Why did they focus on that particular graphics card and not others, and not common laptops used by developers, or something like that?

The repo is very vibecoded (Claude is co-author, READMEs are obviously AI).

This is the output of someone with a 3090 who pointed Claude Code at some research papers and possibly the upstream implementations of these techniques and then posted the output as original work.

That's a pretty popular budget friendly GPU people use for local AI, it actually seems like an excellent choice IMHO.

Depends on your definition of budget friendly, I suppose. I was looking around the other day and the cheapest working 24GB RTX 3090 on eBay was $1800 CAD after exchange rate, shipping and all the rest.

Hugely inflated from the $700 they were once going for. Maybe there are still deals around.

Actually budget friendly is RTX 3060 12Gb.

With one you can run 9B/12B models which are fine for text tasks like chatting or summarisation. Not for precision like tool calling or code.

With two of them you can run models up to Qwen 27B and 35B with a few-turn context window (8k-16k). Dense at 14t/s and MoE at 68t/s.

With three of them you can run 128k context, though you'll need a large format case and the right motherboard or PCIe riser.

I'm running three and even with a new case this setup cost me less than one 3090.

That's insane. I bought two in December for ARS 1.2M (a little less than USD 1000). Maybe OpenClaw raised the demand.

Wild I paid $1000 CAD for mine 2 years ago, I guess things have changed.

Because they are hugely more useful now than running some stupid game at 240 fps instead of 60 fps.

They're not a particularly fast card compared to something like a 5070, they have lots of VRAM.

That's why they were cheap before.

Also "Some stupid game", who woke up and made you king of hobbies.

Radeon 9700 pro or intel arc b70 (both $1000-1400, 32GB, 650GB/s bandwidth), or ryzen AI max 390 (more vram, less bandwidth)

The local inference space is pretty good nowadays.

They get to skip 20 years of figuring out how work gets done in different industries if they can train on the information from everyone’s Jira and Confluence and whatever else Atlassian sells. That data makes Cowork and Claude better. And then you have an advantage that is more than just having the best model (which others are catching up on). It’s harder to catch up on this data moat.

Related front page discussion about Atlassian changing their data policies to train on customers’ data: https://news.ycombinator.com/item?id=47833247

There have been rumors all over social media that the exits from Atlassian’s C suite (allegedly firings) and this aggressive privacy violating change are all because their stock is down 80% and they’re trying to find the most valuable path to selling the company.

From what I read, Atlassian has data from most of the Fortune 500. And this metadata that they are wanting to train on, is data that Anthropic can use to build software for specific verticals (like they just did with Claude Design), or simply to make things like Cowork function better using this proprietary data collected over the last 20 years.

Karp thinks extremely highly of himself above all. He has been outspoken for a while but he’s been becoming increasingly unhinged recently. We just had a post about how Palantir is basically advocating for mandatory military service (https://news.ycombinator.com/item?id=47828218). He’s also been sharing his views about how neurodivergent people - as vague as that term is - are superior to others. He thinks they will be the only people who can be successful outside of trades (or something weirdly supremacist like that).

To some extent I understand and am okay with people sharing their views even when they’re drastically different from others or plain unhinged. But this is a person whose company participates in surveillance and holds many government contracts. That makes me less comfortable.

The thing I can’t get over is how sort of non-specific and unideological it is. Someone somewhere is dangerous and/or not as good as us therefor we need to be prepared to kill millions of people.

This seems exemplified in practice by going into Iran to support the civilian resistance than transitioning to, we’re going to kill them all and turn Iran into a sheet of glass.

I don’t have a problem with the idea of national service (though maybe we should start with domestic service). But this ideology that’s like, we’re libertarian billionaires who want to rule the USA, but you should sign up to die for your country?

Maybe in a Democracy.

This just feels like bizarre marketing, to try and make an LLM seem more human than it is. And if users have to treat it nicely or manage its emotions, I would argue those are flaws.

Is there any requirement that a business who claims a refund must pass it on to customers? If not, this is just another theft of tax payers money.

Have you seen the near daily complaints about how Claude is getting worse or more expensive? I feel like there have been many recent posts like that, but it’s not limited to just here either. It seems like a lot of people are feeling like Anthropic is at least being not transparent, although many would say deceitful.

The later parts of this article listing out the dark patterns and security issues and privacy issues is great. Spyware may not be the right term but there is a lot that is wrong here and Anthropic absolutely should be called out for it. Many people and businesses are trusting what appears to be a mix of vibe coded slop and aggressive anti user growth hacks. So much for Anthropic’s high and mighty moralizing.

The problem is Trump’s family and friends and donors and people who have otherwise bribed him all can benefit from actions the administration takes. It’s not as simple as restricting the current officials.