Plus: Amazon and Google take large Anthropic stakes at IPO; Neural networks realize Wess-Zumino-Witten models
Welcome to a Saturday edition where the machines are quietly doing nearly half the work.
RESEARCH
馃敩 OpenAI agents complete nearly half of research tasks
Image source: openai.com
OpenAI's coding agents completed 23% of two-to-four-hour AI research tasks without human help in March, rising to 46% in October. As of September 2026, the API-equivalent value of agentic AI use in OpenAI's research organization was about twice its non-equity compensation.
The details:
The median researcher at OpenAI used around $2,400 USD per day on agentic AI tools at API list prices.
Agent workdays per researcher at OpenAI rose from 0.5 to 5.7, an elevenfold increase.
The median respondent to an OpenAI survey would give up 40% of team headcount to restore frontier coding agents after a three-month loss.
The paper develops neural realizations of positive-level Wess-Zumino-Witten models using bosonic Wakimoto and fermionic Fourier-Berezin fields. Level-rank duality connects the charged sectors of the fermionic construction.
The details:
Explicit neural observables recover conformal block data in the fermionic construction.
The neural construction enables the transfer of information about unknown correlators between paired theories.
The paper constructs the Virasoro algebra from a neural free-field architecture.