p1esk

22 ms·

391.
by p1esk 2y ago
He means it clusters around the distance from a pole to the equator.
392.
by p1esk 2y ago
Whose fault is that?
393.
by p1esk 2y ago
AMD had 12 years to become competitive. Deep learning revolution started in 2012.
394.
Intelligence at the Edge of Chaos(arxiv.org)
3 points by p1esk 2y ago|0 comments
395.
by p1esk 2y ago
I would not call them “incredible”. Video generation is still at gpt-1 level. It will need several orders of magnitude scaling to get to gpt-2 level. We’ve just started making very first baby steps in this direction.
396.
by p1esk 2y ago
The thing that people strangely still seem to not get is that these systems will continue to improve rapidly. This baffles me as well. So many low hanging fruits to pick, we literally just started with these large models. One obvious dire
397.
by p1esk 2y ago
Sounds like you had a shitty job. Mine is great, I’d do it for free if I didn’t need the money.
398.
by p1esk 2y ago
30% unemployment is bad, but 100%, or even 90% should be great - the age of abundance, where robots are doing all the work, and everything is so cheap that anyone can have the basic necessities for free.
399.
Ask HN: Why are so many people leaving OpenAI?
26 points by p1esk 2y ago|12 comments
400.
by p1esk 2y ago
This argument could not be made before March 14 2023, when the first actual AI was released (GPT-4). I remember that day very well because I was extensively testing every GPT model before that (part of my job as an AI researcher). The ent
401.
by p1esk 2y ago
Because it’s already able to do almost everything a brain can? State of the art AI models can already learn, reason, communicate, and even create - better than most humans. Using a lot less neurons, and much simpler neuronal structures. All
402.
by p1esk 2y ago
What about AI software running on silicon chips? Soon it will reach levels of complexity vastly exceeding any human brain. To these systems we will look like bugs, or maybe even just cells - they might not even classify us as being alive, l
403.
by p1esk 2y ago
In my opinion, he is more notable for inventing a variational autoencoder.
404.
by p1esk 2y ago
It depends on what would actually change.
405.
by p1esk 2y ago
would be… mostly office suite, with a little LLM inference sprinkled in. No, it would be LLM inference with a little bit of an office suite sprinkled in.
406.
by p1esk 2y ago
Very true, once you're senior enough, couple of hours of focused effort is enough to make decent (visible) progress.
407.
by p1esk 2y ago
Maybe he wants to make the world a better place?
408.
by p1esk 2y ago
Is GPT-4 not a “big thing in AI”?
409.
by p1esk 2y ago
Sam’s ambitions are way beyond “salary”.
410.
by p1esk 2y ago
You can post an ad at your local uni ECE department and hope some grad student would have enough FPGA experience to handle a “simple” project.
411.
by p1esk 2y ago
I agree. Though it's good to have options for GPU accelerated Numpy. Especially if Google decides to discontinue Jax at some point.
412.
by p1esk 2y ago
Google “llm scaling laws”.
413.
by p1esk 2y ago
Did you mean you'll be a SWE manager of LLMs? Yes, hopefully for a couple more years until such managers are themselves automated.
414.
by p1esk 2y ago
So, “do as I say, not as I do”? I just looked at Equinox and it does look a little better than flax, but flax seems to be more widely used. Both haiku and trax seem to be on the way out. All four were intended to do exactly the same thing.
415.
by p1esk 2y ago
Use any Too much overhead when I just want to get shit done.
416.
by p1esk 2y ago
Exactly. O1 is just a minor gpt-4 tweak before Orion.
417.
by p1esk 2y ago
When SWE management’s no longer a thing I’m afraid there won’t be many professions left. By that time I really hope we will have some sort if UBI, so at least we won’t starve to death.
418.
by p1esk 2y ago
I will probably be forced to go into SWE management then, something I don’t enjoy. Why do I feel this way? Because I expect significant improvement from Orion, and I don’t think it will stop improving after that.
419.
by p1esk 2y ago
My guess is software engineers still have about 2-3 years of employment left. After that they will look like those switchboard operators who manually plugged cables to connect phone calls.
420.
by p1esk 2y ago
I wish Jax had everything I need to experiment with DL models built in, natively - like Pytorch. Instead there are many third party libraries (flax, trax, haiku, this one, etc). I have no idea which one to use. This was the case when I firs

More ›