p1esk

41 ms·

901.
by p1esk 3y ago
Zero shot means a model trained one one dataset, or for one task, performs well on another dataset, or on another task - without any finetuning.
902.
by p1esk 3y ago
Written by none other than Jaron Lanier. That probably should be in the title.
903.
by p1esk 3y ago
You’ve been in the industry 19 years and you still work 12 hour days? Why?
904.
by p1esk 3y ago
In that case, building things might be a better idea than building a network. Build something, put it on GH, post it here as ShowHN. If you’re a good developer people will notice.
905.
by p1esk 3y ago
If you like playing with LLMs, $20/mo for GPT4 is definitely worth it.
906.
by p1esk 3y ago
By ChatGPT do you mean GPT4 or GPT3.5?
907.
by p1esk 3y ago
Do you have 12 months or more in savings? If not, that’s why. 12 months should be enough to find another job, even when economy is bad.
908.
by p1esk 3y ago
How does memory bandwidth of M2 Max compare to memory bandwidth of Ryzen 8700G? 400GBs vs 80GBs? Exactly my point. 400GBs approaches lower end graphics card territory (e.g. 3060)
909.
by p1esk 3y ago
A regular CPU supports Pytorch and has lots of RAM at low price. The problem is memory bandwidth.
910.
by p1esk 3y ago
It’s iGPU. There’s no GPU memory.
911.
by p1esk 3y ago
Yes. He would have become Steve Jobs.
912.
by p1esk 3y ago
Yes, but it would be more expensive than current methods.
913.
by p1esk 3y ago
if someone writes a new back-end to PyTorch If that was easy to do surely AMD would have done it by now? After many years of trying?
914.
by p1esk 3y ago
Only with domain experience you can tell that something might be wrong. Obviously. How is this different from any other field of science or engineering? you need to understand the theoretical motivation of things and check all operations
915.
by p1esk 3y ago
I've been an ML researcher for the last 11 years. Last week I spent 3 days debugging an issue in my code which had nothing to do with ML. It was my search algorithm not properly modifying the state of an object in my quantization algor
916.
by p1esk 3y ago
It depends on what is it you do when you “do ml”.
917.
by p1esk 3y ago
All of these hypotheses will take on the order of days to check. OK, but you can check them, right? How is that different from a regular software bug?
918.
by p1esk 3y ago
I did consulting a few times in my career (for startups, not FI). The last time: 1. My rate was $350/hr, which was about 3x my salary at the time. I would have laughed if someone offered me to do full time exclusive consulting for them
919.
by p1esk 3y ago
OK, so it is reasonable to assume that the chance the progress will continue in the next two years is about 50%, would you agree? If so, why are people in this thread dismiss my concerns as unreasonable?
920.
by p1esk 3y ago
Is there an indication it won't be?
921.
by p1esk 3y ago
In two years the job of a senior software engineer will be completely redefined, so I'm not sure how we could test your prediction. Would someone who tells an AI model to build something still be an engineer, or would we call that pers
922.
by p1esk 3y ago
unprecedented disruption at a macro scale that humanity has never seen before Exactly. How likely do you think this will happen? I estimate the chance of this happening after the release of GPT-5: 50%. Chance of this happening after GP
923.
by p1esk 3y ago
Progression GPT1-2-3-4 in last 5 years, with no indication of any slowdown in the near future.
924.
by p1esk 3y ago
The obvious thing is gpt-6 (or even gpt-5) will be able to do your job better and faster than you. This will happen in 1-2 years. That doesn’t scare you?
925.
by p1esk 3y ago
How to provide value if gpt-6 (or even gpt-5) can do your job better and faster than you?
926.
by p1esk 3y ago
An article on language shaping one’s worldview and it does not mention Wittgenstein?
927.
by p1esk 3y ago
It also includes the cost of Amazon Prime video content. I don’t know why he called it R&D.
928.
by p1esk 3y ago
Is Numpy or Pytorch an array language? With things like einsum and elaborate slicing/indexing mechanisms.
929.
by p1esk 3y ago
Everything is learnable but most things are not worth learning.
930.
by p1esk 3y ago
he is willing to lie and make false promises in order to gain support This describes pretty much every politician who ever had to be democratically elected. Has any other American president behaved differently?

More ›