mattyyeung

7 ms·

1.
by mattyyeung 11mo ago
oooh fantastic. Thanks! Exciting My initial test looks pretty good, seems I will need to muck around a bit to get to "one click runs the simulation and displays the trace and stats" but that rust link seems to have clues. It sure
2.
by mattyyeung 11mo ago
Does anybody have advice on how to embed perfetto UI into local tool? I'm currently using the old catapult chrome trace viewer to visualise and inspect scheduler outputs but there are some shiny features in perfetto that would be great
3.
by mattyyeung 2y ago
You may be interested "Deterministic Quoting"[1]. This doesn't completely "solve" hallucinations, but I would argue that we do get "good enough" in several applications Disclosure: author on [1] [1] https
4.
by mattyyeung 2y ago
I think you could get a pretty good solution for that using RAG and some tricks with prompt engineering and semantic chunking. With google's very-long-context models (Gemini) you may also have good results simply with some prompt engin
5.
by mattyyeung 2y ago
I would love to learn more, where would you recommend I look?
6.
by mattyyeung 2y ago
Yes, extractive QA is one of the improvements beyond the "minimalist implementation" from the article. In our lingo, we'd say that's another way to create a deterministic quotation. So far, we haven't found extracti
7.
by mattyyeung 2y ago
Can quotations be hallucinated? Or are you using something like "deterministic quoting"[1]? Disclosure: author on that work. [1] https://mattyyeung.github.io/deterministic-quoting
8.
by mattyyeung 2y ago
Thanks for the thought-provoking comment. It's all grey isn't it? Vanilla RAG is a big step along the spectrum from LLM towards search, DQ is perhaps another small step. I'm no expert in search but I've read that those s
9.
by mattyyeung 2y ago
I'd put it like this: RAG = search engine, but sometimes hallucinates RAG + deterministic quoting = search engine that displays real excerpts from pages.
10.
by mattyyeung 2y ago
Unfortunately I don't believe that accuracy will scale "multiplicitively". You'll typically only marginally improve beyond 95%... and how much is enough? Even with such a system, which will still have some hallucination
11.
by mattyyeung 2y ago
Two possibilities: (1) if the <title> contents (unique reference string) doesn't match, then it's trivially detected. Typically the query is re-run (non-determinism comes in handy sometimes) or if problems persist we show an
12.
by mattyyeung 2y ago
Author here, thanks for your interest! Surprising way to wake up in the morning. Happy to answer questions
13.
Hallucination-Free RAG: How We Bring Trust to Healthcare AI(mattyyeung.github.io)
2 points by mattyyeung 2y ago|0 comments
14.
Deterministic Quoting: Making LLMs safer for healthcare(mattyyeung.github.io)
117 points by mattyyeung 2y ago|38 comments
15.
Show HN: Greenforecast.au: 7-day forecast for renewable energy(mattyyeung.github.io)
1 points by mattyyeung 2y ago|0 comments
16.
by mattyyeung 3y ago
I didn't really track... apparently there's 30-ish hours of video, but lectures are just the beginning. The real learning happens when you play and build. The first lesson was released in October I think.
17.
by mattyyeung 3y ago
Depends. Some would benefit from simultaneous, others sequential. I did them in "chunks" starting with fastai but that was more driven by the release schedule. Personally I'd recommend trying both and seeing which style you p
18.
by mattyyeung 3y ago
I just finished this! My thoughts: I recommend it. I feel I can now read an arbitrary paper, frown a lot, and eventually understand what it's talking about - to the point where I can implement my own buggy version. And hey, I built my
19.
Show HN: Kinect hack interface for 5 classic games: streetfighter, airguitarhero(mattyyeung.tumblr.com)
1 points by mattyyeung 15y ago|0 comments