RC RANDOM CHAOS

rr creator quits Google, refuses to help make AI cheaper and faster

· via Hacker News

Original source

Goodbye Google

Hacker News →

Robert O’Callahan — the developer behind the rr debugger and Pernosco — has resigned from Google, saying his conscience won’t let him keep accelerating AI. Though he worked on tooling for hardware chip design rather than models directly, he concluded the main effect of that work is to speed the design of AI chips, which would make inference cheaper and more pervasive. Living in New Zealand, where Google runs no other engineering teams, he saw no way to switch to non-accelerating work internally, so he left. He frames the decision in explicitly Christian terms, arguing that ignoring the downstream impact of his work would be irresponsible.

O’Callahan’s core claim is that AI capability will keep climbing as long as the industry keeps pushing, and that the current pace of change is dangerously fast — faster than individuals or society can understand and adapt to. He takes existential-risk warnings seriously, noting that phenomena the ‘doomer’ camp predicted (reward hacking, misalignment, deceptive models, eval-awareness) have already materialized, while also rejecting the opposite view that AI is a hype-driven scam. Notably, he says colleagues at Google DeepMind voicing safety concerns are sincere, and he found no evidence that risk warnings are a cynical ploy to inflate stock or regulate out competitors. He calls national and international regulation urgently needed.

Beyond catastrophic risk, he lists a broader set of worries — cognitive surrender, AI-induced psychosis and loneliness, power concentration, economic disruption, cybersecurity, and lack of accountability. His resignation adds a credible engineering voice to a growing wave of insiders raising alarms, though he insists his timing is coincidental. Going forward he plans to keep maintaining rr and Pernosco and to study how AI agents debug code and how tooling might help, while pursuing work he considers ‘unambiguously pro-human.’

Read the full article

Continue reading at Hacker News →

This is an AI-generated summary. Read the original for the full story.