ControlsFreak
No bio...
User ID: 1422
ControlsFreak
1d ago
ControlsFreak
3d ago
The topic of the use of LLMs in various situations is always super complicated. Sometimes it's magic; sometimes it's bollocks and all that.1 I don't yet have fully-consistent rules for myself, as things are always changing as well. Terrance Tao just listed a possible set of criteria to consider:
I believe that the creation of visualization apps to illustrate mathematical or scientific concepts is a particularly favorable use case for modern coding agents, as many of the downside risks attached to other LLM use cases are limited:
- Not mission-critical. As such apps are not authorative sources of truth and only used for secondary purposes, a small positive error rate in the output can be acceptable.
- Stand-alone. As the applets are not destined to be incorporated into a larger codebase or literature, the technical debt incurred by delegating all the coding to an LLM agent is bounded.
- End product is deterministic (and sandboxed). As the applets run on a deterministic language (Javascript), are sandboxed against file or internet access, and do not make any LLM calls at run-time, security and privacy concerns are minimal, and the applet can be maintained without continued premium LLM access or resource-intensive compute.
- Not replacing primary skills. While deskilling is the tradeoff one accepts when relying on these tools to accelerate output, I am perfectly willing to forego the opportunity to keep my Javascript skills at a high level, as this is a tertiary skill for me at best in my chosen profession. (I continue to manually program in Lean and in Python to keep in practice with programming in general.)
- Not competing with humans. To my knowledge, there is no existing human effort that is being duplicated by these applets (the activity in this direction appears to have peaked two decades ago).
I would however caution against unrestricted LLM use when one or more of the above five favorable situations is not in effect.
I don't know if these are the right set of rules (#5 in particular seems insufficiently justified). But I guess the culture war is all about bickering over lists like this (or, I guess, whether any list like this should be used vs. just letting an LLM do literally everything for you everywhere).
1 - For example, in writing this, I noticed that I always screw up the markdown for numbered lists inside of blockquotes. I typically still just leave blank lines between lines/paragraphs in blockquotes; no particular reason why I do it, but it's most noticeable when it breaks numbered lists. An LLM trivially told me a couple different ways I could do it that will display properly. On the other hand, I spent a decent amount of time yesterday trying to troubleshoot something that was crashing intermittently. It gave me some certainly reasonable troubleshooting steps, but when the basic, good ones ran out, it sort of went insane. It never occurred to the LLM to note that there were some additional debug tools I could enable; I happened to find that in a web search leading to a forum leading to documentation shortly after giving up on the LLM.
ControlsFreak
3d ago
Unsurprisingly, given that Scott is the Rightful Caliph, I agree that accusations of stochastic terrorism are usually bunk. Moreover, he properly identifies the true fault line:
This liberal solution isn’t trivial. It requires the separate liberal norm of always being against extra-state violence - a norm which is currently less than entirely secure. 39% of young people have a favorable opinion of Luigi Mangione, and during the George Floyd protests several mainstream newspapers flirted with condoning violence in the name of racial justice. If your worldview says that it’s acceptable to lynch sufficiently bad people, then yes, accusing people of being bad is equivalent to calling for their murder, and you have no alternative but to make sure nobody is allowed to criticize anyone you like. This puts you in the position that Winston Churchill called “riding a tiger from which you dare not dismount”; you had better invest all your energy into making extremely sure that you and your friends are the ones calling the shots about who can and can’t be criticized. It sounds exhausting, which is why the liberal solution - bilateral controlled tiger-dismounting - is the choice of most functional societies.
I guess the major question is whether we just need a separate keyword to describe "not sufficiently against extra-state violence in a coherent and consistent way". Follow-on questions that I care less about would be whether people who satisfy this separate keyword are necessarily open to the charge of stochastic terrorism on the terms of their own position on that matter.
ControlsFreak
4d ago
It's been a while, but back in the day, Logical Increments was often pointed to as a pretty good guide for approximately how to get in at various price points. It looks like it's still being updated. Not sure if the specific recommendations are now dominated by ads or anything, but even if so, it helps give you a sense of what is a reasonable expectation at different price points. If you're interested in gaming, they give some FPS performance at various resolutions in some games. I have no idea if the games they're referring to are meaningful today. I think they get their performance numbers from other sources. You can use it as a jumping off point for any other specific tradeoffs you prefer. When I last did it, I used them to get myself into the ballpark of about what I wanted and then used PC Part Picker to narrow in on specific components and do a bit more optimization on current prices.
As far as predicting the future, I can't do that. But just briefly looking, it seems like capable enough gaming PCs are currently not obscenely unreasonable. Probably still on the "somewhat overpriced" side, but honestly not as insane as one might think.
Ok, that is by far the more reasonable option, even if I find it less natural as a matter of language.
It is, of course, difficult to strictly evaluate the extent to which a third party is "increasing" in a sense of reliance/confidence/trust. That said, I would note that a couple days ago I referred to this recent blog post by Tao, where he presented some criteria that he says he uses to guide his usage. I don't necessarily agree with them all, but I do get the sense that he maintains a distrust of them, even while trying to get as much value out of them as possible. In the comments on that blog post, he re-endorses his comments on using "AI on the red team" from this paper, which also reads to me as having a sense of distrust and trying to describe ways of working with the thing that he has a sense of distrust in, while leveraging its strengths.
More options
Context Copy link