Johnny Devriese

@kepano was excited about this but seems like it's not working

Defuddle now returns Youtube transcripts! Paste a YouTube link into defuddle.md to get a markdown transcript with timestamps, chapters, and pretty good diarization! ...or if you just want to read it, try the new Reader mode in Obsidian Web Clipper powered by Defuddle.

📣 Excited to share my first work @Princeton : 𝗧𝗼𝘄𝗮𝗿𝗱𝘀 𝗮 𝗦𝗰𝗶𝗲𝗻𝗰𝗲 𝗼𝗳 𝗔𝗜 𝗔𝗴𝗲𝗻𝘁 𝗥𝗲𝗹𝗶𝗮𝗯𝗶𝗹𝗶𝘁𝘆 AI agents keep getting more capable. But are they actually reliable? 📄 Paper: arxiv.org/abs/2602.16666 📊 Dashboard: hal.cs.princeton.edu/reliability 🧵👇

Slides for my lecture “LLM Reasoning” at Stanford CS 25: dennyzhou.github.io/LLM-Reasoning-… Key points: 1. Reasoning in LLMs simply means generating a sequence of intermediate tokens before producing the final answer. Whether this resembles human reasoning is irrelevant. The crucial insight is that transformer models can become nearly arbitrarily powerful by generating many intermediate tokens, without the need of scaling the model size (arxiv.org/abs/2402.12875). 2. Pretrained models, even without any fine-tuning, are capable of reasoning. The challenge is that reasoning-based outputs often don’t appear at the top of the output distribution, so standard greedy decoding fails to surface them (arxiv.org/abs/2402.10200) 3. Prompting techniques (e.g., chain-of-thought prompting or "let’s think step by step") and supervised finetuning were commonly used to elicit reasoning. Now, RL finetuning has emerged as the most powerful method. This trick was independently discovered by several labs. At Google, credit goes to Jonathan Lai on my team. Based on our theory ( see point 1), scaling RL should focus on generating long responses rather than something else. 4. LLM reasoning can be hugely improved by generating multiple responses and then aggregating them, rather than relying on a single response (arxiv.org/abs/2203.11171).

rebranding Linear Algebra as Artificial Intelligence was the most successful marketing campaign of all time.

Well Claude Sonnet can't do Jackson E&M problems. So I think physics grad students continue to get to suffer.

Will Barta WINS the last stage of @VueltaCV , his first professional victory after a heroic 50 km solo attack from the breakaway! One of my favourite Movistar victories EVER. Huge congrats, @willbarta ! 👏 #75VCV

Thought it was april 1st. LFG!!

By far the most important issue on GH github.com/warpdotdev/War…

Really need an extension that redirects all medium articles to webcache.googleusercontent.com/search?q=cache:<URL>

The DoD gets it! archive.ph/jsb07

excited to try out dino dinov2.metademolab.com

#Ideogram is amazing

Kinda funny that FB has kept facebookresearch on GH because someone else already has their username github.com/metaresearch

This is a cool way to visualize decision tree splits. gitlab.tue.nl/20040367/pybao…

Multimodal learning is really cool. CMU also has an excellent course on the topic cmu-multicomp-lab.github.io/mmml-course/fa…

Oppenheimer didn’t need Jira or a certified scrum master

I don't think Shafer Butte is 7950 ft @BOGUSBASIN media.bogusbasin.org/uploads/2022/0… 🧐

Are RNNs even still a thing? 🫠

Here’s how the first #WEC grid of the season shapes up for tomorrow’s race 🤩 #️⃣5️⃣0️⃣ - P1 #️⃣5️⃣1️⃣ - P4 #FerrariHypercar #Ferrari499P 🇺🇸 #1000MSebring