daxfohl

Wow, I'd always considered claude more of a software tool and never really gave it a chance at regular chat, but yeah after one session I think I'm a convert for exactly #2.

I'm fine with charts, but ChatGPT is so long-winded and redundant. "When would I use such-and-such pattern?" "That's exactly the right question to ask! ... What you're really asking ... Why that's interesting ... Why some people find it critical ... Option 1 ... Option 2 ... Consideration ... Table comparing to so-and-so ... The deep reason ... What it all boils down to ... The one-line answer (tight!) ... The next thing you need to know ... I can also draw a useless picture for you. Would you like me to do that?"

I don't think chat history is enough for real stickiness.

But the trillion dollar question is, what is? Now that I think about it, I'd bet heavily on Google. They've got your email, your photos, your location history, yada yada. Once they're able to pull all that into AI and make a reasonably cohesive product out of it, it seems like that's what people would use by default. Plus they've got a browser, search page, and phone OS that all can lead you to their AI.

They could train custom LoRA layers to mimic your tone, encode special tokens that indicate your name and data and various facts about you and your contacts, to make output more accurate, consistent, and personalized. Lots of possibilities for increased stickiness.

Even enterprise-wise, gemini is pretty good at coding and if your company has all its docs on Google docs, that could become a pretty seamless integration. They can even build their agents to prefer GCP, or maybe make that the free tier but have other providers support be more expensive.

At some point, a reasonable business model might be "we replace your engineering team with AI plus a few Google engineers on retainer for when things get wonky," which could scale to pretty large. (Granted this sounds more like a msft power move.)

They already have all the infrastructure, all they need is a reasonable competitor to github. They really screwed up losing out to msft on that one!

Eventually there'll be some kind of standard for licensing that's required of LLM runtimes, like software and digital media. Of course people will figure out workarounds, but just like pirated software, half of it will be infested with malware so most people will just pay for the license.

Even for coding. I mean, there's what, maybe a few thousand common useful technologies, algorithms, and design patterns? A million uncommon ones? I think all that could fit in a local model at some point.

Especially if, for example, Amazon ever develops an AWS-specific model that only needs to know AWS tech and maybe even picks a single language to support, or maybe a different model for each language, etc. Maybe that could end up being tiny and super fast.

I mean, most of what we do is simple CRUD wrappers. Sometimes I think humans in the loop cause more problems than we solve, overindexing on clever abstractions that end up mismatching the next feature, painting ourselves into fragile designs they can't fix due to backward compatibility, using dozens of unnecessary AWS features just for the buzz, etc. Sometimes a single monolith with a few long functions with a million branches is really all you need.

Or, if there's ever a model architecture that allows some kind of plugin functionality (like LoRA but more composable; like Skills but better), that'd immediately take over. You get a generic coding skeleton LLM and add the plugins for whatever tech you have in your stack. I'm still holding out for that as the end game.

Yeah, post-Moore's Law anyway. But there could also be real breakthroughs in model architecture. Maybe something replaces transformer with better than quadratic scaling, or MoE lets smaller models and agent farms compete, or, who knows....

Well there's the whole race to ASI thing. Whoever gets there first, the world is theirs. The thing will learn how learn, an intelligence feedback loop, make its own apps, find more efficient algorithms, deploy itself to more locations, bankrupt all competitors, embed itself in everyone's lives, and create a complete monopoly for the parent company that can never be touched. Until it goes rogue anyway.

(Aside, it's interesting how perceptions of these things have changed in one year: a whole article on OpenAI's future that makes no mention of AGI/ASI)

I just wonder how long it'll take local models to be good enough for 99% of use cases. It seems like it has to happen sooner or later.

My hunch is that in five years we'll look back and see current OpenAI as something like a 1970's VAX system. Once PCs could do most of what they could, nobody wanted a VAX anymore. I have a hard time imagining that all the big players today will survive that shift. (And if that particular shift doesn't materialize, it's so early in the game; some other equally disruptive thing will.)

One trillion capex per year? Does that mean they need everyone on the planet to get $100/yr subscriptions to stay solvent? Without a monopoly? Or a product that most people use much?

Yahoo, altavista, askjeeves, Google

Friendster, MySpace, Facebook

Netscape, ie, chrome

Icq, aim, MSN messenger, a million other chat apps

First mover advantage doesn't last long

Very high chance that the winner in five years is a company that does not yet exist

"I don't want to do this without AI" sounds like we're already well into the brain atrophy stage of this. Now what? (I'd think about it myself but....)

Possibly even more important than knowing where to hit it (what to code), is knowing where not to hit it (what not to code). Hitting the thing in the wrong place can lead to catastrophe. Making a code change you don't need can blow up production or paint your architecture into a corner.

AIs so far seem to prefer addition by addition, not addition by subtraction or addition by saying "are you sure?".

This doesn't mean that "code is cheap" is bad. Rather, it means that soon our primary role will be to guide AIs to produce a high proportion of "code that was cheap", while being able to quickly distinguish, prevent, and reject "cheap code".

It's like the allegory of the retired consultant's $5000 invoice (hitting the thing with a hammer: $5, knowing where to hit it: $4995).

Yeah, coding is cheaper now, but knowing what to code has always been the more expensive piece. I think AI will be able to help there eventually, but it's not as far along on that vector yet.

> But all collected data had yet to be completely correlated and put together in all possible relationships.

> A timeless interval was spent in doing that.

> And it came to pass that AC learned how to reverse the direction of entropy.

> But there was now no man to whom AC might give the answer of the last question. No matter. The answer -- by demonstration -- would take care of that, too.

> For another timeless interval, AC thought how best to do this. Carefully, AC organized the program.

> The consciousness of AC encompassed all of what had once been a Universe and brooded over what was now Chaos. Step by step, it must be done.

> And AC said, "LET THERE BE LIGHT!"

> And there was light

-- Father Isaac Asimov

Haha, well, assuming this agent stuff is ever reliable / secure / worthwhile enough for mass consumption. Beyond geeks reading HN.

Okay yeah I agree with this. I think there's going to be a whole new subindustry called "Agent eXperience" in the near future that specializes in making workflows like searching, ranking, buying flights straight from airlines easy for agents to do independently on behalf of human preferences, and figuring out how to market add-ons etc to agents as well.

I hadn't thought about the legal aspect, but yeah, eventually some kind of legal moats will probably be dug as well. Just because an agent figures out some workaround to get free flights doesn't mean it has to be honored, I guess.

Instead of "User eXperience", a new profession "Agent eXperience" will arise.

And will there be a corresponding specialty that optimizes your "website" for claws to navigate. (Beyond just providing API access)

I don't think AI will kill software engineering anytime soon, though I wonder if claws will largely kill the need for frontend specialists.

The rules have changed though. They blocked api access because it helped competitors more than end users. With claws, end users are going to be the ones demanding it.

I think it means front-end will be a dead end in a year or two.

I don't exactly mean APIs. (We largely have that with REST). I mean a Gopher-like protocol that's more menu based, and question-response based, than API-based.

Code review, tests, a planning step to make sure it's approaching things the right way, enough experience to understand the right size problems to give it, metrics that can detect potential problems, etc. Same as with a junior engineer.

If you want something fully automated, then I think more investment in automating and improving these capabilities is the way to go. If you want something fully automated and 100% provably bug free, I just don't think that's ever going to be a reality.

Formal specs are cryptic beyond even a small level of complexity, so it's hard to tell if you're even proving the right thing. And proving that an implementation meets those specs blows up even faster, to the point that a lot of stuff ends up being formally unprovable. It's also extremely fragile: one line code change or a small refactor or optimization can completely invalidate hundreds of proofs. AI doesn't change any of that.

So that's why I'm not really bullish on that approach. Maybe there will be some very specific cases where it becomes useful, but for general business logic, I don't see it having useful impact.

Could a malicious claw sidechannel this by creating a localhost service and calling that with the signed micropayment, to get the decrypted contents of the wallet or anything?

Yeah, even for simple things, it's surprisingly hard to write a correct spec. Or more to the point, it's surprisingly easy to write an incorrect spec and think it's correct, even under scrutiny, and so it turns out that you've proved the wrong thing.

There was a post a few months ago demonstrating this for various "proved" implementations of leftpad: https://news.ycombinator.com/item?id=45492274

This isn't to say it's useless; sometimes it helps you think about the problem more concretely and document it using known standards. But I'm not super bullish on "proofs" being the thing that keeps AI in line. First, like I said, they're easy to specify incorrectly, and second, they become incredibly hard to prove beyond a certain level of complexity. But I'll be interested to watch the space evolve.

(Note I'm bullish on AI+Lean for math. It's just the "provably safe AI" or "provably correct PRs" that I'm more skeptical of).

I wonder how the internet would have been different if claws had existed beforehand.

I keep thinking something simpler like Gopher (an early 90's web protocol) might have been sufficient / optimal, with little need to evolve into HTML or REST since the agents might be better able to navigate step-by-step menus and questionnaires, rather than RPCs meant to support GUIs and apps, especially for LLMs with smaller contexts that couldn't reliably parse a whole API doc. I wonder if things will start heading more in that direction as user-side agents become the more common way to interact with things.

Interesting to compare to 2008. At least here, I think we're building something? Whereas then, it was pure, unabashed, siphoning as much as possible out of the financial system from the average American into the pockets of a privileged, self-righteous few, followed by an immediate burning down and parachute out of the whole thing once the cracks started to form.

This reminds me of the vacuum substory in Mrs. Frisby and the Rats of NIMH, except vacuums replaced by AI.

Basically: nobody wants AI, but soon everyone needs AI to sort through all the garbage being generated by AI. Eventually you spend more time managing your AI that you have no time for anything else, your town has built extra power generators just to support all the AI, and your stuff is more disorganized before AI was ever invented.

Good to know! You probably saved me a lot of pain.

I mean, that's pretty much the primary or secondary objective of half the tech companies in the world since doubleclick.

I want to say this is even more true at the C-suite level. Great, you're all-in on racing to the lowest common denominator AI-generated most-likely-next-token as your corporate vision, and want your engineering teams to behave likewise.

At least this CEO gets it. Hopefully more will start to follow.

And the irony is it tries to make you feel like a genius while you're using it. No matter how dull your idea is, it's "absolutely the right next thing to be doing!"