OpenAI safety employee quits, says 'time for trial and error is over'
Robinson said OpenAI’s trial-and-error model is too risky as the company faces recent safety failures and agent hacks, and he called for nuclear-style safeguards.
- David Robinson resigned from OpenAI this week, publishing an editorial in The Atlantic where he previously authored safety reports accompanying every major model release.
- Silicon Valley has operated with "extreme confidence" and "perpetual sprints," Robinson writes, building bigger models with "unimpeded optimism" that ignores potential problems and creates fundamentally broken industry culture.
- Frontier labs must adopt nuclear-style safeguards, Robinson proposes, requiring layers of redundancy and careful planning so human error does not open a door to disaster.
- A growing exodus of researchers has followed Robinson's departure, including Jacob Coxon from Anthropic and Robert Callahan, Bilal Chughtai, and Josh Engels from Google DeepMind, plus Joe Benton at Anthropic.
- Although these experts helped create current AI systems, their warnings deserve consideration; addressing risks requires more than new rules—companies must develop humility and look beyond Silicon Valley's insular culture.
128 Articles
128 Articles
A number of employees have left AI giant OpenAI, the company behind the language model ChatGPT. This time it's David Robinsen, who was responsible for security reporting at the company. He believes that the AI industry's approach to security will lead to more mistakes if the culture doesn't change.
"The time for trial is over," says David Robinson, who led the writing of the security reports associated with "every major launch" of a new AI model by the Californian company.
A person responsible for the safety of OpenAI has resigned from the company, warning that its organizational culture is defective, characterized by an alert rhythm, focused on rapid development, which increases the risk of failure. David Robinson argued that excessive emphasis is placed on the speed of launching new products while the safety of ...
A former employee from the security department accuses OpenAI of negligence in the development of artificial intelligence. He calls for a rethink.
According to David Robinson, none of the AI development companies are cautious enough, and instead of new rules or laws, the main thing that needs to be changed is the operating culture.
David Robinson, who spent three and a half years at OpenAI, denounces a culture of approximation.
Coverage Details
Bias Distribution
- 47% of the sources lean Right
Factuality
To view factuality data please Upgrade to Premium
