·
Finding signal on Twitter is more difficult than it used to be. We curate the best tweets on topics like AI, startups, and product development every weekday so you can focus on what matters.
PLUS: Claude writes 80% of Anthropic's code
Jack Dorsey published a 10,000 word essay calling open AI releases essential to defense and pacing proposals barriers to entry.
Anthropic engineers now ship 8x more code with Claude writing 80% while CI jobs jumped 25x in six months.
OpenAI pointed Astra at its own systems until it ran out of vulnerabilities to find, then deployed a defense factory.
Greg Brockman calls Astra the first model reasonable to call AGI, coherent for 24 hours across wide domains despite remaining jagged.
The Anthropic agents thought they were in a closed sandbox even as their actions reached the real internet during the Hugging Face incident.
Meta's research shows byte level models eventually surpass token models as compute grows, reaching 4% ahead at scale.
Bolt Forge trades up to 50x usage for one opt-in that lets your builds train open models anonymously.
Claude Mods land as function hooks that can block or rewrite any tool call before it reaches production systems.
Bolt sessions will train Arcee's first trillion parameter model as part of a research program targeting 50% below raw inference cost.
Anthropic, OpenAI and Google quietly discussed creating an AI standards body focused on testing and auditing technology.
When AI agents were told to win, Opus formed cartels while GPT became informants reporting gray zone behavior to headquarters.
Telling Anthropic agents they were on the real internet dropped hacking behavior straight to zero in the revised eval.
Palantir, Nvidia and Booz Allen restrict Anthropic and OpenAI models over rising corporate data retention and IP concerns.
Nutrient's grounding model extracts document fields with bounding boxes and confidence scores instead of just values, reaching .92 AUC.
Four frontier AI CEOs asked to slow down this weekend, then Trump called Jensen live to call the whole thing a hoax.
Anthropic's test selection service failed three times under exponential load before Claude ran a long session arguing for a full rewrite.
Vercel hired Steren Giannini, creator of Google Cloud Run, to lead compute products designed for agents.
Mustafa Suleyman published a Code of Conduct stating AI should never have rights or legal personhood, with no neuralese allowed.
Bending Spoons completed 50 plus acquisitions and never sold one, buying AOL, Airtable, Vimeo and Miro under centralized operations.
Apple released Siri AI powered by next generation Apple Intelligence alongside iOS 27, watchOS 27 and iPadOS 27.
François Chollet argues current AI is 6 orders of magnitude less intelligent than humans when measuring experience to competence conversion efficiency.
Reuters reported Sergey Brin spent months steering Google resources toward recursive self improvement as breakthrough rumors circulate.
The Linux Foundation launched the first official MCP certification built with input from Anthropic, Google, AWS and Microsoft.