On the whole “just use loops”
Outside of the increasingly few people who
1) have unlimited AI token budgets
2) feel like prompting agents are holding them back (usually thanks to no #1)
I don’t think many have a use case for them. I’m more than content prompting (esp w #1!)
Positive users cite concrete benefits from AI agent loops like test stabilization and autoresearch, while negative users accuse promoters of self-serving hype or dislike the co-opted terminology.
If you built something specific that can be shown with loops that is useful, feel free to share.
I see experimentation happening for those who don't have budgets (either working at eg an AI lab, or just being happy to burn hundreds on an idea that is throwaway output)
On the whole “just use loops”
Outside of the increasingly few people who
1) have unlimited AI token budgets
2) feel like prompting agents are holding them back (usually thanks to no #1)
I don’t think many have a use case for them. I’m more than content prompting (esp w #1!)
@GergelyOrosz I was just trying to reverse engineer a fingerprint driver and dynamic workflows were helping - it was trying out different image formats used by other models of the same manufacturer one by one and trying to figure out the first few bytes of the captured image.
id respectfully disagree here
w adequate caching the cost is not very prohibitive
i ported pokemon ruby to WASM in about 6 hours in a continous agent loop, 85%+ cache hits, 240M tokens before cache, total cost was $125 for the entire run
now maybe $125 sounds expensive, but this task would have otherwise taken months of work from very senior devs and cost tens of thousands of dollars
loops can be cool if used correctly 😎
tell us you dont know what you're talking about without telling us lol typical wannabe non technical tech influencer 🙄
for your info, this is a full ROM binary to WASM port completed autonomously by my ai agent harness. months worth of elite level code recompiled from binary from scratch done in 6 hours
if i have to spell the ramifications of this out for you, you are not worthy of your tech following.
read the repo hot shot and do better, I bet $125 right now you cant.
https://github.com/agnt-gg/gba-pokemon-rom-to-wasm
@GergelyOrosz jeez, shut the fuck up my god do you ramble.
read the fucking repo and reproduce it yourself you retard.
https://github.com/agnt-gg/gba-pokemon-rom-to-wasm
@NathanWilbanks_ is this some pet project? I am trying to understand what you achieved doing this besides burning the $125 and generating this much revenue for the labs
what is the output? Can you link to it? Can I play it in the web now?
@GergelyOrosz or if you're too fucking retarded for that, ask grok.
@grok pls tell @GergelyOrosz why he is grossly mistaken
https://github.com/agnt-gg/gba-pokemon-rom-to-wasm
@GergelyOrosz @NathanWilbanks_ I use AI to build local-first local-only desktop-mobile regtech/fintech apps. The relay with -p is cool if you want to follow a sequential, well scoped plan but it still burns a shit ton of tokens more than if you just delegate each step manually. The loops is some.version of it
@NathanWilbanks_ @GergelyOrosz to be fair ive gotten this far in about 20 mins without any loops.
@GergelyOrosz It actually made good progress - figured out what type of reader it is (swipe or scan) and then figured out what the magic bytes in the frame were and then figured out which sequence to send to the device to start and stop capture etc
Works when you want to brute force to a goal
Here's one project linked as an example of loops.
But it feels super artificial to me. Someome "one shot agent looped" a Pokemon Ruby game to WASM to prove that you can... used to advertise their AI startup. It feels $125 spent on marketing tbh...
@GergelyOrosz Right. That's my personal experiment for this afternoon.
I want smarter prompts from agents to me (to solicit information), not more prompts from agents to agents.
@GergelyOrosz Human in the loop > loop looping the loop
@GergelyOrosz 3) you need to have something adopted at scale to even have enough new and meaningful events for a loop to be meaningful
mostly irrelevant to the vast majority of people building early stage projects
@ImLunaHey @GergelyOrosz see if it gets passed the internal battery and game based evental. also verify timing across all registers, timing issues can build up over playtime
@GergelyOrosz @NathanWilbanks_ Based on his extremely rude response I’d ignore it completely
@NathanWilbanks_ @GergelyOrosz Classy. Great advertising for your startup.
@GergelyOrosz 💯 A loop is only an advantage when the task actually benefits from iteration, validation, and stateful correction otherwise it just adds latency and token burn. Its important to know when a clean prompt is the more elegant, cheaper, and faster system.
@robertobmello @NathanWilbanks_ yes it was a mistake to engage!
@GergelyOrosz The dividing line isn't budget, it's whether the task has a cheap verifier. With a type-checker or test suite the loop converges in 2-3 iterations and pays for itself. Without one it just burns tokens guessing. Loops need a ground truth to loop against.
On the whole “just use loops”
Outside of the increasingly few people who
1) have unlimited AI token budgets
2) feel like prompting agents are holding them back (usually thanks to no #1)
I don’t think many have a use case for them. I’m more than content prompting (esp w #1!)