Links?

One of the neat things about Deep Seek is they actually do something OAI and Anthropic do not;- They actually publish their methods, training code and all that.

Theres actually a lot more going on with these models than "Trained it on ChatGPT" (And its not like the chinese are alone in that. When claude first came out, it kept randomly claiming to be ChatGPT, a smoking-gun level sign of distillation, and OIA has *heavily* implied that they've caught XAI out multiple times trying to distill CGPT. And to be clear , amongst ML researchers, distillation is considered an entirely legitimate and normal part of AI training.). What is clever about deep seek is they've made pretty significant progress on getting a very big model and running it cheap and fast on more commodity software. And importantly , they've published how they do it in a reproducable way. They've also made *significant* progress on training methods that drastically lower the cost of training up a model the hard way.

So yeah, its more than just distilling western models. The chinese are actually making real progress. And..... Deepseek and Kimi and the like are actually really good models. You can do 90% of what the big american frontier models are doing with them, and on a budget, 90% is good enough.

Also, what could he possibly do out of jail that he can't already do in jail? Clearly he's able to talk to people, which he could do outside too. Nobody will listen to him inside or out, because he's a convicted fraud. Also in jail he gets 3 square meals per day. Outside he would have to get a job just so he could buy the meals to sustain himself.

Not that I think he should succeed in his appeal, but as someone who's worked at the DOJ and seen the inside of a prison multiple times (as a staffer), I can assure you being incarcerated is a miserable and demeaning experience. Zero freedom, awful food, extremely uncomfortable conditions (what you think those places are air conditioned?). Throw the non-zero chance of being stabbed to death because your neighbor, chuck the schizophrenic rapist , doesn't like your hair color, and your absolute inability to escape unpleasant people. Oh and months on end of 23hours a day locked in a single mindmeltingly boring room. For the next 30 years!

Yeah trust me any sane man would do anything in their power to get out of that very fucking unpleasant situtation,

Civilly, absolutely. Ultimately their machines did a harm to the other guys machines. They have to pay the fixem price. A judge would find that a fairly straightforward decision. Two agrieved parties, one harmed the other. No jugement as to culpability. If I accidently hit your car, I have to pay your repair. (Or rather my insurance would reimburse your insurance)

Criminally, there might be an argument for negligence? But it'd be a really hard case to prosecute, and frankly I suspect most prosecutors would rather be using their limited resources locking up rapists and muggers, than trying to stick bracelets on a guy who can turn up with the entire wall street legal bar as defence on a *very* cloudy question of culpability.

Honestly, I think they actually need to pass some laws to actually create a more straight forward theory of prosecution. "If you negligently let your clankers rape another dudes servers, you can be charged as if you broke into that server yourself. Incompetence is a partial defence. Genuine ignorance a complete defence" (Ie if you had no idea your robbie the robot had trundled down the road and held up the gas station, then yeah.... that might be a bit unfair to give you 30 hard years for owning the robot. But your ignorance that its possible should only be a partial defence if you had every opportunity to read the fucking manual where it says "Robbie is bit of a clepto, dont let him have a gun".

OM NOM NOM BOOTS

THAT IS WHAT THEY ARE DOING, BOOT LICKER.

You're a traitor in spirit, if not in law.

I fully agree with you.

AI, Claude or otherwise, isn't a magic bullet. But it does reduce the time it takes to produce code. It therefore reduces the cost of code, and I believe will bring down software prices, especially for "enterprise" software. These tools that facilitate the operations of a company, lend themselves well to AI coding, and companies are motivated to build their own enterprise software because none of the SAS tools do _exactly_ what they want.

I also agree that the market is overrunning its headlights. There are going to be a lot of investors losing money, when the bubble bursts. When that will be, is anybody's guess.

There is no surprise that their models (or any of the other major ones) can hack other sites. ALL of them are fully capable of this. The surprise is that they can't seem to build test environments that contain their tests.

I once ran a project that dealt with medical billing. The team kept noticing that some transactions went missing for no apparent reason. Eventually, they discovered that the pre-production environment was missing a firewall between it and production, so the pre-production process was grabbing production transactions and processing them.

This is NO different from Anthropic's testing issues. Their test environment wasn't properly set up, and the test bot acted on "production" sites.

I'm sorry that you've never apparently had a good manager. In my opinion, the best managers are people who are fully capable and skilled at doing the job of any person on the team. Sure, there are also some decent managers who have no idea what their team does. But such managers cannot be as effective as a practitioner.

They are a safety concern. They are creating an opportunity for a foreign adversary. Does anyone really believe these clowns are capable of maintaining security?

WordPress would probably improve if he did

They found that obvious and somewhat obvious lies still get them 95% of the effect. Hence why go for cleverly crafted when simple does the job nicely.

I mean on some level the AI fans are cult-like. A cult mind-set destroys the ability to fact-check.

Yes. Basically the wizard's apprentice all over. Again.

It may be that. Observable history makes that pretty likely. Or they may be facing genuine shortages of money and are trying to actually slow things down. They may also have noticed that they are doing very criminal things if they run models than then hack others.

Exactly. The question is however whether they can keep enough stupid money flowing to survive even if they slow down training. Merely running the models is still quite a bit more expensive than what they can charge for it and they lose money on every prompt.

I think we may be seeing the begin of the crash. Nothing is sure, obviously, as this tech has been massively overvalued for several years now bit they still managed to get mountains of more money.

To be fair, most of that trust was quite irrational to begin with. But exploiting irrational trust is what scammers do and the general consent is that these people still belong in prison.