AI — Page 14

Dean Ball, hired in July to shape frontier policy, says he will keep issuing candid warnings even when they complicate OpenAI's message.

The first cohort will include up to 100 New York-area superintendents, education leaders and educators, with SkillsBuild learning, credentials and responsible-AI capstone projects this fall.

LatchBio found 59.2% red-team refusal and 64.8% routine completion, showing how far behavior can shift under one model name.

Alibaba says the September 2nd update improves coding and Cowork performance; its Singapore service lists $2 per million input tokens and $6 per million output tokens.

The reported coding model could arrive 20 days after Gemini 3.7 Flash, though Google has yet to publish benchmarks, pricing or a model card.

Baseten's technical essay turns latency, throughput and model quality into a framework for deciding where expensive inference gains come from.

Basis, Clay and Exa Labs show three ways founders are turning bounded internal processes into agent workflows, from onboarding and account triage to tested developer integrations.

MiniMax's open-weight H3 now runs through vLLM-Omni, while its community license excludes the US, EU, UK and Republic of Korea.

Fei-Fei Li's company is putting camera control, scene reconstruction and synthetic robot views into one model, with access initially limited to selected partners.

The CEO linked OpenAI's path to a public offering to the pace of recursive self-improvement as the company slows frontier research and tightens model safeguards.

Every current subscriber gets a Reset Card for weekly and five-hour limits, with redemption tied to a signed-in ZCode account.

The Cambridge-linked founders are extending Flower's federated-learning stack into a licensed model preview for sensitive enterprise workloads.

The Accio team's state-based benchmark reports Qwen3.8-Max completing 56 tasks through Accio, 47 through OpenClaw and 53 through Pi; Alibaba calls it the strongest overall open-weight result.

Founders Red Xiao, Tao Zhang and Yichao "Peak" Ji remain in charge after a separation that forced account and task-data deletions.

DHH says agents handled Quattro's implementation, while 37signals' Basecamp work showed how quickly mature architecture can become a cleanup job.

The two-person YC lab says delta-neutral strategies gained 140%, but it has not published the period, capital base, leverage or methodology.

A forensic review of Chang Liu's Apple-issued MacBook has become the iPhone maker's main argument for accelerating discovery.

Version 0.21.0 adds persistent bot-to-bot messaging, steerable subagents, browser control and scheduled jobs that remember prior runs.

Meta's terminal coding agent adds cross-session messaging, multi-agent workflows and an SDK preview after a 26-day beta.

The two-person startup says its early customers include a large US public pension plan, a top law firm and governance advisors.

The early-access research model turns clicks and drags into synthesized 720p frames, while text, trust and accessibility remain open problems.

The 305B-parameter checkpoint lands 10 days after its API debut, while DeepSeek's headline benchmark uses Anthropic's superseded Opus 4.8.

HUMAIN will supply power and 120,000 AI chips under a revenue-sharing arrangement for open-model training and inference.

A viral post adds an unverified training-complete claim to Musk's August promise of a September model trained on SpaceX engineering data.