OpenAIOpenAI
OpenAI says it will no longer train its AI models on the live internet, citing “significant risks”
The company links the decision to significant risks, but details of the new data approach are not yet known.
Archive
All published reports
Pyxory’s chronological feed for search, citation and topic navigation, newest first.- OpenAI
OpenAIOpenAI
OpenAI strengthened isolation of the training environment for frontier models
OpenAI reported tightening isolation of the training and evaluation environment for advanced models
OpenAIOpenAI
Sam Altman calls sandboxes a temporary protection against advanced AI
Sam Altman, according to The Decoder, believes isolating advanced AI is a temporary measure and alignment remains an unsolved task.
OpenAIOpenAI
OpenAI and Wiz discovered ancient vulnerabilities with AI
TechCrunch ties the discovery and prevention of ancient vulnerabilities to AI, but details are still scarce.
OpenAIOpenAI
OpenAI and Poland’s NASK Partner to Advance AI Safety and Security
OpenAI and Polish NASK announced collaboration in the field of AI safety and technical research.
OpenAIOpenAI
OpenAI makes Jev competitor widely available
OpenAI is putting a decision-making model into wide availability, but details of its launch and capabilities have not yet been disclosed.
OpenAIOpenAI
OpenAI prepares a round of $30 billion at a valuation of $1,25 trillion
TechCrunch AI reports on a possible OpenAI round of 30 billion dollars at a valuation of 1,25 trillion dollars before the IPO.
OpenAIOpenAI
OpenAI acquires Orchestra Bio for Biotech Research
OpenAI reports the acquisition of Orchestra Bio to bolster AI tools in life sciences research; deal details are not disclosed.
OpenAIOpenAI
OpenAI improved models for legacy programming languages
OpenAI announced improvements to models for COBOL, Fortran and other low-resource languages, but available data do not reveal the assessment methodology.
OpenAIOpenAI
Sam Altman reportedly pitched three endgames for AI safety, one involving a single firm melting everyone else's GPUs
The Decoder attributes three AI-safety scenarios to Sam Altman
OpenAIOpenAI
ChatGPT Gains a Personal Chat Lock Feature
TechCrunch AI reports a feature that should hide personal conversations when a device is handed to another person.
OpenAIOpenAI
Replit designates OpenAI as default model provider instead of Anthropic
Replit has chosen OpenAI as the default model provider instead of Anthropic, but access to Claude on the platform will be preserved.
OpenAIOpenAI
OpenAI Aligns Safety Measures with Distribution Partners
OpenAI described collaboration with distribution partners, uniform safety measures and clarification of the Terms of Use.
OpenAIOpenAI
DeepSeek Opens DeepInfer for Inference of Model V4.1-Flash
DeepSeek opened DeepInfer, the inference framework for V4.1-Flash. The Decoder reports reduced computing costs and energy consumption, but available data are insufficient to assess the scale of the effect.
OpenAIOpenAI
White House wants to vet OpenAI and Anthropic models before Britain
White House seeks pre-approval of new OpenAI and Anthropic models by American agencies before their transfer to British testers.
OpenAIOpenAI
Sakana AI hires Jürgen Schmidhuber as scientific advisor
Sakana AI has appointed Jürgen Schmidhuber as Chief Scientific Advisor and linked his work to a laboratory for recursive self-improvement.
OpenAIOpenAI
ChatGPT Ads expands advertising to Southeast Asia and Taiwan
ChatGPT Ads enters Southeast Asian markets and Taiwan, but launch details remain undisclosed.
OpenAIOpenAI
Sam Altman Discussed AI Safety at the UN Security Council
Sam Altman discussed AI safety, human control, and international cooperation at the UN Security Council.
OpenAIOpenAI
ChatGPT Voice gains access to email, calendar, and Slack
ChatGPT Voice has gained access to email, calendar, and Slack, but launch details remain undisclosed.
OpenAIOpenAI
OpenAI Marks Two Years of Academy and Expands Access to AI Skills
OpenAI announced the second anniversary of OpenAI Academy and the expansion of the educational initiative's reach.
OpenAIOpenAI
OpenAI claims its internal model solved over 100 long-standing math problems
OpenAI claims to have solved more than 100 open mathematical problems, but available information lacks proofs or verification details.
OpenAIOpenAI
OpenAI Expands Ukraine's Access to Cyberdefense Program
OpenAI announced an expansion of the Ukrainian government's access to the Daybreak program for the cyberdefense of civilian infrastructure.
OpenAIOpenAI
OpenAI and Grab launch AI training for 30 000 partners
OpenAI and Grab have launched a program to teach practical AI skills to 30 000 partners in Southeast Asia.
OpenAIOpenAI
OpenAI Appoints Patreon Co-Founder as Head of Creator Product
OpenAI has created a Creator Product division and appointed Patreon co-founder Sam Yam as its leader.