As someone who has audited dozens of safety-critical systems, built static analysis tools, and used most formal verification and security tools, here are some red flags that should be a caution in taking these claims at face value:
1. There are no comparison benchmarks with 1/

https://twitter.com/AnthropicAI/status/2041578392852517128

traditional static analysis (SA) tools well-used within these orgs. This includes Amazon's own tooling, and advanced tools well used in safety-critical settings like CodeSonar.
2. The rate of false positive is not disclosed, when assessing security and and SA tools, this is 2/

Codex paper is out! I'm grateful to have led the Safety and PL workstreams for Codex/Copilot, working along Policy @OpenAI. There's many questions about limitations and implications (BI section), a thread on some of our findings:
arxiv.org/abs/2107.03374 First, to assess Codex, I developed a framework aiming to assess the capabilities of all code synthesis models to generate code for increasingly complex and intentionally higher-level specifications (PL folks: a bit more, but not all, info in Appendix D and last tweet here)

I've been thinking about the 30 under 30 threads that occurred last month, and the inspirational stories I read regarding those who found success after 30. But something that wasn't emphasized enough was how achieving success at a younger age often required massive privilege. Those who faced hardships and lacked privilege, no matter the age, should be recognized for the unspoken successes they've achieved. Whether it be fighting through mental illness, or being an immigrant. I really want to hear your story and what you're proud of. #unseensuccess

In my humble few months verifying safety critical code in industry, I now wish I knew quite a few things during my PhD: A criticism I often received was that our tool (T2), needed to go beyond verifying C. C was simple, archaic, unsafe, and no longer used.

Growing up in Egypt, the idea that women can't be engineers or scientists did not exist. This may be due to the fact that "nerd" or "geek" Is not in the Arabic vocabulary. It was common to see a 50/50 ration in CS courses, and the top performers in universities were often female