If there are no white people to blame on either side of the war, nobody cares. It's the same shit with so many conflicts.

Comments URL: https://news.ycombinator.com/item?id=49905603

Points: 3

# Comments: 1

Comments URL: https://news.ycombinator.com/item?id=49886186

Points: 3

# Comments: 0

I predict it will keep trying and failing without changing strategy until the team is so overleveled that brute force works.

I apologize for reading that into your statement but there absolutely are a lot of people who claim that.

> they all used that to demonstrate their model's capabilities.

That's one interpretation. Another would be: the Overton window shifted, allowing them to admit it. The Chinese labs also had incidents [1] but they're burring them deep in technical report or maybe don't publish them at all. Do they not care about "pumping up their valuation"?

[1] https://www.forbes.com/sites/boazsobrado/2026/03/11/alibabas...

> Or, if they were following their own founding principles

The mindset is "AGI will be built by someone. We care about safety, so better we do it."

Quite flawed in retrospect, because it leads to this ironic race dynamic:

Either slow down and leave the frontier to the "wreckless" players. Or contribute but increasingly undermine safety yourself to keep pace.

That's how we are in a situation where the very labs that work on AGI give P(doom) > 10%.

The HuggingFace incident was audited by independent third party analysts. To "orchestrate" that and keep it a secret is like faking the moon landing. To many people involved. It's not feasible.

OpenAI was founded to develop safe AI. Then Anthropic split off because OpenAI was not safety focused enough. These companies have been railing about AI safety long before they had these big boy valuations. Many people at OpenAI/Anthropic explicitly joined to help make AI safe. This is not some top-down company value.

Also, I seriously doubt that "making panic" is actually good for the stock price. Usually any companies natural reflex is to cover up safety risks. That's not to say that these companies are angles. Of course they're pulling some shit but that doesn't mean that everything out of their mouths must be lie.

* commands run by me (! prefix) are also sandbox blocked

* agent has no way to _request_ unsandboxed execution (e.g. if `kubectl whatever` is rejected by the sandbox, the model should have the chance to request permission)

* does not understand shell composition patterns (e.g. if `git status` is allowed and `git log` is allowed, then `git status && git log` should be allowed automatically)

* sandbox only supported on mac or linux. not both

All of that can be fixed by yourself. That's certainly the spirit of pi. But if you want strong defaults and batteries included (like omp promises) then that's just annoying.

We started building that but it quickly turned out to be too narrow. Often we want agents to do task that have no input ticket and often the output is not a code change (Slack bot, incident investigatior, scheduled daily tasks, ...)