Publication
Can Models Prove Their Own Work?
Models are making it cheaper to generate more attempts: faster kernels, possible protein binders, new data-center demand, and software output from coding agents. The harder question is whether models can prove their own work, or where tests, labs, buyers, infrastructure, and human judgment still have to decide what is real.
Sources and further reading
- Finding high-severity security issues with publicly available models
- How Makora Generates CUDA Kernels That Beat Hand-Tuned Code | Researcher Conversations at GTC
- Finding Miscompiles for Fun, Not Profit
- @SemiAnalysis_: Anthropic Growth and Bedrock Mix Drive AWS Margins Higher While Peers Lag
- Laguna M.1 and Laguna XS.2 Technical Report
- The Bitter Lesson is Coming for Proteins - Alex Rives, BioHub
- @xai: Grok-build-0.1 in Kilo Code
- Thank God For Data Centers
- Building self-improving tax agents with Codex
- Waste Tokens, Save Time
- @gdb: ChatGPT bug-fixing thread with Codex
- Coding agents in the social sciences
- @SemiAnalysis_: Sub-agent usage stat image
- AI Factories: The New Infrastructure of Intelligence
- Anthropic Growth and Bedrock Mix Drive AWS Margins Higher While Peers Lag
- From data overload to actionable insights: How Verizon Connect scaled agentic AI to 100,000 users
- Building AI agents for business support using Amazon Bedrock AgentCore
- Warp's big bet on building open source with GPT-5.5