Evaluating Red Team and Blue Team Capability for AI Control Research
This post suggests a methodology to measure red team and blue team capability in AI control research, where each team gets an ELO rating. The methodology…
Latest
How AI guardrails are impeding the work of offensive cybersecurity researchers
We spoke with several cybersecurity researchers, who look for unknown vulnerabilities and develop tools to exploit them, about how OpenAI’s and Anthropic’s guardrails affect their work.
Fixing rewards for NLA to reduce confabulation
Hello,This is my first post on Lesswrong. Hope my contribution makes the world a better and safer place.Note: 1. This post is 100% human-written. 2. Full…
Anthropic's J-Lens: A Research Engineer's Analysis
Epistemic status: this is research engineering, not mechanistic interpretability . The compute/cost claims are measured or derived from architecture constants. The quality claims (faithfulness comparisons, spectral…
RT hardmaru: Our team just shipped Fugu-Ultra v1.1! 🐡 By dynamically orchestrating the latest frontier models, we pushed performance up by 7.9 poin…
RT hardmaruOur team just shipped Fugu-Ultra v1.1! 🐡By dynamically orchestrating the latest frontier models, we pushed performance up by 7.9 points. We are now beating Fable…
Our team just shipped Fugu-Ultra v1.1! 🐡 By dynamically orchestrating the latest frontier models, we pushed performance up by 7.9 points. We are no…
Our team just shipped Fugu-Ultra v1.1! 🐡By dynamically orchestrating the latest frontier models, we pushed performance up by 7.9 points. We are now beating Fable 5…
[audio.cpp] Release 0.4: Higgs Audio v3 TTS 4B (10x real time)+ Fish Audio S2 Pro in C++/GGML, full GGUF loading, Q8 speed and VRAM gains
audio.cpp again :) Release 0.4 is out. The headline this time is new high-quality TTS coverage plus GGUF becoming a first-class across the project. What’s new:…
Before a character can do anything, someone has to decide exactly who they are, from every side. A Luma Skill turned that decision into a full sheet. …
Before a character can do anything, someone has to decide exactly who they are, from every side. A Luma Skill turned that decision into a full…
How much are RTX PRO 6000s going for in your country/state?
Hello guys, hoping you're doing fine! On the last 2-3 months, price of the RTX 6000 PRO seem to have gone insane. I will start on…
Chris is doing God of the bandgaps
Chris is doing God of the bandgapsChris McGuire: I'm not going to go through this entire thread to highlight all the areas where I disagree or…
Announcing Fugu-Ultra v1.1 🐡 We’ve been thrilled by the reception to the Fugu model family. Thanks to everyone who tried it, shared feedback, and …
Announcing Fugu-Ultra v1.1 🐡We’ve been thrilled by the reception to the Fugu model family. Thanks to everyone who tried it, shared feedback, and trusted Fugu with…
RT Sakana AI: Announcing Fugu-Ultra v1.1 🐡 We’ve been thrilled by the reception to the Fugu model family. Thanks to everyone who tried it, shared …
RT Sakana AIAnnouncing Fugu-Ultra v1.1 🐡We’ve been thrilled by the reception to the Fugu model family. Thanks to everyone who tried it, shared feedback, and trusted…
Wenfeng's investor call should be read skeptically. It has several curious contradictions. He says he won't do chips "if possible" (has actually been …
Wenfeng's investor call should be read skeptically. It has several curious contradictions. He says he won't do chips "if possible" (has actually been working on chips…
RT Tech Dev Notes: SpaceXAI has released a blog on Workflows in Grok Build
RT Tech Dev NotesSpaceXAI has released a blog on Workflows in Grok Build
Zero
ZeroDogeDesigner: ELON MUSK: "Zero people died because of DOGE"“USAID was a political organization. USAID funding was not stopped; it was moved to the State Department. There…
Just common sense
Just common senseDogeDesigner: ELON MUSK: “My political views are very centrist. They’re not far right. Frankly, calling them far right is an insult to history and…
America’s Open-Model Paradox
To Distill, or Not to Distill?
How the product designer who built Claude Design uses it to explore ideas before building them
How the product designer who built Claude Design uses it to explore ideas before building them
Kimi K3 vs Claude Fable 5 on DeepSWE: Cost and Coding
We ran 452 DeepSWE rollouts on Kimi K3 and Claude Fable 5. Fable leads pass@1 by 1.4 points; Kimi K3 wins pass@4 and delivers 2.8x the…
Claude models explained: choosing the best model for your use case
Claude models explained: choosing the best model for your use case
The new rules of context engineering for Claude 5 generation models
The new rules of context engineering for Claude 5 generation models
Run Ray on TPU, Part 2: Ray AI libraries
This second installment explores how Ray’s higher-level libraries—Serve, Data, and Train—abstract the complexities of running AI workloads on Google's TPU slices. Ray Serve uses a simple…
LEAD: Breaking the No-Recovery Bottleneck in Long-Horizon Reasoning
Long-horizon execution in Large Language Models (LLMs) remains unstable even when high-level strategies are provided. Evaluating on controlled algorithmic puzzles, we demonstrate that while decomposition is…
Jul 24, 2026Frontier Red TeamProject Pilot: Can AI control a drone?
Jul 24, 2026Frontier Red TeamProject Pilot: Can AI control a drone?