Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Agents need control flow, not more prompts

bsuh.bearblog.dev

Post reply on HN
Print view View original

291–300 of 348 posts

Re: Agents need control flow, not more prompts

#291

Post by gman2093 »

jimbokun wrote: post #218
devin wrote: post #216

Earlier quoted context omitted.

And then you realize that what you’re using the smaller models for is ALSO decomposable and part of it is just a few if statements, and then you realize that for this feature you don’t actually need or want a model because the performance, reliability, reproducibility are cheaper and better for you and your users.

So you have the model write the if statements and put itself out of a job.

Alternatively, and sometimes more cost-efficient: you can find a developer who can write bespoke if statements. There are dozens of us!

Re: Agents need control flow, not more prompts

#292

Post by gf000 »

Neywiny wrote: post #270
gf000 wrote: post #246

Earlier quoted context omitted.

I'm being deliberately pedantic, but depending on what kind of representation we use for the neural network (due to rounding) as well as the choice of inference (that is, given a distribution for next token, which one to choose), it can absolutely be reproducible and completely deterministic. Though chaotic, which I believe is the better word here - a single letter change may result in widely different results. We ju…

With determinism you're not wrong. The problem is that you'd need to make sure all your seeds, temperatures, and other input parameters are exactly the same, and importantly that all context is cleared. But people don't do that. And I'm not sure every if even any provider lets you set those parameters.

Even with temperature set to zero, I believe due to FP operations not being commutative you may still get non-determinism, so what I am talking about (as mentioned, very pedantically) is mostly the theory.

Re: Agents need control flow, not more prompts

#295

Post by morpheos137 »

It speaks to how dumbed down the human userbase has become that these kind of articles are even presented as insightful. "Agents" are not intelligent. they are pattern extrapolators. If you want a reliable deterministic output you need a deterministic harness. Think of agents as a montecarlo sampling tool. The harness defines the result over noise. it is hilarious to me the industry is going head long into more "intelligent" agents while ignoring that intelligence is an adaptation to constraints not some magical abstract general thing that just appears and can do useful work. AGI is a lie. Stocastic parrots + harness is a useful tool.

Re: Agents need control flow, not more prompts

#296

Post by whattheheckheck »

LikesPwsh wrote: post #271
krzyk wrote: post #258

Earlier quoted context omitted.

Could you elaborate what does "compiling orchestration prompt" mean?

Rather than telling the LLM "loop through these files", tell it "write a script to loop through these files", then hard-code that script somewhere.

The models will eventually be able to know that they need to do that to get the thing done from natural language

Re: Agents need control flow, not more prompts

#299

Post by juanre »

Absolutely agree. However, if you do not need absolute reliability pairs of agents are much better than single agents. These days I always have one agent coding and another code-reviewing. The code reviewer is also the holder of the lamp, keeping track of the final goal. This is applicable to whatever task you want your agents to achieve: one works, the other looks over the shoulder.
Post reply on HN