rolisz

13 ms·

181.
by rolisz 3y ago
Blood test for what exactly? I'm not OP, but last year I also had a health issue and after going to three doctors, eventually ChatGPT told me to do some tests and one of them was the thing I was deficient in.
182.
by rolisz 3y ago
Recently tried Flask for a project started by someone else, made me want to run to FastAPI right away.
183.
by rolisz 3y ago
Also family of four: we fill a 120l garbage bin per week, to the brim, and a 120l compost bin every two weeks (this one maybe is not full, but it stinks by that time) and a ton of "recyclables" (cardboard, paper, metals, plastics)
184.
by rolisz 3y ago
With RAG (so when you give it source text), hallucinations are greatly reduced (not completely, but greatly).
185.
by rolisz 3y ago
I don't think they always retrain models from scratch. Sometimes they might do continual learning (take the old model and train it on newer data)
186.
by rolisz 3y ago
BERT has around that many parameters and it runs on CPU in 200ms
187.
by rolisz 3y ago
We're in agreement :)
188.
by rolisz 3y ago
Not quite. The GPT4 offered through the chat completion API will answer questions without any special system prompts. What these prompts do is try to do some extra steering of the chat model, on top of the steering done via RLHF.
189.
by rolisz 3y ago
React was developed by Facebook. I don't think they did it for small components on otherwise static sites.
190.
by rolisz 3y ago
If I understand this announcement correctly, they removed support for JSONP
191.
by rolisz 3y ago
I've been self employed for 3 years now. I help companies get started with machine learning. ChatGPT is the best marketing for me: everyone and their dog wants ML because of ChatGPT. I do advisory work and development too.
192.
by rolisz 3y ago
I enjoy life a lot more with kids. As Paul Graham put it in one of his essays, kids are joy on tap.
193.
by rolisz 3y ago
Any sufficiently advanced technology is indistinguishable from magic. Pile in enough abstractions and yeah, the cloud becomes magic. Are there many mystical experiences that if you peeled away enough layers you could have rational explanati
194.
by rolisz 3y ago
Joscha Bach also talks about this a lot. He calls the consciousness the monkey with a stick controlling the elephant. For a starting point, listen to his Lex Fridman interviews.
195.
by rolisz 3y ago
Are they worried about you (the startup) or about OpenAI? If it's the former, there's nothing new, startups have fought this battle for a while. If the reluctance is around OpenAI, there are several options: Azure OpenAI services
196.
by rolisz 3y ago
There are no RCTs to prove that using a parachute leads to better outcomes than not using them when jumping out of a plane. Does that mean you wouldn't use a parachute if in that situation?
197.
by rolisz 3y ago
Do you mean using brute force (exact search) and not approximate nearest neighbor search? From my unscientific benchmarks, doing an exact search with 100k vectors takes about 1 second. Usually chatbots take much longer to generate the text,
198.
by rolisz 3y ago
But Loxone is also a very different system, living in it's own bubble. When I talked to my local Loxone representative, they told me you have to get their special bulbs (costing 80$ a piece), running over their own wiring (24V, not 240
199.
by rolisz 3y ago
I mostly know Python. Is there something similar in Python? Or the backend language is not relevant, because you mostly use this from JavaScript anyway?
200.
by rolisz 3y ago
But that still makes it great for RAG applications, where I want the answer to be based on my data, not on whatever it learned from the web.
201.
by rolisz 3y ago
So you're suggesting that instead of asking an LLM, we should spend time on Fiverr/Upwork to find someone to do random coding tasks that might not fall under our expertise? Can you do that for less than 20$/month?
202.
by rolisz 3y ago
Simple fix (at least for ChatGPT): ask it to avoid drawings with similarities to copyrighted characters.
203.
by rolisz 3y ago
Oh c'mon, those prompts were not generic. Italian plumbers? How many other Italian plumbers do you know? What's the most popular soda in a red can?
204.
by rolisz 3y ago
Trivial change, except for also having to change all the prompts to work well with Mistral. LLMs are not equal, you have to ask each one "nicely" in a different way.
205.
by rolisz 3y ago
Photos and videos are very different from text. 300 GB of text is not comparable to 300 GB of photos. You can do something using image embeddings to get what you want.
206.
by rolisz 3y ago
What are you trying to achieve with that dataset from LibGen? I kinda expect that GPT4 was trained on the data that is available on LibGen
207.
by rolisz 3y ago
Oh, it is Scott's problem. I got the references and I still found them to be too much.
208.
by rolisz 3y ago
How many memory leaks were discovered only during the winter code freeze, because there were no pushes being done, so no server restarts
209.
by rolisz 3y ago
Mistral MOE model is 8*7B parameters, so you should compare training time to similar sized models, not to 7B ones.
210.
by rolisz 3y ago
FYI, you might be out of luck in other countries: in Romania (where I live) semaglutide is made available for free only for diabetics. If you want it for off label usage, you'll pay for it. In Hungary, it's illegal for doctors to

More ›