rolisz

14 ms·

211.
by rolisz 3y ago
Maybe the regulations will be Mistral shaped.
212.
by rolisz 3y ago
Can you get it running using Cosmopolitan? :D
213.
by rolisz 3y ago
Have you tried Azure Document Intelligence? I've had very good results with it.
214.
by rolisz 3y ago
If we get 1-bit quantization, wouldn't it be basically a bunch of nested if's and else's?
215.
by rolisz 3y ago
What is up with that eval @32? Am I reading it correctly that they are generating 32 responses and taking majority? Who will use the API like that? That feels like such a fake way to improve metrics
216.
by rolisz 3y ago
Following the advice of chatgpt without double checking? Bad idea. Using ChatGPT as a starting point? Sounds really good to me, been there, done that.
217.
by rolisz 3y ago
0 and 1 don't support your claim that WhatsApp discloses message content. those two links explicitly say WhatsApp doesn't give messages to the FBI.
218.
by rolisz 3y ago
I think many designs in tech have two states between which they oscillate. We don't go back exactly to the same thing, we do improve on things, but things are very similar to the way they were done 20-30 years ago (or 40 in this case)
219.
by rolisz 3y ago
Why do you think cloud providers can undercut OpenAI? From what I know, Llama 70b is more expensive to run than GPT-3.5, unless you can get 70+% utilization rate for your GPUs, which is hard to do. So far we don't have any open source
220.
by rolisz 3y ago
No, it's a customer who'll be footing the bill. Sometimes that customer can go somewhere else, sometimes not (eg. Internet providers in the US)
221.
by rolisz 3y ago
You can remove videos from your YouTube history and then they won't influence the recommendations you get.
222.
by rolisz 3y ago
Yes, this is called triplet loss and has made embeddings much better.
223.
by rolisz 3y ago
Interesting. We'll see in 2026 when it's finished how much it actually costs and how well it actually works :)
224.
by rolisz 3y ago
> It's surprisingly effective to store thermal energy over several months. Do tell more.
225.
by rolisz 3y ago
I don't think you can do that easily. If you already have a list of embeddings from a different model, you might be able to generate an alignment somehow, but in general, I wouldn't recommend it.
226.
by rolisz 3y ago
Eliminating deserts doesn't make cravings go away for me. I feel hungry (or a weird similar feeling) until I eat something sweet, no matter how much I eat of a main course. But I tried the potato diet and my cravings were gone. For two
227.
by rolisz 3y ago
You can do the one after the other. When on vacation, no tech. When at home, one weekend a month (or some afternoons), do some tinkering and learning
228.
by rolisz 3y ago
So you're proposing replacing a hypothetical risk (Geoffrey Hinton puts the probability at 1%) with guaranteed pain and destruction, which might escalate to nuclear war? Thank you, but no thank you.
229.
by rolisz 3y ago
Nope. A glass of freshly squeezed oranges has more calories than a can of Coke and the impact on your blood sugar is similar.
230.
by rolisz 3y ago
And... How do you do that? Seriously, people have been talking about stoping CO2 emissions for what, 50 years? We've barely made progress there.
231.
by rolisz 3y ago
That's a known issue with the BGE embeddings, the authors warn about that in the model card. Their recommendation is to choose more carefully the thresholds for similarity (which will be much higher than for other embeddings)
232.
by rolisz 3y ago
Why did you switch back to Chrome? What features did you miss?
233.
by rolisz 3y ago
Happy paying user here. Managed to even convert some friends to it.
234.
by rolisz 3y ago
Do they? Aren't EAs mostly doomers? I haven't followed the debate closely.
235.
by rolisz 3y ago
That's not quite tfidf though. I agree you can get better results than that with Ada embeddings, but I would argue you can get even better results with embeddings from smaller chunks.
236.
by rolisz 3y ago
Then a lot of stuff is not open source. Have you tried reproducing random GitHub repos, especially in machine learning?
237.
by rolisz 3y ago
I think the point is: if you compress 25 pages of text into 1024 floats, you will lose a ton of information, regardless of what the use case is, so you're probably still better of with chunking.
238.
by rolisz 3y ago
I guess Polish people had it worse than Romanians and Hungarians. All our accents are simple to strip away and they still kinda sound like the base letter. But I agree with you: I don't trust translations and I think the benefits from
239.
by rolisz 3y ago
Would that mean that if we had decentralized pulmonary control, we wouldn't be able to hold our breath?
240.
by rolisz 3y ago
Your browser is also automating things, you're not resolving DNS and writing all the http requests by hand.

More ›