kadushka

16 ms·

181.
by kadushka 2y ago
But I want their models to know as much as possible about me - it should improve my experience using them.
182.
by kadushka 2y ago
I'm having trouble understanding, or perhaps, accepting, your point of view. When you say "It’s physical properties would be emulated in software", what really happens is still electrical signals flowing between transistors.
183.
by kadushka 2y ago
ideally, I would want the model to remember what I told it a few million tokens ago Yes, you can keep finetuning your model on every chat you have with it. You can definitely make it remember everything you have ever said. LLMs are excell
184.
by kadushka 2y ago
I've been discussing a philosophical topic (brain uploading) with all major models in the last two years. This is a topic I've read and thought about for a long time. Until o3, the responses I got from all other models (Gemini 2.5
185.
by kadushka 2y ago
I’m paying $20/mo, and I’m paying the same for Gemini and for Claude. What’s wrong with my argument? You questioned the performance of the model on non-STEM tasks, and I gave you my impression.
186.
by kadushka 2y ago
users aren’t finetuning models after every chat Users can do that if they want, but it’s more effective and more efficient to do that after every billion chats, and I’m sure OpenAI does it.
187.
by kadushka 2y ago
I had a philosophical discussion with o3 model earlier today. It was much better than 2.5 pro. In fact it was pretty much what I would expect from a professional philosopher.
188.
by kadushka 2y ago
I don’t - if I can use a tool that saves me 10 hours a week, that’s 10 hours more beach time for me.
189.
by kadushka 2y ago
So to you AGI == terminators? Interesting.
190.
by kadushka 2y ago
Google has the leading model on pretty much every metric Correction: Google had the leading model for three weeks. Today it’s back to the second place.
191.
by kadushka 2y ago
Long term memory in an LLM is its weights.
192.
by kadushka 2y ago
Imagenet had improved the error rate by 100*11/25=44%. o1 to o3 error rate went from 28 to 19, so 100*9/28=32%. But these are meaningless comparisons because it’s typically harder to improve already good results.
193.
by kadushka 2y ago
We should clarify the definition of "uploading". I see two cases: 1. We gradually replace each neuron in-place - without any external connections. This means a biological brain slowly turns into a silicon brain (or whatever materi
194.
by kadushka 2y ago
Let's explore the concept of "cloud based ones" neurons. Let's say your brain has just three neurons, connected with each other. You remove one of them, and connect the two remaining one an external terminal - a physical
195.
by kadushka 2y ago
I see a huge progress made since the first gpt-4 release. The reliability of answers has improved an order of magnitude. Two years ago, more than half of my questions resulted in incorrect or partially correct answers (most of my queries ar
196.
by kadushka 2y ago
upload yourself You mean “upload a copy of yourself”, right? I guess I would be happy for my uploaded copy to live forever while I continue to age.
197.
by kadushka 2y ago
To make some progress in ML you might not need a lot of theory, but to understand why things work – you absolutely do. Moreover, the DL field as a whole desperately needs theories explaining what’s going on in these large models.
198.
by kadushka 2y ago
In our field (AI) nobody can see even 5 months ahead, including people who are training a model today to be released 5 months from now. Predicting something 5 years from now is about as accurate as predicting something 100 years from now.
199.
by kadushka 2y ago
So, how do we solve questions like these? How about collecting a lot of data and looking for patterns in that data? In the process, scientists typically produce some hypotheses, test them by collecting more data and finding more patterns, a
200.
by kadushka 2y ago
There are LLMs which do not generate one token at a time: https://arxiv.org/abs/2502.09992 They do not reason significantly better than autoregressive LLMs. Which makes me question “one token at a time” as the bottlene
201.
by kadushka 2y ago
Theory is still needed if you want to understand things like variational inference (which is in turn needed to understand things like diffusion models). It’s just like physics - you need math theories to understand things like quantum mecha
202.
OpenAI change of plans – o3/o4 models are coming(techcrunch.com)
4 points by kadushka 2y ago|0 comments
203.
by kadushka 2y ago
If you have a bottleneck in terms of memory bandwidth utilization, this method is great - it would utilize the idle compute.
204.
by kadushka 2y ago
I'm guessing the overlap of those who know how to fix it with those willing to pay is not that large.
205.
by kadushka 2y ago
My wife used ChatGPT to fix a garage door opener while I was out. Uploaded a picture, and followed the instructions. She's not a manual labor inclined person by any stretch and knew nothing about garage door openers.
206.
by kadushka 2y ago
the industrial revolution made the horse redundant I love this analogy. People hope they'll be like carriage drivers who transitioned into driving cars. Not this time. On the other hand, I'm not sure how this transition actually
207.
by kadushka 2y ago
People ain't gonna sign large contracts with a bot. True. Bots will sign contracts with other bots.
208.
by kadushka 2y ago
Did you read the article? A bureau spokesperson issued a statement: “I can confirm we conducted court-authorized activity at the address in Carmel today.
209.
Nvidia GTC Recap(blog.glennklockwood.com)
1 points by kadushka 2y ago|0 comments
210.
by kadushka 2y ago
For me, it’s the opposite – speed is there, intelligence is still lacking sometimes. I’m OK waiting 10 minutes with o1-pro, but I want a deep insight into the issue I’m brainstorming. Hopefully GPT-5 will deliver.

More ›