🚨 41 years in the making — and Dijkstra is no longer unbeatable. A Tsinghua, Stanford, and MPI for Informatics team has achieved the first deterministic algorithm to break the O(m + n log n) bound for single-source shortest paths in directed graphs with real non-negative

Are frontier AI models really capable of “PhD-level” reasoning? To answer this question, we introduce FormulaOne, a new reasoning benchmark of expert-level Dynamic Programming problems. We have curated a benchmark consisting of three tiers, in increasing complexity, which we call

There is psychological well-being benefits to pair programming with frontier LLMs, that would not be offered by perfect (meaning, not making mistakes) but partial (meaning, not AI-complete) GOFAI systems.

Watch Starship's tenth flight test → spacex.com/launches/stars… x.com/i/broadcasts/1…

GPT-2, Opus 4.1 na-palcach.blogspot.com/search/label/G…

super cool to compare the outputs from GPT-1 through GPT-5, given the same prompt: progress.openai.com

ppx_minidebug might reach v3 before it reaches its 3rd user. #ocaml

Ok hear me out - the sun keeps sending it's unwanted radiation to America. This is unacceptable, we need to put a tariff on those solar rays. Let's build a wall that will absorb that unwanted light and turn it into electricity. But we need to make the wall horizontal and put one

Watch Starship's tenth flight test → spacex.com/launches/stars… x.com/i/broadcasts/1…

Since everyone is talking about RL Environments and GRPO now but no one knows how it works we thought it would be cool to make an explainer video + code you can run: This is an example of using GRPO to train Qwen 2.5 to play 2048 (code in thread) 🧵:

I just finished reading this interesting book by Druv Pai, Sam Buchanan and colleagues. It's fairly "heavy" but provides a very satisfying theoretical explanation for many different empirical approaches currently used in "generative AI", such as denoising diffusion models,

I'm currently having a Claude Code phase. I stopped using Cursor agents, but Cursor tab complete is by itself totally worth the price.

Starship

tab-o: sorry, I blindly pressed autocomplete on this

Updated & turned my Big LLM Architecture Comparison article into a narrated video lecture. The 11 LLM architectures covered in this video: 1. DeepSeek V3/R1 2. OLMo 2 3. Gemma 3 4. Mistral Small 3.1 5. Llama 4 6. Qwen3 7. SmolLM3 8. Kimi 2 9. GPT-OSS 10. Grok 2.5 11. GLM-4.5

My old university page is dead :-| Truth be told it was an attack vector, insecure wiki content management setup. But a bit sad to lose it.

Here's Nathan Taylor who'll be speaking at FUN OCaml 2025 - and it's already next Monday! ✨🐫

Huawei proposed Tree-OPO! They explore how MCTS trajectories can fuel Group Relative Policy Optimization (GRPO), enabling preference-based RL without value networks. By staging training with partially revealed rollouts, they create tree-structured reward signals that better

Which life do you prefer: at the frontier, or out of distribution?