AI Daily
TOOLIN DAILYGPT-6 Safety Model and Regional Data Center Restrictions
OpenAI releases GPT-6 Cyber and image updates, Anthropic launches personal finance, and local votes plus US stock sell-offs trigger compute expansion concerns.
The essentials
OpenAI Releases GPT-6 Cyber Safety Model
OpenAI released the safety research model GPT-6 Cyber. Officials stated it improves alignment while enhancing capabilities and drives vulnerability research, with teams like Mozilla and Cisco already using it. The team used the model to generate 948 Linux kernel privilege escalation patches in a single run, and system cards show its violation rate decreased in tasks such as exploitation.
Source ↗Claude Launches US Personal Finance Feature Beta
Anthropic launched Personal Finance in Claude, available for US Pro, Max, and Ultra users to try starting this week. Users query spending and bills through Plaid read-only connections. Investors accused Anthropic of launching a similar product after contacting competing teams, and Anthropic has not commented.
Source ↗Local Data Center Restrictions and US Tech Stock Decline
Local voters in multiple US locations voted to support restrictions on new projects, Wall Street market cap dropped significantly in a single day, and the Nasdaq and semiconductor sectors moved sharply lower. Investors harbor AI safety concerns and worry that Democratic control of Congress will slow high-leverage AI spending, with executives such as those at OpenAI discussing compute capacity limits in private.
In brief
ChatGPT Images 2.6
OpenAI updated its image product to support GIF and reference image creation, with API generation speeding up by 50%.
Source ↗Gemini Omni 1.2 Flash
Google released a new video generation version, raising the limit to 15 seconds, with API prices down 30%.
Source ↗Perplexity Teammates
Launched work-oriented agents, shifting from conversational prompts to proactively issuing high-level goals.
Source ↗Baidu ERNIE 5.5 Flash
Open-sourced a sparse MoE model with 280B total parameters and 14B active parameters per token.
Source ↗Thinking Machines Lab Releases Inkling 2 Small
Released an MoE model with 594B total parameters and 17B active parameters, opening Apache 2.0 weights next week.
Cat Wu Joins OpenAI
Cat Wu announced joining OpenAI as Head of Product Management for Core Products.