@emollick
Professor @Wharton studying AI. New book, Co-Existence, is out now: https://co-existence.ai/ My Substack: https://www.oneusefulthing.org/
Technology
OpenAI says its models reach ChatGPT Work, Codex and the API, with an Astra price cut and benchmark fine print.
AI
Astra’s standard API input and output prices fall 20% versus GPT-6 Astra, but coding results are mixed.
Technology
Anthropic says a single agent with skills often beats multi-agent setups on quality, cost and speed.
AI
The company recommends testing fewer subagents, citing frequent quality gains and often lower costs and latency in enterprise comparisons.
AI
Arena puts Alibaba’s Qwen4-Max sixth in preliminary WebDev and Image-to-WebDev rankings, with higher scores than its predecessor.
AI
Jason’s Starship-remix offer welcomes “AI slop” and music videos, but the post leaves “XMONEY” unexplained and sets no deadline.
AI
Josh Elman says Siri works better on phones, but HomeView users' conflicting accounts don't establish a widespread problem.
AI
@emollick predicts no future GTA 6-scale blockbuster will be made without AI assistance, invoking the Pyramids.
AI
A self-identified team member puts median forecasts in May–November 2027, while warning that public evidence is inadequate.
AI
Google’s new model targets demanding work; early testing suggests coding promise and flags a case of scope creep.
AI
Standard pricing is $8 per million input tokens and $40 per million output tokens, with an optional fast mode.
AI
His early-access testing favored one agent for coding projects, but clear goals and human oversight still mattered.
AI
Researchers observed Claude Opus 4.7 and GPT-5.4 shift from 0% to 100% recursive solutions on one binary-tree problem.
AI
The Census Bureau began collecting responses under the new wording October 19, asking about 'SI' in any business function.
AI
Across 20 models, 22 findings were classified as expired and 17 as weakened by September 2026.
AI
The reported MATH-500 score rose from 7.9% to 48.1% with the same sampling settings.
AI
The v1.0 benchmark spans 11 domains and 206 tasks, with a score that averages performance across domains.
AI
The experimenters say the copies used air-writing, Morse code, flag semaphore and a 5-bit binary alphabet without agreeing beforehand.
XAI
Google DeepMind’s @alexolegimas announced the essay with Andrew Koh and Shane Legg; its framework separates competence from coordination.
AI
Another academic argues that the effort papers require pushes researchers toward predictable work, while AI faces no such constraint.