brap

I’m sure they can hire a whole team of world-class UX folks and it will pay for itself

Anyone surprised by this is incredibly naive.

By all means use whatever works for you, I’m not even going to try to make an argument on ethics (and honestly I’m not even sure where I stand, given the behavior of American AI companies).

But I just cringe every time I see people acting like any of this is done in good faith.

Open source coming out of China is a state-sponsored criminal enterprise, built only for the benefit of the Chinese regime, one of the worst to exist in human history.

It’s crazy how much these companies invest in their models but when it comes to UX they do fuckall

And you an do it without even changing the model!

I think if you showed these before/after menus to people from 5 years ago, most would tell you it’s a massive improvement.

But now many of us have been trained to detect slop from miles away so it just comes across as cheap.

I often feel the same way when I glance at a project’s documentation and it’s long and full of markdown styles and emojis: “the author didn’t think about the users at all, they just wanted to get this over with quickly, so now I have to be super suspicious of everything I read here and it might end up being a huge waste of my time”. A few years ago I’d probably think “wow, emojis!”

From my experience, this thing is crazy fast.

Spawn 10 on the same problem and have them debate to reach a consensus, you’ll get Fable-like results but 100x faster.

I think many of us engineers (including myself) are suffering from AI-induced procrastination. Fixing that pesky bug in production isn't fun. Building a harness/workflow that detects and fixes pesky bugs in production on its own is way more fun (especially when your employer pays for tokens), and it feels just as important, if not more! So what if it didn't actually fix that pesky bug in production on its own...

The Chinese ecosystem is largely built on stealing intellectual property from and/or distilling of US closed source models.

I’m not getting into the ethics/comedy aspect of it, but let’s not pretend this isn’t the case. Plenty of evidence, the motive is obvious, and the numbers are in plain sight (valuations, salaries, etc).

As long as this continues I think closed source will continue being a few steps ahead, but the steps will probably get smaller over time.

Having said that, this is already priced in, the market predicts this gap will be large enough for the US companies to profit from (astronomically).

Because it's mostly a load of crap.

Most of these reusable, automated workflows sound really cool in theory, but in practice either they require a lot of customization (not really reusable) or a lot of babysitting (not really automated).

AI has improved my productivity significantly, I'm not a hater, but we also have to be honest about its (current) limitations. The whole "software factory" thing is just a pipe dream people sell to execs / investors / online courses.

I bet that, all things considered, training 10 new Chinese frontier models costs _significantly_ less than a single modern fighter jet.

I'd expect them to shift to a model where Chinese models are hosted on Chinese infra.

You assume these Chinese companies need to make a profit. Nothing is really private in China, nothing that truly grants power, anyway.

CCP can just write off these loses as defense budget, which they basically are.

I bet that, all things considered, training 10 new Chinese frontier models costs _significantly_ less than a single modern fighter jet.

How dare they not buy things

How dare they buy things

This is very interesting. I never thought of myself as a “consumer” in that sense, but I guess it is 100% correct, I’ve always been a consumer and I have in fact taken things for granted.

I don’t see that changing, to be honest, but it’s interesting that this never even occurred to me. It seems so obvious in hindsight. Quite the blindspot.

I can totally see myself asking questions on SO even these days, but there’s a good chance they’ll lock my post so why bother.

Something about the SO incentive system created the most hostile platform imaginable.

won't stop until it achieves or exhausts your prompt

This is overselling it. In all implementations of this that I’ve seen, a more correct phrasing is “won’t stop until it pinky-promises it achieved your goal”.

It usually relies on something like a hook that refuses to let the agent stop unless it includes a “promise sentinel”, which is basically the model having to include “I swear I’m done” in the response (usually as a Markdown comment so the user doesn’t see it). If that sentinel isn’t in the response, the hook rejects the completion and hands it back to the agent (sometimes it also restates the goal to keep it focused). It’s useful for sure, but it fails many times for obvious reasons.

More robust implementations use the hook to spawn another agent that verifies the goal is achieved, but that too isn’t prefect and sometimes performs worse.

Of course if you have a problem that can be deterministically verified, you could have the hook run this verification instead, but this is usually not the case (and even when it is, agents will often cheat the verification e.g. by deleting a test file).

We keep seeing this. If you had to point out the fundamental problem, what would it be?

I think it’s the disconnect. Each persona is an expert in their own field but is completely oblivious to other critical areas.

The founder knows how to raise money but doesn’t really understand the customers. The engineer knows the tech but doesn’t really understand what it takes to keep the business afloat. The salesperson knows what customers want but doesn’t really understand what’s possible to make. The investor knows the numbers but doesn’t really understand how poorly the business is run.

I suspect if you look at successful startups you’ll often see a very small (1-3) group of founders who are very close, each can do more than one thing really well, and their combined expertise means that together they have very few blindspots.

Just today I was thinking why isn’t there an optional timeout parameter for this tool

I sometimes wonder if at some point we’ll discover that the fundamental laws of the universe are simply far beyond anything we can even remotely comprehend, like explaining a microprocessor to bacteria.

But then what if we end up with ASI that can comprehend and make it useful for us?

The thing I struggle with wrt to taste is that LLMs just don’t get it.

Even if I write down every single thing it did wrong and how I’d do it, and even if I turn those into rules, it will know how to follow these specific rules, but for some reason it can’t seem to generalize beyond that. And the real list of rules seem truly infinite.

Agree, and I also don’t see why government should be involved in book rental. Why books and not say, car rental? What about jetski rental?

(If the argument is that subsidizing books helps the poor, I’m all for it, a nonprofit or a charity would be a much better framework)

This is the public sector M.O, instead of admitting something is obsolete they grab more scope and funding.

My local post office now sells iPhones. And why shouldn’t they? Nobody stopped them when they just sold SIM cards, and then cases and chargers. It’s like a law of nature.

I thought we’re over this collective delusion called MCP

OP after completing this essay https://i.kym-cdn.com/photos/images/original/001/240/075/90f...

So: after AI takes everything, what remains is not some second-best refuge — it is the place where the sense of value was always meant to live. AI is a receding tide. It washes away all the external anchors we carelessly threw out over the years — title, output, the feeling of being needed — and forces us to swim back to the one center that the tide cannot reach.

Yeah for sure and that’s great and all but I think what they were really asking is how they’ll feed their kids

This makes a lot of sense, thank you!

Of course, you have to define what "has the same behavior as" means

And that's really my issue, for example when you define "has no trailing whitespace", you are basically writing a piece of the implementation. Cover all behaviors, and you have basically re-implemented the function, no?

In other words, if I have the full formal spec of f(), isn't that the same thing as having f()?

Whenever I read about formal specs it always seems to me like “write the same tests just in a different way”, or worse, “write the same implementation but in a different way”.

I guess doing things twice can help catch errors, but I fail to see what’s so special about formal specs if they can suffer from the exact same bugs as the tests/implementation.

I guess the root of the problem is if you want to formally prove that a program does something, you have to be very specific (heh) about what that something is, at which point you are basically just writing tests/implementation all over again.

I have been looking into this on and off for years now, and I keep feeling like I’m missing something, but I don’t know what it is.

Can anyone enlighten me?

I gave it a task of scanning 6 markdown files and finding issues in the prose (contradictions etc). It ran for over 2h, exhausted my max plan session limit and crashed. I did not get any issues back.

About 2 years ago, my then girlfriend, who is a medical doctor (read: likely not an idiot), had ChatGPT plan a trip for us in a foreign country (naturally I had to nip it in the bud). And this was 2 years ago, when accuracy was way worse.

You would be surprised what normies are willing to delegate to AI.

It always seemed very natural to me that AI will move “down the stack”, where Open AI and Anthropic don’t really have a foot in the door.

Who makes consumer devices? Google

Who makes operating systems? Google

Who makes browsers? Google

Who makes the world’s most popular websites? Google

By the time 90% of average internet users get to chatgpt.com or whatever, they already went through several Google chokepoints, each layer is one more place Google can answer their questions.

And that’s not even getting into the chips, the data centers, the data, the talent, the consumer apps, the enterprise apps, the cloud platform, the brand, and of course the biggest cash printing machine in human history.

You would honestly have to be insane to bet against G.