I don’t think the good programmer bad programmer framing is helpful. I think AI fits the working style of some developers better than others. Where it fits I’ve seen some developers become so productive that they suck all the air out of the room and drown everyone else in review - not clear if this is good or bad. Where it doesn’t fit I’ve seen developers shut down because they find the workflow so depressing and their former productivity craters. Similarly I’ve seen perfectly good developers turn into slop generation machines and go largely unchecked because everyone else gets alarm fatigue from reviewing their PRs.

Yeah the review fatigue thing is a huge problem at the moment. When your work becomes 90% reviewing AI code, it is very demotivating.

I have been focusing a lot more on testing and "manual" validation than on reviewing individual lines of code. (And on setting up rules for what the lines of code should look like that are easily checked with deterministic tools or by llms.)

For large diffs, I’ll have the human driver walk me through the code in a call and if they understand it well enough to explain it to me, and explain the implications of the changes, that’s good enough for me; if they can’t answer how something works, I’ll respectfully turn down their contribution.

This cost shifting is how I stay sane and avoid burn out at work as well as stay apprised of major behavioral changes in the software we run.

I also think it’s important to emphasize that in an organization, individual productivity is secondary to group productivity, and group productivity is different than the sum of its parts.

We solved this at work by creating a system that has an LLM read a Jira ticket assigned to Mike, create an implementation by itself and post a PR using a service account. Then Mike can come along and review it, thereby satisfying peer review as it was authored by a different user (in our SCM) than the one that reviewed it.

Needless to say it is getting really hard to care about my job anymore

> Ten years ago, the kind of person who implemented a new compiler because their compiler was slow was the kind of person who started programming a new 3d game engine when they wanted to make an indie game, i.e., the kind of person who never ships anything.

Jonathan Blow is a glaring exception to this idea. In order to make his latest game, which has shipped and is great, he wrote not only his own engine but his own programming language and compiler to build it in. Zero use of AI.

I think his stuff is cool and i enjoyed the new game, but it is pretty striking that as a consequence of writing his own engine/language, it has taken him a decade of time and many millions of dollars burned in order to ship that game. Hopefully he recoups it all, but what about all the less popular devs who will spend a decade and ship a game that doesn't turn out to be as commercially successful?

He was able to do it because he had millions of dollars from selling previous games he had written in his own game engines (one of them 3D). If other devs want to try this approach and don’t succeed then so be it, but they’re free to use whatever approach they want. I was simply pointing out that the stereotype presented in the essay has exceptions.

> as a consequence of writing his own engine/language, it has taken him a decade of time and many millions of dollars burned in order to ship that game.

…But he's also shipped the engine and the language, yes?

I think he’s the exception that proves the rule. A truly rare, exceptional programmer who not only has the:

- technical skills to build an engine, language, and game

- interpersonal + communication skills to build, manage, and lead a team

- business acumen to take the above 2 skills and create profitable businesses (plural emphasized)

Not trying to lay the idolization on too thick, but an average developer having any one of the above would likely be a multiplier in their career.

When I think of games developed by a single developer I always have to think of Chris Sawyer, who initially created two of my favorite games, Rollercoaster Tycoon and Transport Tycoon Deluxe, with both being written almost entirely in assembly.

Which is a great pastime but not a feasible way to actually make games.

But he literally made a real game and released it. It was a long process for sure, but you can’t say it’s infeasible.

He had previously built and sold two games, each of which made him millions, funding the next effort. Not infeasible, but building on repeat success.

I don’t see why this weakens the example. He has repeatedly built and released games from his own engines and now language/compiler. Money was increasingly required but he had made that money through the same process.

I don't think you can say that making wildly successful games one after the other is "feasible" for anyone who's interested in making games.

Impossible ≠ unfeasible.

It's extremely unfeasible.

I've been thinking about building a GPGPU language, but it's definitely going to take a while, even with AI. AI, unfortunately(??), cannot replace taste at the moment, and generally cannot unless it becomes a lot more disagreeable.

Taste is disagreeable.

Maybe Claude should not let me build a two-level type theory with just a "You're absolutely right!". In fact, now it will say "You're absolutely right, and it would be so amazing to use cubical types for blah blah", and adamantly defend it with several paragraphs of vomit.

Why would you want a new language if you're using AI? Training set size more than compensates for token usage, so unless you plan on writing in this language by hand I don't really see a point.

I find that frontier AI is very good at understanding new syntax/semantics, for whatever reason. But that's moreso "why is it not just a huge waste immediately".

I want a new language because current GPU (and really, parallel programming) languages are just not it. Also, I think we'd be better off if it weren't just CUDA dominating the developer experience.

Moreover, I feel that we should be embracing advanced type systems with AI. The past couple decades have shown that people do want advanced typing even without AI. And now we have stuff like "Lean is faster than Rust" (https://kim-em.github.io/blog/2026-7-24-why-lean-is-faster-t...).

Programming is hard. Programming when you can't just rely on the classic RAM model is harder. Parallel programming at scale on vector processors with small amounts of cache (and slow cache at that) is even harder. We can do better. Rust has already shown this for normal systems programming.

So complile lean to GPU, maybe?

That may be a not-terrible idea (possibly even great idea), though it's not quite the base language I'm looking for. Maybe I need to just experience it more.

Also Ken Silverman, Tim Sweeney, John Carmack, etc. This is easier to do with small-scale indie games or earlier generation games

Only took him 10 years.

Sounds like he shipped the game.

It might be the case that the models internally predict the nationality of the user given other indicators, e.g. tells in the way they write.

Just because AI development moves fast and breaks things continually it doesn’t mean there’s no value in studying older models. Unless you think we understand all the behaviors of those older models.

I think if you're going to claim that Chinese AI is making things less secure for people out of China you need to back that up with models that are actually in use. Is this a real problem today, or a problem with a specific model and it's training? It's pretty freaking relevant if you ask me.

You are taking too narrow of a view. I agree that if you want to say something about the behavior of models in use today then you need to use those models. But that’s not the only reason to study questions like this.

I agree, they should have studied a range of open weight models, including non-Chinese models.

As others have said it's mostly just consistency of practice. There's a wealth of resources available freely online for these topics and you learn by doing. I would caution setting goals/expectations based on what others have achieved in youtube videos though - this guy was/is a computer science undergrad at Cambridge, so he certainly has a high capacity for learning hard stuff and has been exposed to some of the best schooling on offer.

Good job that only honest people post things on the internet otherwise we’d have a problem

falsifying statistics alone without the added stealing on top, was 3 years of labor camp in places. that was after they course corrected and stopped giving capital punishment for it.

good thing nobody falsified statistic just for the sake of it -- so being a sabotager and a foreign spy was always part of it too.

I don’t understand why this article makes you ask that, can you elaborate? The simple answer to why we didn’t have LLMs in 2005 is that nobody had the requisite ideas that make them work effectively (independent of any agentic harnesses and Schmidhuber notwithstanding) and there wasn’t sufficient clarity in the direction to scale for anyone to build anything at scale (also a requisite for them working well).

We had genetic programming though, and with the right harness you could accomplish impressive results: https://dl.acm.org/doi/abs/10.1145/2771783.2771796

You could view LLMs as a super-charged mutation function inside a genetic programming harness.

>You could view LLMs as a super-charged mutation function inside a genetic programming harness.

There's probably something there, though im not sure what. I'm thinking about LLMs token-filling and the 4 amino bases

Surely the only problem isn’t the knowledge of how to train and the confidence it would work. Wouldn’t it also have required a 1 or 2 order of magnitude increase in computing power over 2005-era supercomputers to be able to train a modern LLM in a reasonable time (“reasonable” meaning “finish training before GPT-3 comes out in 15 years”)?

I think it’s plausible that attention could have been invented earlier and GPT demonstrated at smaller scale on earlier hardware. Then hardware timelines might have accelerated (or alternative approaches explored) and we’d have gotten to where we are earlier.

NVIDIA moved as fast as possible back then, there were huge performance gains with each new GPU generation, hard to imagine they might have gone even faster.

And also "the whole Internet" (including scanned books etc.) on which to train was much smaller.

On the other hand, Google was openly collecting data with a view towards building AI.

Excellent, thanks for the paper suggestion, I’ve been needing another unnecessary luxury expense in my life!

I wonder how much bias there is in the science here. I certainly can think better with pen and paper versus computer but I’m 45 and my whole schooling through to postgrad was done with pen, paper and blackboard. It seems very plausible that digital natives from childhood will think just as well at a keyboard (maybe better since they’ll be less practiced with pen and paper).

Are you suggesting this is isn’t currently true in the conventional use of the word?

Yes, currently means "right now" not "for the foreseeable future".

Right. The person you responded to made a statement entirely about “right now” and used the word “currently”. You added in the discussion about the future, they said nothing about that.

People are saying "AI took our jobs" or are concerned it imminently will. A lot of white collar tasks are doable by AI today.