He holds the record for the largest amount of money lost (in dot com), looks like he's aiming higher this time.

> I would think an RSI could count the number of R's in strawberry

Can you count the circles in that sentence?

Like, I think that RSI et al is way over hyped (where's the data gonna come from?) but the number of R's question is very silly given how tokenisation works.

It is very silly to expected RSI to be able yo count the number of R's in strawberry?

Maybe its not very useful then.

Dungeon Crawler Carl is marvelous, though. Threshold is also really good. Most of its terrible, but it's within error bounds for a lot of book quality.

And her husband is a hedge fund guy. Like, the prior here should be that she took advantage of his knowledge, rather than vice versa.

Those devs were clearly a lot smarter than me, then.

They do return their services if you search for the name of some AWS services, which is helpful.

Did opus have an oracle for the original compiler? If not, then it's expected that qwen would do a better job.

As is known, LLMs are amazing when verification is cheap, and much less useful when it's expensive.

> Did opus have an oracle for the original compiler?

It actually did. There was an older yacc-based implementation that had a lot of architectural limitations. We decided that it was better to start over with a better architecture, but we did have that older one around to compare generated microcode initially.

Ok, that's interesting. My prior would have been that if Opus didn't have an oracle, then there's still a pretty large gap.

But if, as you say, both had an oracle then this is a really interesting outcome.

To be fair, I think the differences between frontier and open models is pretty small (I've been swapping back between Claude and GLM over the last week).

The one thing that I notice is that GLM keeps trying to blame external factors for any non-matches (we're rebuilding an external tool for internal use), whereas Claude adheres to my prompt much better (i tell it to only blame these factors if it's ruled out everything else).

I'm not convinced that change is worth 3.5-5x cost differentials though (I only used Claude because we have the small business package, which is way cheaper than API rates).

Agreed. Like lots of this AI speculation appears to require the belief that AI will figure out how to build supervised training environments without more data.

I think this is very unlikely, but world changing if true.

Additionally, what does super intelligence even mean and why would we expect it to be obtainable from the data that we have?

I'd be delighted if someone can provide some definitions for these terms.

> For highly paid employees (high impact, high competence, whatever the reason), reducing inequality goes against their direct financial interests. Inequality gets reduced by dragging their compensation down, not increasing everyone else's.

Speaking as someone who's been a super high performer in some roles, and a terrible performer in others, I definitely want to reduce inequality.

Like, it's entirely unfair that I (because I negotiate well) get better options than others. If you want to pay someone outside of band, give them a different title.

If someone performs unexpectedly well, give them a bonus/promotion etc.

It's deeply sad (but unsurprising) that so many people can only focus on their particular benefit (especially when the vast majority of that benefit is context dependent).

Like, if you hired John Carmack or Fabrice Bellard and just told them to take tickets, they'd probably be an average performer at best. If you told them never to take initiative, they'd perform badly. So lots of performance is dependent on context, there are very few "good" employees in a vacuum.

Wanting to live in a better society should not preclude me from doing better in this one.

And the transaction costs on such an approach would be too large, as I'm sure that you know.

M I note that you don't address the notion of context dependent high performance.

They call them revolutions because they come round full circle.

GDPR mostly. In general, the easiest way to get ZDR for the models was through the hyperscalers. Additionally they're a German company so it's probably about all the data and tools being hosted in the EU, which Anthropic direct don't provide.

Why does dev work need to comply with GDPR? Are devs handling real customer data?

I mean, they shouldn't be, but lots of orgs, particularly in Germany are very very sceptical of any kind of data retention.