This page is a roundup of several news items. Claims that large numbers of ChatGPT accounts were suspended sparked debate in Chinese-speaking communities, but OpenAI has not yet made an official statement. Binance's CFO resigned one week after a hacking incident, and the company said the reason was personal.
OrcaRouter has released an API for a Hy4-based model with reduced refusal responses. It is not a service anyone can use right away. Users must go through approval steps such as business verification and agreeing to terms. Connecting is done by changing only the address and key in existing OpenAI tools.
Since 2023, OpenAI has signed content deals with more than 20 media publishers. These publishers are often cited in ChatGPT answers, along with their sources and original links. According to one analysis, 10 of the 50 sites that ChatGPT cites most often are these partner sites.
OpenAI's coding tool Codex has released a new test version, 0.169.0-alpha.3. The release post only lists the version number, so it is unclear what has changed.
Stefano Fabbri and Oliver Steffmann of OpenAI's applied AI team present a hands-on workshop video on building financial workflows with ChatGPT and Codex. The video covers how to set up AI agents (AI programs that carry out multi-step tasks on their own) for finance work.
The YouTube channel Two Minute Papers has released a video that explains OpenAI's 'Astra weight hack' in simple terms. The topic appears to involve weights, which are the numerical settings AI models learn during training. The video description does not give details, so you need to watch the video to see the specific method.
OpenAI CEO Sam Altman has said that his comments about memory in home PCs, made on a podcast, were quoted without their full context. He said he has long argued that memory production needs to increase. However, the article notes that as supply runs short, memory is still flowing toward data centers.
'KAIL Fault Ratio,' a traffic accident fault-ratio analysis service made by a Kookmin University research lab, has passed OpenAI's review and is now officially available as a ChatGPT plugin (an add-on feature). Users can start using it right away without signing up separately.
In tests by AI evaluation firm Vals, some publicly available models completed a significant share of simulated penetration testing tasks. The results were measured under specific conditions in a simulation, so they do not indicate actual hacking success rates.
Trader Arthur Hayes warned that individual investors who buy shares of SpaceX, Anthropic, and OpenAI at the IPO price could suffer big losses when insiders sell after lock-up periods end. He argued that if share prices fall below the IPO price, the funding flows for AI company bonds and the crypto market could also be shaken.
OpenAI has raised $11 billion as the first stage of a funding round that values the company at $1.25 trillion. This is still short of the original goal of up to $30 billion. The company's valuation target is also lower than it was at the start.
Emirates airline is shifting its 2026 marketing focus from selling flight tickets to promoting itself as a premium lifestyle brand. Content made specifically for TikTok and Instagram, sports sponsorships, and the use of AI (artificial intelligence that thinks and responds like a person) are moving forward together.
OpenAI has released BioBasicsBench, a benchmark that tests whether AI agents correctly interpret raw data obtained in laboratories. It consists of 100 tasks created with Ligo, an enzyme design company. The top-performing model scored 66.7%.
OpenAI's coding tool Codex has released version 0.169.0-alpha.2. Alpha versions are test releases that people can try before the official launch. The announcement gives only the version number and no details about the changes.
Abdulrahman Alfozan has denied charges of eavesdropping on OpenAI. He could face up to five years in prison if convicted. More details can be found in the original article.
OpenAI has introduced LifeVal, a new benchmark (a test standard for comparing performance) that measures AI performance based on tasks that matter in daily life. According to the video title and description, it was built with feedback from thousands of ChatGPT users.