p1esk

11 ms·

241.
by p1esk 1y ago
Was this some sort of dress code? Yes
242.
by p1esk 1y ago
5090 has 32GB of RAM. Not sure if that’s enough to fit this model.
243.
by p1esk 1y ago
a lot of research could be invalidated, so obviously would make huge news. A lot of research is unreproducible crap. That’s not news to anyone. Plus, bugs usually make results worse, not better.
244.
by p1esk 1y ago
Me: What color is a green eggplant? ChatGPT: A green eggplant (also called Thai eggplant or green brinjal, depending on the variety) is actually green on the outside — sometimes solid green, sometimes streaked with white or pale green. The
245.
by p1esk 1y ago
Exactly. Five years ago I posted here on HN that AI will pass Turing Test in the next 3 years (I was impressed by Facebook chatbot progress at the time). I was laughed at and downvoted into oblivion. TT was seen by many as a huge milestone,
246.
by p1esk 1y ago
Tensor cores use lower precision, so small numerical differences should be expected.
247.
by p1esk 1y ago
this would be the world’s most catastrophic data leak. Why?
248.
by p1esk 1y ago
I don’t really use LLMs for coding - I still enjoy doing it myself. I use them for brainstorming ideas, reviewing papers (I’m a researcher), practical questions (e.g. how to replace a cooling bypass valve in my car), or curiosity questions
249.
by p1esk 1y ago
You probably just forgot how bad GPT4 was. It was failing half of complicated tasks or questions I asked. GPT5 has roughly 90% success rate for me.
250.
by p1esk 1y ago
GPT5 compared to the original GPT4 is a huge improvement. It exceeded all my expectations from 2 years ago. I really doubted most GPT4 limitations will be resolved so quickly. If they manage a similar quality jump with GPT6, it will probabl
251.
by p1esk 1y ago
Did he say that?
252.
by p1esk 1y ago
I just gave chatgpt this prompt - it produced a picture of a glass filled to the brim with water.
253.
by p1esk 1y ago
I stopped eating desserts and sweets, stopped overeating, stopped eating after 7pm, switched to eating more veggies and unprocessed foods. I started taking supplements (ag1 powder, fish oil, turmeric). Once inflammation started to reduce I
254.
by p1esk 1y ago
I had the same problem with my elbow, electrotherapy did not help. Turned out it was systemic inflammation in my body that was preventing it from healing. Change of diet fixed it.
255.
by p1esk 1y ago
So you’re saying a brain is a computer, right?
256.
by p1esk 1y ago
Do you need differential equations to understand what’s happening in a transistor?
257.
by p1esk 1y ago
Do these numbers compare performance at the same cost?
258.
by p1esk 1y ago
Would you actually have a gun loaded and willing to fire? Or would you lie to the cops?
259.
by p1esk 1y ago
How does it compare to state of the art models? Does it scale?
260.
by p1esk 1y ago
7 million parameters
261.
by p1esk 1y ago
Surely SamA doesn’t actually think that they’ll more than 20x their compute in a few years? If their goal is to train say, a 100T model on the whole youtube dataset they will need 20000x more compute. And that would be my goal if I were h
262.
by p1esk 1y ago
You test different models on your real world problem, and pick the smallest one that works.
263.
by p1esk 1y ago
The paper does not make any clear conclusions about LoRA vs FullFT performance, beyond "the two methods seem to be learning different things".
264.
by p1esk 1y ago
Even that paper itself does not provide any "clear" conclusions about which method is better.
265.
by p1esk 1y ago
What late 20th century tools and technology were used to reconstruct it that weren’t available during its construction?
266.
by p1esk 1y ago
How can inflammation be measured?
267.
by p1esk 1y ago
Out of all things they could have improved, they picked one that didn’t need changing.
268.
by p1esk 1y ago
I think the old logo was distinctive and memorable. I used to go there because of food and because of atmosphere. If the atmosphere is gone I’m less likely to return in the future.
269.
by p1esk 1y ago
The test accuracy in figure 6b shows a marginal increase, and a gentler transition to the overfitting regime, suggesting the regularization is working. Higher LR does not mean there’s overfitting.
270.
by p1esk 1y ago
https://www.dwavequantum.com/blockchain/

More ›