holbrad

I can't believe more people don't point out his unbelievably bad track record of predictions. For a public figure, this should be incredibly damning.

Their recently released S-1, though I don't think they do a detailed breakdown.

I couldn't agree more around the ideological friction slowing things down, especially code review. I'd be really curious to get people's thoughts.

Personally I do code review, it's what we do at my workplace, and we regularly improve the agents' code and design though human review, even with Opus 5.5.

But looking at some of the people who are pushing AI development the furthest, I do see the trend towards it just not being that useful anymore.

Real world example:

I use T3Code as my agent orchestrator and it's consistently pretty high quality and works well + regularly gets new useful features.

And with just a handful of maintainers, they're doing anywhere from 100 to 200 PRs a day!

And there are a ton of users of their nightly, which is a rolling 5-hour release, and nothing really breaks. And they have more real-world users than almost all the projects I've worked on before.

So I just can't really in good faith say that this extremely challenging to my world view.

This is such a hilarious comment when we've literally just had Sol 6 and Opus 5.5 drop that are massive wins for efficiency. I mean, how out of touch can you be? Especially if you think power saving and efficiency come from local AI enthusiasts. What a fucking joke.

I was under the impression that GitHub Copilot was much more expensive because you're always paying API rates?

I think this is just a case of the bitter lesson that increasing compute just makes all these predictions meaningless. LLMs just keep going when everyone predicts them to fail constantly.

I've heard very little positive press around Sol 6, with a ton of people preferring Sol 5.6 instead.

I haven't used it much yet, but I have much higher hopes for Sol 6.1, as it seems to be based off of a completely different base, it's not just a tune.

Yes, all those things are fine.

People should be able to build where they want.

We can have reasonable restrictions around noise and pollution, but not on the land use itself.

It seems pretty clear that this is a much larger model than Sol 6, and you can see this in the much lower generation times. I think this is also the main explanation for the $200 plan being cut in terms of API usage.

This is because they have really aggressively priced a larger model to compete with Opus 5.5, so their margins are much worse. Consequently, the equivalent API spend on the subscription is much less.

I think the only answer to this is you're just not using agents enough, because even with the very cheap pricing, it's still easy to rack up a large bill.

That's only true if you're paying API prices, but you really, really should be using the subscriptions. Both the personal and business ones are still really good value.

Well, we're stuck between a left-wing party and a left-wing party, both of whom, despite what they say, hate economic growth and try to restrict it at every opportunity.

And that's largely because the average Brit has terrible beliefs that actively make them poorer. A shockingly high percentage of people believe that building more houses won't lower costs.

>truly air-gapped environments

I don't think people have given much thought about just how hard this would be for large AI models, that need super powerful hardware/cooling etc.

Are you going to air-gap your entire data center?

>token-hungry model

It's kind of funny how this is the exact opposite of the truth. It's one of the most token-efficient models ever.

The claims aren't bullshit. Every conceivable benchmark and test you can throw at it shows Sol being good for token efficiency.

If your benchmark shows Opus 5 winning, I really question the validity of it.

As someone from the UK, Techy, nerdy people can know, but normies don't.

Also, I'd add that gym bros definitely have heard of them, especially if they are familiar with steroids.

The GLP-1 drugs are probably the main cause of this. We have this amazing class of compounds that work really well, But are priced pretty high, at least considering lower UK wages.

So there is a big incentive to look at getting these much cheaper from China. You hear about this happening, and then you see the fact that these companies have big catalogues full of other peptides that are a crypto payment and a click away.

I think it's pretty clear that most of the discussion around AI is just generally bullshit.

I've seen some articles on some data centres that were very loud because of gas turbines, but realistically, what percentage is that? Surely, I would assume it must be a minority, and it just seems like the state failing to actually follow reasonable laws around noise levels.

The US getting better power infrastructure and more energy seems like a long-term good thing. They've underinvested in it for a long time.

Obviously everyone likes the think "this time it's different" but with LLM's and AI it certainly seems to be the case.

Having a machine that's even as smart as most human's + cheap robotics, does seem to lead to us having no jobs in near-ish future eventually. (5-15 years)

Though obviously AI progress could hit a brick wall tomorrow.

I think that's an incredibly dangerous slippery slope.

If you effectively control what is monetised, you effectively control what is said en masse. It's a distinction without a difference.

I mean it's absolutely true that China is loving the American AI backlash and is fueling it.

How purposely they are doing this is unclear, but they have the motive and means.

Let your biggest rival shoot themselves in the foot over the most important technology potentially ever, is a great strategy.

Precisely, if we expect AI/LLM progress to continue as it currently is, then data centers seem like a sure fire bet. They will just print money if AI reached that level.

If you believe AI will become increasing important, which most people in tech seem to agree on.

Having data centers in your country seems incredibly important, especially to the EU. You'd really hope to see the exact opposite of this behaviour.

I mean bribes can be good.

Sure you can build a data center near me, if your going to fund a new swimming pool and library with the taxes you pay. Also I get reduced local taxes? Even better!

Infrastructure building would be a whole lot easier, if people could directly tie it to tangible benefits.

I think it's in large part because the public is being fed Chinese propaganda about all the supposed issues. There are real issues, but nowhere near to the scale people seem so think.

Combine with the fact they have zero clue about what AI is capable of*, they think we are pouring billions into technology to write emails.

(They don't have any idea about current models, let alone an intuition about what future models will be capable of)

Ah yes, the majority of people obviously don't know about climate change!

It's far grimmer to realise that people know, they just don't think we have an acceptable answer yet.

I'm not willing to accept any climate action, if it means I can't go on holiday abroad or it makes my life meaningfully more expensive etc.

It's selfish, but we need to solve it in a way that makes people globally richer.

(This is one of the great things about renewables coming down in price)

>Substack activiely promotes and pays cash money to white nationalists and Nazis

Good. They should be an open platform, if that's what people want to pay for they should be able to.

They'd also allow the polar opposite views.

You are misreading the benchmark.

https://artificialanalysis.ai/#aa-omniscience-hallucination-...

If you look at the results 3.0 hallucinates an awful lot, when it's wrong.

It's just not wrong that often.

(And it looks like 3.1 does better on both fronts)

This isn't actually correct.

Gemini 3.0 gets a very high score because it's very often correct, but it does not have a low hallucination rate.

https://artificialanalysis.ai/#aa-omniscience-hallucination-...

It looks like 3.1 is a big improvement in this regard, it hallucinates a lot less.

I'm just genuinely really confused by this take.

It's a 100x easier to build products today than it was in the 1990's. (I don't think that's an exaggeration in the slightest)

It would be basically be impossible to build anything like Maps, Excaidraw, Chat GPT etc.

Arguably people are reaching for the tools without those interactive requirements ?

There's an awful lot of people running these tools in YOLO mode.

https://embracethered.com/blog/posts/2025/the-normalization-...