OpenAI’s ex-policy lead criticizes the company for ‘rewriting’ its AI safety history

High-profile Ex-Openai policy researcher, Miles Class, Taken to social media On Wednesday, to criticize Openaii to “rewrite the date” of the placement approach to potential risk AI systems.

This week ago, Openai made a publication document The process of designing the existing philosophy in AI security and adaptation, desirable and explainable ways. The document said that as a “sustainable road” as a “sustainable road”, “a continuous way,” as a “sustainable road”, which requires “a continuous way,”, which requires “a continuous way.”

“In an undisputed world […] Safety classes are clearly from treating today’s systems to the forces that are clearly visible. [which] The approach we received [our AI model] GPT-2, “Openai wrote.” Now we see the first AGI as a point in a number of systems in growing usefulness […] The path to make the next system safe and useful in the world is to learn from the current system. “

However, CLuntage, GPT-2, in fact, this claims that this is “100% consistent” with Openai’s iTedative placement strategy today.

“The open release of the GPT-2 I participated was 100% consecutive [with and] Openai’s Iterative Placement Philosophy, “Weak wrote a post in x. “The model has been released as classes shared with classes in each step. In this case, many security experts thanked this caution.”

As a research scientist of 2018, he joined Openai and became the head of the company for several years. Openai’s “AGI training” team, Openai’s AI Chatbot Platform Platform Platform Platform was in a special focus on the responsibility of language generation systems.

GPT-2What Openai announced in 2019 was a generation of strengthening EU systems Chatgpt. GPT-2 can answer questions related to a topic, can summarize articles and sometimes can sometimes create text in an integral level from humans.

While GPT-2 and its consequences can be seen today, they were also advanced. Openai, who wants the risk of harm, first refused to issue a source code of the GPT-2 source code, instead of giving you access to a demon instead of the selected news stores.

The decision was met with mixed reviews from the AI ​​industry. Many experts claimed to be the threat created by GPT-2 It was exaggeratedand said that there is no evidence that it can be abused on the ways described in the clear way and there is no evidence. A Published Published AI went up to date to publish skin open letter The release of Openai’s model, claiming that it is very technologically important for it.

Openai, in a few months later, after a few months later, six months after the opening of the model, GPT-2 left the partial version of six months. The choice thinks this is a right approach.

“Which part [the GPT-2 release] Considered or placed in the suspension of AGI? None of this, “said, X.” Was there a post about this “this caution” was an unathealed ‘ex Ante? Ex post, this prob. It would be good, but it doesn’t mean it is responsible for it [sic] then informed. “

In addition, Openai’s goal is to create a proof of proof of “concerns” and “anxious threats to move themselves”. This claims that he is a “very dangerous” mindset for advanced AI systems.

“If I would still work in Openai, why would I ask for it [document] It was written on the road and openly opentai hopes to achieve poo-pooing caution in such a lop party. ”

Openai has historically Defended Prioritize “Bright Products” to the security account and Hurry product releases Rival companies to beat the market. Last year, Openai solved the AGI training team, and a string company of AI security and policy researchers set off for competitors.

Competitive pressures are just exposed to the ramp. China ai lab deepseek He openly caught the world’s attention in his hand R1 Model that adapts a number of “basic” model “substantiator” in a number of basic criteria. Openai CEO SAM Altman has receiving This DeepSEEK reduced Openai’s technological device and gossip He “pulls some releases” to better compete Openai.

There is a lot of money on the line. Openai loses billions every year and has a company It was reported to have something He predicted that the annual cases will increase to $ 14 billion by 2026. The faster product release period is the term close to Openai’s bottom line, but perhaps the safety costs. Specialists as a luxury question, regardless of trade.