The 52-Page Memo That Nearly Destroyed OpenAI: Ilya Sutskever's Deposition

medium.com

37 points·SilverElfin··7 comments

I wish I lived in the timeline where Sam was fired and OpenAI merged with Anthropic. I think the world would have been A LOT better.

How would less competition make the world better?

Yes. Competition is not the goal.

"How would less competition make the world better?" In the context of Umbrella Corp and bayer. Both undoubtedly bad, but one so evil it is irredeemable.

Hmm. I wonder how openAI, in the direction it was taken under Altmans leadership, is a fundamentally and deeply rooted evil thing to allow to exist. It really boggles the brain.

It's like saying "Therapy needs competition, this is why we're allowing PsychoTheRapist to continue practicing, to prevent psychotherapist's from getting to cocky"

Which is odd because the two seem to have fundamentally different purposes with immediate and long-term impacts of the former being unambiguously negative

Besides, strangling OAI in it's cot, as the natural order of things would be, wouldn't get rid of competition. Competition is fierce via Chinese and European offerings.

Someone would have connected gpt to bash. As it turns out that someone ended up at Anthropic.

I can't use Claude Code because it runs out of memory during installation. I'm running my development environment inside a VM for sandboxing and I'm unwilling to allocate more RAM to it. What you're suggesting would basically lock me out of agentic coding or into getting a computer with 64GB RAM to deal with the perpetual memory leaks Claude Code is famous for.

I run into quota violations, which eventually kills my terms and emacs. The agents gets killed with the terms of course.

Root cause: tests running in loops, filling up space in /tmp, but I see it only with lsof, not ls nor df. Claude didn't use timeout.

Reply on news.ycombinator.com