In my mind, the government would be fully happy to use this to surveil citizens (and indeed anyone) with or without any legal basis, but the issue was that Anthropic has a safety stack / training and inference protocols that it follows. Refusals, abuse models, and manual guardrails. They didn’t want to shut those off. Likely there were some very basic technical reasons, some being that the team’s safety posture is fully ingrained in the model itself and thus difficult to remove.

In this document, OpenAI admits that while they are not “turning off” their safety stack, they are completely willing to provide the government with a different model, different guardrails, etc. That should be incredibly concerning. Anthropic was unwilling to do this, cited their ToS, and ultimately had to walk away from the deal. Given that the government (DoW really) framed this in terms of a hilariously stupid position (surveillance and autonomous weapons), Anthropic felt that this was something they could voice to the public and therefore the entire guardrails discussion turned into a “we want the language changed”. Also the government can’t actually compel Anthropic to create new guardrails so they had no choice but to raise the stakes, make this a moral thing, and basically accuse Anthropic of being woke.

IMO this is really sad for OpenAI employees. Yet again Sam Altman proves that he wants to weasel his way around public perception. Folks at the company have to grapple with working for someone of that disposition.

Another way to frame it is what would you do in a low trust environment where corporations and the government were not to be trusted. You would likely avoid things like bubble bursting AI stock investments, jostling for rank in a company, etc.

_at this power density, cooling isn’t mandatory - it’s physics_

EDIT: I understand now. $30 for input, $150 for output. Very confusing wording. That’s insanely expensive!

Comments URL: https://news.ycombinator.com/item?id=46720999

Points: 4

# Comments: 0

Of course, this kind of interactive deep engagement with a topic is fast becoming obsolete. But the essence to me of “knowing” is about doing and experiencing things, updating my bayesian priors dialectically (to put it fancily)

The problem however lies in who or what does this rebasing in a multi-tenant environment. You sort of need a system that can do it automatically, or one that gives you control over the process. For example, jj can often get tripped up with branch rules in git since you might accidentally move a bookmark that isn't yours to move, so to speak.

I’m not a database neckbeard but I’ve always been confused how GraphQL doesn’t require throwing all systems knowledge about databases out the window

Another pro not mentioned is that native integrations (i.e. obj-c on macos) are much easier to do since rust has great ffi integration with other native libraries.

The biggest pro to electron is that it has extensive plugins that are often widely used in production by large companies. But Tauri is definitely winning and any new project should use Tauri no matter what essentially.