rolisz

11 ms·

151.
by rolisz 2y ago
Oh, I'm sure plenty of recent PhDs don't know about RNNs. They've been dropped like a hot potato in the last 4-5 years.
152.
by rolisz 2y ago
BERT was not large (it was under a billion parameters). And it wasn't an autoregressive language model like GPT
153.
by rolisz 2y ago
Oh, you will have to deal with bullshit corporate politics, but in a different way. You will get projects where there will be conflicts between the stakeholders and you'll be sitting there waiting for the final final final final decisi
154.
by rolisz 2y ago
Lee Cronin is working on Assembly theory, which should enable measuring complexity in substances, with high complexity implying some life there. His interviews with Lex Friedman are pretty good.
155.
by rolisz 2y ago
There are open source variants: https://huggingface.co/prithivida/Splade_PP_en_v1
156.
by rolisz 2y ago
> Public health guidelines targeted at non-professionals need to be simple. And that's how you lose their trust. Just telling me to do simple X when it's clear to me from my own experience that sometimes not-X is better for me
157.
by rolisz 2y ago
Hah. My toddler gladly calls her former walking aid toy a "lawn mower". Random toys become pie and cakes she brings to us to eat.
158.
by rolisz 2y ago
That's a bit outdated. At the scale LLMs are trained, quality starts to matter again.
159.
by rolisz 2y ago
But then how do you test out the latest model that came out from who knows where and has the weirdest dependencies and a super obscure command to install?
160.
by rolisz 2y ago
I took Rybelsus (pill version of Ozempic) for 7 months last year. I lost 8 kg and then I slowly tapered off over the course of 3 months. I'm now +10 kg. :/
161.
by rolisz 2y ago
Source for culture without green: https://archive.nytimes.com/6thfloor.blogs.nytimes.com/2012/... And I agree with you on the second part: everyone agrees the sky is blue and that grass is green. But the fuzzy edg
162.
by rolisz 2y ago
I think you two agree: there are millions of successful projects out there, but they are not standardized. But the guy who made 1000 of them knows exactly how long it will take him to build the 1001th
163.
by rolisz 2y ago
I don't think you can build a machine to identify what is considered green. My wife and I disagree about whether our baby daughters eyes are green or not. There's the joke that men see 16 colors and women see 24 million. Some cult
164.
by rolisz 2y ago
Yes, but nobody got fired for buying IBM
165.
by rolisz 2y ago
This reminds me of finetuning LLMs vs using RAG: In RAG you "know" what the model knows and it's easy for it to give sources - you literally give it to the model in the prompt. In finetuning, the model learns something, but i
166.
by rolisz 3y ago
I'd say that using LLMs to write boilerplate falls under "automation"
167.
by rolisz 3y ago
Well, regular search means switching to a different application, with an implied context switch. It definitely takes longer for many things than just using GitHub copilot.
168.
by rolisz 3y ago
Up until a month ago, the OpenAI embeddings where very poor. But they recently released a new model which is much better then they're previous one. Now it depends un specific usecase (domain, language, length of texts)
169.
by rolisz 3y ago
my concern with grammar based sampling is that it makes the model dumber: after all, you are forcing it to say something else than what it thought would be best.
170.
by rolisz 3y ago
Check out Michael Levin's work, who's done some experiments with skin cell and has shown that they learn to do stuff.
171.
by rolisz 3y ago
Heh, I think my record is three level deeps (dream inside of a dream inside of a dream). Inception ftw.
172.
by rolisz 3y ago
Or OSS will be going in a different direction then what you need, so if you are using it you'll either be stuck on an old version or you have to keep fighting around it. ML libraries in particular have this annoying habit of not being
173.
by rolisz 3y ago
> that are each 1-200 LOC. i can inline that easily in my app and retain full control Isn't that true of most of langchain as well though?
174.
by rolisz 3y ago
Think about hakarl: icelandic rotten shark. It stinks to high heaven: I'm not kidding, a colleague brought a small sealed cup to the office and opened it. The whole floor (where 80+ people were seated) stank like crazy. Regular shark m
175.
by rolisz 3y ago
It's weird to you. Who says the universe must be isotropic?
176.
by rolisz 3y ago
How can you prove that? We literally can't see further than 98 billion light years. How's that not also a religion? :p
177.
by rolisz 3y ago
I'm happy to accept that I have two apples of my desk if all my friends and the guy I bought them from agree with that definition. At some point some will start disagreeing, but with two big full apples, all of them will agree. Or is
178.
by rolisz 3y ago
According to the Big Bang model the universe started with a super small volume. When did it get to infinity? As far as I know, we can see/measure that the universe is about 93 billion light years across. We literally can't know wh
179.
by rolisz 3y ago
Well, I have two apples on my desk. I can't have an infinity of anything. In that sense 2 exists more tangibly than infinity. Our best physical models are just that - models. All models are wrong, some are useful. Infinity in physics i
180.
by rolisz 3y ago
You'd be surprised how much creative tasks like 3D modeling and creating artwork can be broken down into tasks that are very well defined and can be estimated quite correctly.

More ›