disgruntledphd2

Yeah, but the Chinese might see them! \s

Like, it's wild to me seeing majority Chinese ML teams for years in the Bay and elsewhere that suddenly in the LLM space they are only capable of competing by stealing the hard stolen goods of our frontier US developers.

Argon is definitely competitive (used it over the past week or so).

It's pretty slow, but comprehensive. Caught a bunch of stuff that GLM-5.3 flash (which is incredible value) missed, and gave me good solutions for some irritating issues.

It's very slow (even by frontier standards), but it's hard for me to believe that Anthropic have much of a moat.

I mean, I like talking to Claude more than basically every other model (the open models are much more entertaining though, especially their thinking) but that doesn't feel like it warrants their valuation.

If they can successfully move towards owning their compute, and optimise it massively then they might do well, but it's a risky proposition from my perspective.

> software stacks that kind of work and no one really understands how they work

We've been heading that way for a long time before AI. AI is definitely making it worse in many cases, by lowering the bar to getting something that "works".

Like, you could write a web server in R. You'd need to bang your head against a wall for a while, which might make you think again, but with LLMs it's just a machine that is insanely goal oriented working very quickly.

This is most apparent when you can see the actual thinking tokens, not the sanitised versions that the US labs show.

Fundamentally if I am paying for these tokens, I should be able to see them but that's not where we are right now.

> On what metric? I just showed you JPMorgan has an equity value a quarter of its liabilities. JPMorgan is not insolvent.

To be fair though, JPMorgan (and banks more generally) are very, very different from Anthropic/other labs.

To be fair to them, if they have an expected output this can be exactly what you need.

They've been super useful to me when comparing datasets during a porting process.

But if they don't have that loop, then they make assumptions and build towers of madness on top of said assumptions.

> Older IPOs like Facebook felt different, like the hard part had been done, and they just had to scale.

Facebook were actually forced to IPO, because they had too many shareholders to remain private. They were super generous with allowing early employees to sell equity, and the laws required them to list once they passed 500 (i think) external shareholders.

This got changed in later legislation, which is part of the reason that companies like Stripe have been able to remain private for so much longer.

That being said, I agree with basically everything else in your comment, including the feelings around this.

LLMs (and Claude more specifically) are wonderful products, but I'm not convinced that they can justify their valuations over time.

> Occasionally there are long periods where returns are low (or even negative), but in terms of risk it’s the least risky place to put your money compared to all the alternatives (if properly diversified).

This is a very US-centric viewpoint, my recollection is that the FTSE didn't beat inflation for like a 20 year period.

That being said, index funds focused on equity are definitely a good way of getting reasonable returns over time (as long as you don't retire into a massive crash).

It's possible that Astra may have been trained on this kind of work, whereas Claude hasn't been.

Also possible that it's leakage (which honestly does seem like the most likely explanation).

As a reminder, if you see incredible performance from any statistical model, it's most likely leakage.

I totally agree, but this is still super cool, and I'm really happy that the hype around this paper brought this line of research to my attention.

> I think it's highly doubtful we're going to see software generally regulated as finance,

Probably not all of software, but basically all of our financial infrastructure, operational infrastructure and communications infrastructure is either already critically dependent on software, or heading that way.

I think that the EU's DORA act is probably the way things will head. We'll regulate the outcomes and ensure that these services can be maintained.

Honestly, the only reason that this hasn't already happened is that the US government is wildly against basically all regulation of software. I figure that if we end up in some kind of cybersecurity crash (likely, given the current trajectory) that we'll get lots and lots of regulation.

And to be honest, this will be for the better in the long term, but it'll probably suck for all of us in the short term (but at least there'll be jobs).

AI in software engineering is very unlikely to go away.

The real question is how much other knowledge work can be sped up by AI, as that's what the valuations depend on.

My guess is lots, but it will take a lot longer than the boosters expect.

Claude has really really good vibes. I'm not sure why, but it feels more pleasant to interact with than other models.

Nah, they were planning for 30bn two weeks ago, so this is a bad look if nothing else.

You should probably read the S1, they showed a mostly nominal loss due to accounting rules.

They have a lot of cash, but very few assets and a voracious hunger for renting compute.

Don't get me wrong, they make a fantastic model, but the business success is yet to be seen.

Even if you're a shareholder, the long term gains of an investment are much higher than the short term ones.

The trouble is that executives are bonused on much shorter time scales, and they run the company.

Yeah, but this is bottlenecked by both your ability to observe and understand the system, which is precisely the thing that's being reduced by careless use of AI tools.

As software intermediates more and more of the physical world, we're gonna end up at least as regulated as finance, and potentially as much as aviation.

The only real question is how much damage will be caused before that happens.

And just to pile on, payments is an incredibly competitive space, and operational excellence is non negotiable for someone like Stripe, as they are a higher priced provider.

I think that it was very strange how this became such an issue. Like, as long as it doesn't hurt anyone, who cares about another's gender expression?

But it's such a tiny proportion of the population, and matters much less than making things better for the populace (which is the whole point of governance).

I'm not a Marxist, but the whole culture war stuff is very very convenient for those who don't want to work on reducing inequality (by taxing the upper earners more). And given how much of the Democratic vote is dependent on these people, I guess it makes political sense in the short term.

Dunno while I'm replying to a random anon, but anyways.

People can not (mostly) change their sex (AFAB etc).

However, there are a vast proportion of potential gender expressions, and it should be entirely uncontroversial that people can choose from them.

And now, given that I have disagreed with both partisan (US) positions on this, I'll withdraw.

We still don't know that, as Anthropic don't split our costs by training vs inference.

Maybe someday.

Sort of. It really does depend on the company and how they're set up. Like, brands using agencies or more non DR advertising only update on marketing mix models which are much slower.

Depends on the big spenders, tbh. Internet native brands are more likely to do this, but larger more staid companies will take a lot longer.

> I think what's happening is Chines models genuinely try to benchmax on it and hence they suffer when a new version comes out.

I read somewhere on Substack that a lot of this was driven by the inclusion of more knowledge work benchmarks, where the US labs are stronger, versus coding which is where most Chinese models appear to put their energies for now.

> They are like almost like malicious compliance geniuses lol.

They're like a monkey's paw or evil genie I think.

You'll get exactly what you ask for, but lots of weirdness you didn't expect.

To be fair to the model makers, deciding what to do with an under-specified prompt is hard, particularly in an RL environment where you can't just ask the user.

For stuff I want to run for a while, I'm defaulting to telling it to ask me the questions upfront to reduce the incidence of these kinds of problems.

The secret here is that building the semantic layer is a ton of work, predominantly done by data people.

I'd say more likely the FAANG type companies, suspect big 4 are much smaller.

But the FAANG customers are the opposite of sticky, given that they all have their own models.

As per the S1, 5 companies were responsible for 25% of revenue in the 9months back from end of September.

Which feels like a carefully chosen statistic.

The US has global taxation so, YS plus the rest of the OECD would probably be stable.

If we're going to tax wealth, then we should probably tax it to just below the level of growth, so that it doesn't spiral towards a singularity.