Norman Mu
@normanmu.com
AI Safety @ xAI | AI robustness, PhD @ UC Berkeley | normanmu.com
System prompts are a critical control plane for LLMs/AI agents, but models vary widely in their robustness. We found a "complexity wall" at ~10 guardrails where prompt adherence declines rapidly even on totally benign inputs
February 19, 2025 at 6:06 AM
System prompts are a critical control plane for LLMs/AI agents, but models vary widely in their robustness. We found a "complexity wall" at ~10 guardrails where prompt adherence declines rapidly even on totally benign inputs
absolutely wild allegations being shared about the author of the NeurIPS 2024 best paper winner: var-integrity-report.github.io.
really feels like the bottom is dropping out from academic ML research with the amount of brazen dishonesty
really feels like the bottom is dropping out from academic ML research with the amount of brazen dishonesty
Ethical Challenges Related to the NeurIPS 2024 Best Paper Award
var-integrity-report.github.io
December 11, 2024 at 8:26 AM
absolutely wild allegations being shared about the author of the NeurIPS 2024 best paper winner: var-integrity-report.github.io.
really feels like the bottom is dropping out from academic ML research with the amount of brazen dishonesty
really feels like the bottom is dropping out from academic ML research with the amount of brazen dishonesty
"AWS is currently deploying a cluster with 200k+ Trainium2 chips for Anthropic called “Project Rainer”" semianalysis.com/2024/12/03/a...
Amazon’s AI Self Sufficiency | Trainium2 Architecture & Networking
Amazon is currently conducting one of the largest build-out of AI clusters globally, deploying a considerable number of Hopper and Blackwell GPUs. In addition to a massive Capex invested into Nvidi…
semianalysis.com
December 3, 2024 at 8:56 PM
"AWS is currently deploying a cluster with 200k+ Trainium2 chips for Anthropic called “Project Rainer”" semianalysis.com/2024/12/03/a...
i've been confusedly semi-following the RL chip design/AlphaChip story for a few years trying to understand what the core disagreement was, and finally found a summary that (seems) to collate and explain all the relevant artifacts
Reinforcement Learning for Chip Floorplanning: The Saga
vighneshiyer.com
December 1, 2024 at 8:45 AM
i've been confusedly semi-following the RL chip design/AlphaChip story for a few years trying to understand what the core disagreement was, and finally found a summary that (seems) to collate and explain all the relevant artifacts