OpenAI at odds: Scientist hits the brakes, company steps on the accelerator

[Financial News] Jakub Pachocki, chief scientist at ChatGPT developer OpenAI, warned of the risks of artificial intelligence (AI) and called for the industry as a whole to slow the pace of development. However, a separate report released by OpenAI the same day emphasized accelerating research, creating an apparent disconnect.
On the 6th local time, OpenAI chief scientist Jakub Pachocki expressed concern in an essay titled “Intelligence from Beyond,” writing, “No one is preparing for the consequences that could result if machine intelligence continues to advance at its current rapid pace.”
He said modern AI is “cultivated rather than designed,” adding that “it is difficult for humans to fully understand how AI works.” Because AI grows more like a living organism than something built according to a blueprint precisely drawn by humans, he argued, it is difficult for people to control AI completely. He also stressed, “Safety measures such as Chain of Thought (CoT) monitoring, which was introduced to monitor AI, are gradually deteriorating as models become more advanced.”
He rejected the AI industry’s argument that development must continue in order to build defensive systems, saying, “Building defensive systems must not become an excuse for justifying reckless behavior.” He continued, “Not a single research lab has yet sufficiently solved the alignment problem of ensuring that AI refuses to take actions that threaten humans,” and added, “I hope voluntary pace-setting becomes widespread until shared safeguards are established.” He further said, “International cooperation on AI development should become the top priority for governments around the world.”
Sam Altman, OpenAI’s chief executive officer (CEO), also quoted the essay by chief scientist Pachocki on social media and expressed his support, calling it “an important essay.” Altman had consistently been skeptical of calls to slow AI development, which have been raised repeatedly over the years. However, following the recent hacking incident involving a Hugging Face GPT model, he has changed his position and now supports calls for a slower pace.
However, a separate report issued by OpenAI that day, titled “Accelerating Research,” drew attention because it contained somewhat contrasting content. In the report, OpenAI said it had “achieved the goal of an ‘automated AI research intern’ capable of improving its own AI and performing coding,” and presented a new goal: “We will improve it by March 2028 and develop an ‘automated AI researcher.’”
The report argued that this kind of automated AI research could instead help solve the alignment problem, revealing a view that contrasts with that of chief scientist Pachocki.
[email protected] Hong Chae-wan Reporter