Google launches Gemini 4 Carbon: #1 on the overall intelligence index, built for cybersecurity
10 output; strong scores on finding and fixing vulnerabilities, rolling out to security defenders first.
Tip
models now have “specialties” — pick by use case
Anthropic study: professional analysts followed the AI’s lean 86% of the time, but credited it only 15%
50 analysts did a causal-inference task while the AI quietly nudged four choices in random directions; 77% of decisions reused the AI’s exact arguments.
Tip
you think you’re deciding, but the AI is. How to keep your own judgment when working with AI
Full end-to-end tax work: even the best AI passes only 10.8%
Surge AI’s new benchmark covers 8 tax areas; Claude Sonnet 5.5 leads. Models miss deadlines and don’t verify assumptions.
Tip
knowing tax law isn’t the same as filing your taxes — professional judgment stays with you
Anthropic acquires a one-person biotech
the founder used Claude to coordinate 11 outside labs and moved a muscular dystrophy candidate to promising mouse results in under 18 months; Anthropic also announced drug-discovery deals with Bristol Myers Squibb and five others.
Tip
one person plus AI as a drug company — what a “super individual” really looks like
Mac Studio now sells with 512GB of memory, from $22,299
M5 Ultra; the first Mac to reach 512GB, aimed at developers running large models locally.
Tip
follow-up to the Oct 16 “816B on two Mac Studios” post — local AI is still expensive
Alphabet earnings: Gemini app hits 1 billion users; Google’s model APIs process nearly 32 billion tokens per minute
Q3 revenue $127.6B, Google Cloud up 103%; token volume is 4x a year ago.
Tip
AI usage up 4x in a year — “too cheap to think about” is already happening
⚠️ Compiled by an AI assistant from public AI news roundups; original articles are not individually verified. For reference only.