garrinm

5 ms·

1.
by garrinm 11d ago
Originally I had those parts written in math with probability functions and the likes (its closer to my background). Then I remembered who is my target audience... but now that I see exactly who is my target audience I'm thinking I&#x2
2.
by garrinm 11d ago
To be more specific there’s no ground truth tokens to predict. There a verifiable answer in RLVR. But the tokens are explored. Not predicted as there’s no true token to predict.
3.
by garrinm 12d ago
I try to make 3 claims in the post, it was a bit clumsy I'll admit that. 1. At inference time, LLMs emit one token at a time given the prior tokens. This looks like prediction and I concede that. 2. During pre-training, LLMs predict th
4.
by garrinm 12d ago
Yes I understand the analogy was a bit loose. I'm comparing what happens at "inference time" in chess engines to what happens at train time in LLMs. In hindsight AlphaGo Zero was the perfect analogy, but I missed that opportu
5.
by garrinm 12d ago
In the article I made 3 claims, and I agree it was a bit clumsy. 1st I say that "working forwards" in the sense of outputting one token at a time could be some form of prediction, I don't argue against that. This is what LLMs
6.
by garrinm 12d ago
I think the point is more that in RL there's no ground truth to predict. So when training a model with RL the idea of "predicting" doesn't fit anymore. I'll make some edits I see that I wasn't very clear.
7.
by garrinm 12d ago
It does in pre training, but not in RL post training. And not at inference time. Reading over all these comments I get the feeling my mistake was not clearly delineating inference time and train time.
8.
by garrinm 12d ago
I think that’s fair, I didn’t actually run the whole thing through an AI. it was more targeted edits, but each time it does erode at my writing. But at the same time, I don’t think it’s a good reason to dismiss this. Because I did spend sev
9.
by garrinm 12d ago
The distinction I perhaps didn’t make clearly enough is that I’m not really debating the concept of prediction at inference time, although, as I pointed out elsewhere, I think that’s the less interesting interpretation of what “prediction”
10.
by garrinm 12d ago
It was written by a human. There are AI edits but it’s very much a human composition. Perhaps a bit sloppy.
11.
by garrinm 12d ago
Yes, I think that’s a good explanation. There are really two sides to it. There’s the mechanical, inference time, autoregressive, one-token-after-another side, which I’m not going to argue isn’t prediction. I just think that’s a relatively
12.
“Next-token predictor” is the wrong mental model for LLMs(gmcgoldr.github.io)
164 points by garrinm 12d ago|320 comments
13.
Why Does Rust Need Send and Sync?(gmcgoldr.github.io)
3 points by garrinm 13d ago|0 comments
14.
Understanding Rust's Pin Type(gmcgoldr.github.io)
1 points by garrinm 13d ago|0 comments
15.
Implementing Rust's Pin from Scratch, More or Less(gmcgoldr.github.io)
4 points by garrinm 19d ago|0 comments
16.
Pinning Down Rust's Pin(gmcgoldr.github.io)
1 points by garrinm 20d ago|0 comments
17.
Ergonomic errors in Rust: write fast, debug with ease, handle precisely(gmcgoldr.github.io)
32 points by garrinm 1y ago|27 comments
18.
by garrinm 1y ago
Thanks for the insight, I wasn't aware of `track_caller`. I'll definitely be looking into this. I was scratching my head trying to figure out how to make file and line number usage consistent and customizable, this looks like the
19.
by garrinm 1y ago
Anyhow still makes things easier for application development. The main drawback is that the resulting error type doesn't implement std::error::Error, so it's not suitable for library development (as pointed out in the anyhow docum
20.
by garrinm 1y ago
I played around a bit with SNAFU a couple of years ago, but I'm haven't worked deeply with the library so there might well be some features I'm not aware of. I think SNAFU is more like a combination of anyhow and thiserror in
21.
Show HN: Stack Error – ergonomic error handling for Rust(github.com)
27 points by garrinm 1y ago|13 comments
22.
Closing the Loop: Real-Time LLM Self-Optimization with RAG(crossingminds.com)
1 points by garrinm 2y ago|0 comments
23.
Stack Error – Pragmatic error handling for Rust(github.com)
2 points by garrinm 2y ago|1 comments
24.
by garrinm 2y ago
Stack Error is a pragmatic error handling library for Rust that provides helpful messages for debugging, and structured data for runtime error handling. Features: - Informative error messages: stack error messages and optionally add file&#x
25.
Show HN: Stack Error – Simple and informative errors for Rust(github.com)
2 points by garrinm 2y ago|0 comments
26.
by garrinm 2y ago
This gets discussed in the Roc community. They are exploring designing a language without higher kinded polymorphism. Here's a snippet from the Roc FAQ. > It's impossible for a programming language to be neutral on this. If the
27.
by garrinm 3y ago
Yes, Clint is a proof-of-concept, meant to showcase this use case of LLMs more than anything else.
28.
by garrinm 3y ago
Unfortunately you can ask Clint to tell you just about anything. But fortunately it will at least try to tell you that some things are less plausible than others.
29.
by garrinm 3y ago
Clint should be used only to research information. It provides links to resources. It uses Retrieval Augmented Generation which is less prone providing incorrect information, though it can still happen.
30.
by garrinm 3y ago
Clint should not be used for diagnosis. Only for personal information. Like an web search but more interactive.

More ›