OpenAI critic says GPT OSS looks safe enough for an uncensored release
aiamblichus · x · 2026-07-21
The poster argues that GPT OSS is not a safety risk and asks why OpenAI-style companies do not release an uncensored version if the model is already safe enough. They claim that, in practice, “AI safety” has mostly been used to reduce reputational and legal risk for companies, not to protect users.
In a reply, Aidan Clark says they should stay far away from frontier model development and defends iterative deployment: they believe advanced models will eventually pose a risk of significant harm, that it is hard to know in advance exactly when that threshold is reached, and therefore deployment should proceed carefully with updates along the way.
Related event: GPT-OSS Openness vs Safety: Aidan Clark Responds to Critics(9 posts)→
More from AGI Musings
- Competition is pushing AI forward faster than ever, the post says — eyishazyer · 2026-07-21
- Why adversaries can make even coin flips predictable — snikolov · 2026-07-21
- Scott Aaronson says AI theorem proofs do not mean P vs NP is around the corner — fortnow · 2026-07-21
- AI makes polished analysis look authoritative, but humans still matter — aloncarmel · 2026-07-21
- AI guilt debates look arbitrary next to the water footprint of dairy milk — PeterDiamandis · 2026-07-21
- A note on David Bessis: mathematical intuition can be trained over time — paraschopra · 2026-07-21