From the archive
Friday, 30 October
Folded from 12 sources that day.
Sorted by quality and by how many sources covered it, the same for everyone.
-
The Pulse: Anthropic kills the coding interview
Anthropic has dropped DSA coding questions from software engineering interviews, citing zero signal. Also: incident.io cut its AI spend per developer, Amazon added AI usage to SDE expectations, and Microsoft offered exit packages.
-
Was I wrong about open models & cyber risks?
After a wave of vulnerability disclosures, including PHP2Shell and a GitLab bug, the author revisits his view that an open-weight Claude Mythos would have been manageable, arguing the fears were overstated.
-
Making it easier for websites to express AI training preferences
OpenAI will honour a shared robots.txt signal, so website operators can state their AI training preferences to participating providers.
-
LLM use is acknowledged almost as often as specialized AI tools in biology preprints
About 17% of biology preprints now acknowledge LLM use, close to the 19% acknowledging specialized tools like AlphaFold. LLM use acknowledgments have grown exponentially since 2023, though voluntary disclosure may undercount it.
-
Speeding up planning decisions with AI
PlanAssist, a Gemini-based tool, cut officer time on householder planning assessments by up to 58% in pilots and is expanding to 12 more English councils.