4 ms·

Why do so many see LLM as the inflection that makes this all possible? So much of the proposed automation can be built in totally deterministic ways for cheaper

Why do so many see LLM as the inflection that makes this all possible? So much of the proposed automation can be built in totally deterministic ways for cheaper. Not everything benefits from being forced to look like a non deterministic language problem. You might argue the biggest barrier to enterprise software innovation is the enterprise mindset vs a more engineering first mindset. Rippling comes to mind as an example of bringing engineering to enterprise.

soulofmischief 2y ago
I can go back to the NES or Atari and probably develop what would have been groundbreaking games and genres well before their time. But it took several small advancements and inflection points for those conceptual mechanics and hardware tricks to now be obvious in hindsight, and that's why no one did them earlier than they were done.
- beardedwizard 2y ago
  I agree there are a bunch of small inflections that make me confident such a solution is possible today, but llms just aren't one of them. Enterprise software is a prisoner of its own old mindset and low resource allocation.
SrslyJosh 2y ago
Yeah, this is where I stopped reading: > LLMs now offer the intelligence layer that makes it all viable LLMs are the opposite of what you want in enterprise software: unpredictable black boxes that go awry seemingly at random. I also can't see how they're at all necessary if you've got a state machine describing a process: the state machine tells you what needs to happen next.
- slowmovintarget 2y ago
  Why code carefully when you can throw gobs of hardware at the problem and push the easy button? This is the literal attitude for many in upper management. Don't build those nasty user interfaces that require design, wrap your APIs, buy your bidness peeps LLM licenses and move on. This is an extreme. Artifact Assistance will be a new style of UI pattern, but it will live right alongside the fill-out forms and CRUD back-ends.
  - jcgrillo 2y ago
    As the youths say, "fuck around and find out". I wish them the best of luck, but I'm not betting it'll work.
  - beardedwizard 2y ago
    Is this really upper management though? I see a lot of what I assume to be engineers here taking the same approach, and many projects in open source.
jcgrillo 2y ago
But don't forget the Actor Model and Durable Execution! That'll surely keep the LLM on track.
PhasmaFelis 2y ago
Reminds me of a few years ago, when a bunch of people were breathlessly claiming that blockchain was the future of videogames, because it would let you transfer items from one game to another. Take your favorite Skyrim sword and use it in Minecraft, or whatever. So I'd point out that that had been technically possible for decades, no blockchain necessary, and the reason it hadn't happened was that no developer actually wants it. And then they'd get super mad and call me names. I'm afraid the AI bros are going to do more real-world damage than the crypto bros did, though.

soulofmischief 2y ago

I can go back to the NES or Atari and probably develop what would have been groundbreaking games and genres well before their time. But it took several small advancements and inflection points for those conceptual mechanics and hardware tricks to now be obvious in hindsight, and that's why no one did them earlier than they were done.

beardedwizard 2y ago
I agree there are a bunch of small inflections that make me confident such a solution is possible today, but llms just aren't one of them. Enterprise software is a prisoner of its own old mindset and low resource allocation.

SrslyJosh 2y ago

Yeah, this is where I stopped reading: > LLMs now offer the intelligence layer that makes it all viable LLMs are the opposite of what you want in enterprise software: unpredictable black boxes that go awry seemingly at random. I also can't see how they're at all necessary if you've got a state machine describing a process: the state machine tells you what needs to happen next.

slowmovintarget 2y ago
Why code carefully when you can throw gobs of hardware at the problem and push the easy button? This is the literal attitude for many in upper management. Don't build those nasty user interfaces that require design, wrap your APIs, buy your bidness peeps LLM licenses and move on. This is an extreme. Artifact Assistance will be a new style of UI pattern, but it will live right alongside the fill-out forms and CRUD back-ends.
- jcgrillo 2y ago
  As the youths say, "fuck around and find out". I wish them the best of luck, but I'm not betting it'll work.
- beardedwizard 2y ago
  Is this really upper management though? I see a lot of what I assume to be engineers here taking the same approach, and many projects in open source.

jcgrillo 2y ago

But don't forget the Actor Model and Durable Execution! That'll surely keep the LLM on track.

PhasmaFelis 2y ago

Reminds me of a few years ago, when a bunch of people were breathlessly claiming that blockchain was the future of videogames, because it would let you transfer items from one game to another. Take your favorite Skyrim sword and use it in Minecraft, or whatever. So I'd point out that that had been technically possible for decades, no blockchain necessary, and the reason it hadn't happened was that no developer actually wants it. And then they'd get super mad and call me names. I'm afraid the AI bros are going to do more real-world damage than the crypto bros did, though.