pu_pe

Yes, you can also see an increasing number of typos, which I suspect would have been corrected pre-AI. What's interesting is that all these signals are really easy to fake with AI as well.

If we go back only 45 days, Opus 5.5 was not even out yet. So for more than a third of the current SOTA model's existence, Google's response to it was under its own embargo. When it was finally out, paying users get 3-4 prompts a day, which means Argon will be experienced by a tiny minority of people, another unforced error on Google's side.

By the time Carbon comes out, I fully expect Anthropic, ChatGPT and/or a Chinese lab to have its answer ready to ship in less time than it took Google to approve a release.

And how do they realize their current index is broken? They either have some recency bias, or indeed favor the big American labs (more likely, since a new Chinese model topping the charts triggers a shakeup).

Income isn't salary either. I'm pretty sure he earns a lot more than that.

The mythical man-month was not only about programming itself, but everything around it. The argument was that projects in large organizations are more akin to pregnancies, with stages that cannot be sped up, rather than a bunch of tasks that can just be solved whenever, by whoever.

This article seems to be instead about whether AI models can be parallelized. And if so, yes, absolutely, in fact you should be trying to break down tasks in a way that they can be worked in parallel. Looking at the frontier, the math results from openAI are coming from a massive parallelization of agents for example, which is why they could crack a millenium problem in like four days.

This is not my experience at all. I see more people having difficulties in extracting value out of AI than the other way around. When I check what they're doing, they paint themselves in all kinds of weird corners, like keeping a mega-session full of unrelated questions, letting the AI model lead instead of managing it, misunderstanding the scope of what to ask for, etc.

Of course all the crap about prompt engineering or whatever is meaningless, and it always was. That said, this idea that if AI gets good enough you will just jump in and use it is flawed. You learn by doing. It's almost self-evident to me that someone who has been working with AI models for the past 4 years will be much better at it than someone who started this week.

Inference will continue to be profitable while we don't have enough compute in the world to serve the demand we have. Also, I don't see why it's a problem that the end game for inference of open models will be relatively low, ie 5% or so. You can have a huge industry running on these margins. Though I would note that hyperscalers have been selling compute commodities with a high margin for a long time.

DS4 Flash is so cheap also, imagine what the numbers look like for a flagship model like Opus 5.5, which charges 100x more. Of course inference for that model also costs more, but I doubt the proportion is the same.

This is a really neat, approachable way to explain the concept of attention in plain words. Thanks for giving us the reference, though in my understanding the formulation in the first paragraph comes from Natalie Wolchover, the columnist, rather than Bahdanau.

Your premise about predictability is wrong. First of all, humans are also unpredictable. We just had a case where two pilots started a brawl in the cabin and almost took the airplane down with them. Second, and most importantly, the code generated by LLMs is just as deterministic as the ones by humans. So if you think we can reasonably test code made by humans, then we can also test code made by AI, regardless of whether its writer is deterministic in nature or not (ie, "predictable").

Human error rates have already been beaten, even in precisely the domains we are discussing here. Just look at safety data from Waymo for example [1].

[1] https://waymo.com/safety/impact/

I don't understand your argument here. All these scenarios you mentioned are testable. If the rate of accidents of autopilot + AI is lower than for human pilots, yes I would fly with them. Same with self driving cars, which have already achieved that.

AI is not error-free and it doesn't have to be. It has to have the same or lower error rate as humans.

> Software teams are capturing a lot of the speedup in the form of more leisure or quasi-leisure time. (To greatly simplify: software teams are offering 2x speedups to their companies. They're then using tools that allow them to work 10x as efficiently, and taking nice long coffee breaks.)

This might be the biggest factor in my opinion. Corporate incentives usually don't reward someone delivering 2x-10x, so an employee who suddenly found a way to multiply their productivity might be keeping that method to themselves and freeing up some time instead.

> In many Western countries, housing is the best performing asset class so it's reached the point of stealing investment dollars that might otherwise go to productive output. Housing is the black hole at the heart of the economy.

This is true, and it is a huge problem in Europe for example, where buying a property to rent is often the best investment for regular people. So instead of opening a business or taking any other risk that may or may not pay off, investment money is going to an asset class that is not productive.

> The findings backed up what anecdotes and surveys have increasingly found in recent years: An overreliance on AI for answers erodes our cognitive fitness.

What is our cognitive fitness without our tools? How capable would we be at intellectual tasks if you remove computers, search engines and other things we now take for granted?

It's clear to me our overall output is better with those tools, and better with AI. So the comparison the paper is trying to make is disingenuous, because it tries to compare a baseline without any modern tools (ie, solving fractions by hand) to a result where the best tools are suddenly removed. But these tools are here to stay, and my cognitive fitness has adapted to their continued existence.

How well would these people doing fractions by hand perform if math notation could not be used? Or if the most common method for solving fractions on paper was suddenly disallowed?

This should have a material impact on SpaceX stock. Wouldn't it be much easier to harden datacenters against attack under ground than in space?

This might be the pets.com moment of this era.

So just when the whole Greenland debacle seemed to be settled, it's time to create another rift with the EU.

Yeah, exactly. British power came from technological advancements that occurred before the Victorian era: steam engine, spinning jenny, railways, electricity, etc. By the time the elite depicted in this article was running the country, they were largely inheriting that technological lead, and arguably had already lost it by the turn of the 20th century.

What's stopping you from using the subscription plan and then topping up with API if you need?

The author did not make this model though, they used Claude to write a blog post about it.

A blog post is not a Show HN topic.

This is an ad for the new startup by the author, who will now focus on reasoning models. His insight seems to be that LLMs should make their assumptions explicit first, then map out a search space and provide reasons for why they should take one path or the other.

I think it's an interesting approach but the overall discussion about reasoning is really pedantic. What the author is describing here is one approach out of many, and in my opinion it doesn't cover what humans colloquially think of when they hear reasoning (while the output from a chain of thought sometimes does).

Everyone started using the term after that article came out, it feels like...

I was listening to his interview with Ezra Klein and I think his arguments are very coherent. If you are interested in actually knowing what he thinks about it, you could check the transcript [1].

[1] https://www.nytimes.com/2026/09/29/opinion/ezra-klein-podcas...

There are far more moving pieces in deploying an application than coding. My impression is TFA is arguing that replacing humans with LLMs for coding would make other things like quality assurance more difficult, in which case I disagree.

I agree, I think this false dichotomy between using LLMs and caring about quality/reliability needs to stop. All these mission critical industries listed in the article rely on extensive testing for quality assurance, with human code review being a layer on top of all that, but far from the most critical one.

Interpretability is the same, our abilities to do that have increased rather than decreased. I think a codebase generated by AI is actually more understandable than one generated by humans at this point, and you can ask clarifying questions whenever you get stuck.

TFA's points only make sense if the mental model the author has in mind is someone who writes a prompt then immediately puts an app into production without any thought behind it.

It is a bicycle for the mind, but the mind still sets the direction. Sometimes it's going faster towards a dumpster.

AI moves too fast for traditional academic pace. During peer review alone, another generation of models comes along and might obsolete your findings. Though, in this case, I am not sure that Github Copilot specifically has improved that much in the meantime.

Surely AI can be dangerous before the point it becomes completely perfect. Very imperfect people manage to do a ton of damage already.

We had people departing from OpenAI, Anthropic and now Google. The pace the frontier open letter was signed by over 1,000 experts. At which point do we believe the message?