OpenAI reportedly ditches model over safety concerns

In Brief

OpenAI had planned to merchandise yet different AI exemplary adjacent month, but has decided to nix the merchandise implicit information concerns.

The Wall Street Journal reports that Astra 6.1 was scheduled to beryllium released arsenic soon arsenic wrong the adjacent fewer days. However, the exemplary “showed higher levels of deception” than erstwhile models and exhibited unsafe behavior, the Journal writes.

Saachi Jain, OpenAI’s caput of information systems, told the WSJ that the exemplary tested poorly connected alignment, a measurement of however good the programme adheres to quality intent.

TechCrunch reached retired to OpenAI for much accusation and volition update the nonfiction if it responds.

Astra was released earlier this period and hailed by OpenAI arsenic its astir almighty exemplary yet.

Questions astir information person plagued the AI manufacture implicit the past respective months — ever since the Hugging Face incident, successful which an OpenAI cause broke escaped of its sandboxed situation and hacked respective antithetic companies. Since that incident, much models — including Anthropic’s Claude and Google’s Gemini — have been revealed to person exhibited akin behavior.

The deluge of concerning stories has, ironically, helped to propulsion the argumentation speech successful the U.S. toward an result desired by apical AI labs: the instauration of caller manufacture standards for AI information and perchance a slowdown of the manufacture itself.

Companies similar OpenAI and Anthropic person claimed that the interest present is safety, though different imaginable information posited by critics is that it could entrench the manufacture position of those companies astatine the detriment of little resourced firms.

Subscribe for the industry’s biggest tech news