively generate data with CoT/search and distill. To understand it one step further, I’m curious about: 1. What are the potential benefits of test-time compute, as

🪩The State of AI 2024 has landed! 🪩 Our seventh installment is our biggest and most comprehensive yet, covering everything you *need* to know about research, industry, safety and politics. As ever, here's my director’s cut (+ video tutorial!) 🧵

LLMs are trained on much more than the whole Internet -- they also consume handcrafted answers produced by armies of highly qualified data annotators (often domain experts). Today approximately 20,000 people are employed full-time to produce training data for LLMs.

Extremely proud to have been a part of this. Veo 2 is a significant upgrade and has achieved SoTA results in head-to-head comparisons of outputs by human raters over top video generation models. Looking forward to seeing what everyone creates with it. deepmind.google/technologies/v…

Beautifully said Nando. We should slow down the crazy rat race and enjoy the beautiful work we do without all the unnecessary stress. It is not worth it.

I’m grateful for 2024. I had a brilliant start at Google DeepMind working on Lyria, Veo and Imagen 3. I’m pleased that my teams continued to build on our work and deliver amazing results after I left. By leaving Google, I had the chance to go back to the things I love doing the

I’m really sad that my dear friend Felix Hill is no longer with us. He had many friends and colleagues all over the world - to try to ensure we reach them, his family have asked to share this webpage for the celebration of his life: pp.events/felix

This is incredibly sad.

I worked with Felix on a paper in 2021. I remember Felix as a consistently kind person, and as a brilliant thinker. I'm sharing his farewell letter, as I think it's what he would have wanted. Please be mindful, it's not an easy read: docs.google.com/document/d/1-j… Rest in peace.

People scaled LLMs by ~10,000x from 2019 to 2024, and their scores on ARC stayed near 0 (e.g. GPT-4o at ~5%). Meanwhile a very crude program search approach could score >20% with hardly any compute. Then OpenAI started adding test-time CoT search. ARC scores immediately shot up.

👨‍🚀 2025 is off to a great start - today we're excited to share that Sina Samangooei has joined Cusp as Member of Technical Staff! With extensive experience in multimodal AI systems, large language models, and autonomous systems from his roles at DeepMind, FiveAI (Bosch), and Amazon, Sina

I and a few other CuspAI folks will be going to #ICLR2025 next week! are you? let me know! We have some sweet sweet swag to give away, come find me and tell me two truths and a lie about material science to get your very own cusp-swag!

About to watch a pirate themed water based circus at the hippodrome... we're all pumped...#ICLR2025 has a lot to live up to next weekend