rolisz

17 ms·

241.
by rolisz 3y ago
Thank you for your answer! The founder wants me to both build the product and eventually grow the technical team.
242.
by rolisz 3y ago
They will lose pretty much all their current users when that happens.
243.
by rolisz 3y ago
I find the breakdown into 3 types of CTOs (tech, people and external oriented) very interesting. A very early stage startup (no product yet, just some basic demos, doing pre seed funding) is courting me for a CTO role. What should I expect
244.
by rolisz 3y ago
It's probably also a calculated misery: why newspapers don't want you to read their articles
245.
by rolisz 3y ago
Yes. Started using Kagi while they were in beta. When they launched the paid service I gave them my money right away. 10/10 strong recommend.
246.
by rolisz 3y ago
Doesn't the Turing halting problem also imply that (some) things are not predictable (the halting of certain algorithms)? But I don't think that interferes with causality.
247.
by rolisz 3y ago
Romanian here: the progress this country has made in every single aspect is huge in the 15 years since we joined the EU. Are there downsides? Sure. But yes, the progress for now is massive.
248.
by rolisz 3y ago
Are they still friends with you? Are they still returning your calls? I've been coding for 15 years, but on my NAS I still use the Synology reverse proxy so I don't have to deal with Traefik and self signed SSL.
249.
by rolisz 3y ago
Oh yes. I'm the same! When I am in a larger group, I have a hard time opening up and having meaningful discussions, whereas on one on one settings it's easier.
250.
by rolisz 3y ago
T5 family is awesome. FastChat-T5 has amazing text generation quality (eg for RAG chatbots) and it can easily run on CPU fast enough to do a live conversation.
251.
by rolisz 3y ago
That's a sort of adapting an embedding to a certain task/domain. But I wouldn't call it asymmetric embeddings.
252.
by rolisz 3y ago
It's embeddings that are generated differently for queries and for documents. The idea of that queries are usually short, while documents are longer, so if you embed them in the same way, the most relevant docs will be far from the que
253.
by rolisz 3y ago
No, you don't need an online degree. I had a colleague at Google who was hired straight after highschool.
254.
by rolisz 3y ago
Except there are some weird feedback loops in the body, such as restricting calories leading to reduced calorie consumption, leading to less weight loss than expected.
255.
by rolisz 3y ago
If you have loyalty/coupon apps for the store, sometimes they also give you digital receipts you can download. One day I will automate downloading the receipts and OCR-ing the items .
256.
by rolisz 3y ago
I've watched some lectures about CT (and they flew over my head pretty quickly) and I did get the feeling that on one hand there is some deep insight there (that I was barely able to glimpse), but on the other hand it felt like that ea
257.
by rolisz 3y ago
Re 2: tell that to the company I used to work with. Management heard microservices are good? Then we should have nano services. They mandated that we split things up super small, by functionality, even if there was only one team working on
258.
by rolisz 3y ago
Spacy has integration with LLMs, so it can leverage them. But in general you can often get by with models that are 10-100 times smaller, but are not as general as an LLM.
259.
by rolisz 3y ago
There are embeddings that are trained to reflect similarity, for example SentenceBERT, where the training process pushes pairs of similar sentences (as defined by whoever built the dataset) to have closer embeddings and dissimilar sentences
260.
by rolisz 3y ago
BERT is encoder only and was designed for classification and natural language inference problems. The original Transformer was encoder-decoder and was designed for translation. BERT can't be used in an autoregressive way because it doe
261.
by rolisz 3y ago
For 1000 points, brute force is super quick. Actually, up to 100k (on my machine), brute force takes less than 1 second.
262.
by rolisz 3y ago
I don't really feel that, so it's not obvious to me.
263.
by rolisz 3y ago
What is the 1 day of overhead? That's sounds a lot per week.
264.
by rolisz 3y ago
ChatGPT warns about gotchas with that line.
265.
by rolisz 3y ago
With code interpreter, ChatGPT can check that it's code at least runs without errors
266.
by rolisz 3y ago
But then in the learning to read example, a 14 year old learning to read has seen a ton more letters around them (on billboards, book covers, phones, etc) so it should be easier for them to learn then for an 8 year old.
267.
by rolisz 3y ago
> If there is a subject where that is the case, I do not know what it would be. Once they're past the age of 7 or 8, most or all children are capable of what they would be at older ages. I don't think, for instance, that an 8 y
268.
by rolisz 3y ago
It still doesn't give guarantees, it just makes hallucinations much less likely.
269.
by rolisz 3y ago
I think there is some caching of some of the computation
270.
by rolisz 3y ago
I don't think so. You don't have enough data for finetuning. How would you even fine tune? If you have less then several thousand examples, I wouldn't even think about finetuning

More ›