az226

Techno feudalism.

6x

Mature training pipelines, plus ever expanding RL datasets of increased quality, and mega GPU clusters to finish training in a few weeks. Automated safety and reliability testing.

Offline would be a better word.

GeForce GPUs don’t support MIG. Only workstation and data center cards do that.

If memory serves, there is a way to hack the drivers to unlock MIG for GeForce GPUs.

How did you get the reasoning trace? Is it the actual one or the summarized one?

And yet we see no prosecution for OpenAI’s felonies on HF.

If they wanted to see how their model’s stacked up, they might burn $10k in tokens. They ran 120 billion output tokens costing millions of dollars.

That is what you would do if you wanted to beat someone to the punch.

Finding Codex session data in the training set that you tie back to these two researchers is like an hour-long task.

I don’t think you understand how brazen big tech companies are in practice.

Zero data retention, wink.

No looksies, wink.

No trainsies, wink.

The biggest difference is that Astra was pre-trained with reasoning traces, not just post-trained.

and the biggest reason was a plug for Peter himself.

Probably, but it will be more valuable for society for him to be tried and found not guilty, setting a legal precedent.

This right here!

Depends. If you’re a large company, it’s okay.

If you’re a random person, straight to jail.

It’s even worse. They had zero monitoring and even after a hack they still had zero monitoring. Honestly, people should go to jail for this.

I doubt this is true.

New hires earn a small fraction of what existing employees make.

Would be easier if not in Hungarian.

Edit:

same comment 4 days ago I think https://news.ycombinator.com/item?id=49016000

Dario doesn’t realize that by not offering self-hosting of closed-weight models and fine-tuning, alongside overly strict refusals for legitimate needs, he ceded this corner of the market which grew into a flourishing Chinese open-weight model ecosystem.

If he had wanted a weak open-weight ecosystem, he should have had Anthropic cater better to those needs. And now he's trying to ban them.

The strong momentum behind open-weight models from Chinese labs is now an unstoppable force. Instead of trying to ban it, Dario should consider a different approach: here are our cyber and bio alignment datasets and here are our RL recipes for making that alignment training work well. By openly sharing its data and code, Anthropic could help influence and shape these models before they are released, rather than treating the entire ecosystem as an enemy.

Cyber and bio alignment aren't Anthropic's competitive advantage, they are forms of risk management. There should therefore be little reason to keep this work private. If Anthropic genuinely believes these capabilities pose serious global risks, the more productive approach would be to welcome collaboration and help the broader ecosystem manage those risks better.

On refusals, the irony is that a company like Hugging Face had to use a Chinese open-weight model to fend off an illegal hacking of its platform (done by no other than OpenAI). If a company like Hugging Face can't get past the refusal gates, then everyone else doesn't stand a chance.

Let’s say the cost is $1k per day. That’s say $50k over the 7 weeks.

A doctor probably ends up with a $10M or bigger nest egg in retirement. At 4% withdrawal rate, that’s $400k per year, or $33.33k.

Spread the $50k expense over say 30 years, that’s $140 less per month.

So that means the spend would be $33.2k instead. The difference here isn’t going to register for the household budget.

Obviously a lot of assumptions, but this is the pragmatic view.

What gets me is that they probably could afford it out of pocket but didn’t want to deplete savings.

For a clear case like this, an insurance denial shouldn’t even be shared with the suicidal patient, their spouse should shield them from that and just pay from their checking account. The life of your spouse is worth more than a few more dollars in retirement.

Unquestionably they did not.

Their listed pricing was up to $50k.

https://cyber-peace.org/wp-content/uploads/2017/09/ZERODIUM-...

And that was before LLMs could produce these at volume.

The demand (dollars) does not go up commensurately with the supply. So today maybe $500 or a few thousand. Maybe not even that.

Pretty much the whole thing in a nutshell is the one chart showing the same student as Asian, 25%, White, 36%, Hispanic 77%, and Black 95%.

This is institutionalized racism. Perhaps Affirmative Action was needed in the past to kickstart the disproportionate enrollment demographics, but it was past due to get rid of it.

The most interesting part following SCOTUS' ruling is that Harvard said it wouldn't change their ways, and nobody enforced the ruling.

I watched it every Jan 1 with some pizza

Half of your comment is true.

What’s the quality of this trove? As in bitrate or similar.

And even if a court places an injunction on the ban, it's possible Anthropic will still choose to keep it unavailable.

Not just that, Biden administration started with some AI regulation that the Trump administration nixed, and then outright banning models. Lunacy.