←back to thread

AI agents: Less capability, more reliability, please

(www.sergey.fyi)
419 points serjester | 7 comments | 31 Mar 25 14:45 UTC | HN request time: 0s | source | bottom
Show context
simonw ◴[31 Mar 25 15:08 UTC] No.43535919[source]▶
Yeah, the "book a flight" agent thing is a running joke now - it was a punchline in the Swyx keynote for the recent AI Engineer event in NYC: https://www.latent.space/p/agent

I think this piece is underestimating the difficulty involved here though. If only it was as easy as "just pick a single task and make the agent really good at that"!

The problem is that if your UI involves human beings typing or talking to you in a human language, there is an unbounded set of ways things could go wrong. You can't test against every possible variant of what they might say. Humans are bad at clearly expressing things, but even worse is the challenge of ensuring they have a concrete, accurate mental model of what the software can and cannot do.

replies(12): >>43536068 #>>43536088 #>>43536142 #>>43536257 #>>43536583 #>>43536731 #>>43537089 #>>43537591 #>>43539058 #>>43539104 #>>43539116 #>>43540011 #
emn13 ◴[31 Mar 25 15:26 UTC] No.43536142[source]▶
Perhaps the solutions(s) needs to be less focusing on output quality, and more on having a solid process for dealing with errors. Think undo, containers, git, CRDTs or whatever rather than zero tolerance for errors. That probably also means some kind of review for the irreversible bits of any process, and perhaps even process changes where possible to make common processes more reversible (which sounds like an extreme challenge in some cases).

I can't imagine we're anywhere even close to the kind of perfection required not to need something like this - if it's even possible. Humans use all kinds of review and audit processes precisely because perfection is rarely attainable, and that might be fundamental.

replies(6): >>43536235 #>>43536390 #>>43536448 #>>43536860 #>>43536868 #>>43538708 #
_bin_ ◴[31 Mar 25 16:30 UTC] No.43536868[source]▶
The biggest issue I’ve seen is “context window poisoning”, for lack of a better term. If it screws something up it’s highly prone to repeating that mistake. It then makes a bad fix that propagates two more errors, the says, “Sure! Let me address that,” repeating to poorly fix those rather than the underlying issue (say, restructuring code to mitigate.)

It is almost impossible to produce a useful result, far as I’ve seen, unless one eliminates that mistake from the context window.

replies(4): >>43537158 #>>43537500 #>>43539768 #>>43547497 #
1. instakill ◴[31 Mar 25 16:55 UTC] No.43537158[source]▶
I really really wish that LLMs had an "eject" function - as in I could click on any message in a chat, and it would basically start a new clone chat with the current chat's thread history.

There are so many times where I get to a point where the conversation is finally flowing in the way that I want and I would love to "fork" into several directions from that one specific part of the conversation.

Instead I have to rely on a prompt that requests the LLM to compress the entire conversation into a non-prose format that attempts to be as semantically lossless as possible; this sadly never works as in ten did [sic].

replies(4): >>43537724 #>>43537745 #>>43539090 #>>43547811 #
2. theblazehen ◴[31 Mar 25 17:50 UTC] No.43537724[source]▶
You can use LibreChat which allows you to fork messages: https://www.librechat.ai/docs/features/fork
3. tough ◴[31 Mar 25 17:52 UTC] No.43537745[source]▶
Google UI supports branching and delete someone recently made a blog post about how great it is
replies(1): >>43539049 #
4. marlott ◴[31 Mar 25 19:47 UTC] No.43539049[source]▶
which Google UI?
replies(1): >>43542605 #
5. mvdtnz ◴[31 Mar 25 19:51 UTC] No.43539090[source]▶
This is precisely what the poorly named Edit button does in Claude.
6. tough ◴[01 Apr 25 03:40 UTC] No.43542605{3}[source]▶
ai.dev AI studio sorry
7. genewitch ◴[01 Apr 25 15:16 UTC] No.43547811[source]▶
LM studio has a fork button on every chat part. Sorry, can't think of a better word - you can fork on any human or ai part. You can also edit, but editing isn't, it essentially creates a copy of the context with the edit, and sends the whole thing to the AI. This can overflow your context window, so it isn't recommended. Forking of course does the same thing, but it is obvious that it is doing so, whereas people are surprised to learn editing sends everything.