deep_etcetera

You can see some examples here https://aipaygrad.es/

Hi, I'm the same age as you, born in the UK but have been working in USA for a few years now. I am also feeling burned out by my job, spend a lot of time fantasizing about retiring early.

This is a shot in the dark but I feel like a lot of my mental state has been caused by covid, missing regularly seeing my friends, family and the alienating nature of interacting with my colleagues only via screen. Because of this I have resolved to not make any large decisions until covid is completely over, since it's hard for me to assess how much differently I will feel once things get back to normal. Till then I save as much as possible to give myself more options.

I think it will be great if they can create some mechanism to compensate people for their data, I just suspect many people conflate the value of their data as training data and say how much they might charge a client to write some similar code.

I'm guessing that since there are hundreds of millions of repositories the typical marginal value of someone's contributions would optimistically be on the order of a few dollars. But since the consensus on HN is that they spend very little time actually coding and there is no use-case for copilot, perhaps it worth a lot less.

Yes, if you don't condition on the past moves then the distribution you're modeling is where you randomly pick a 1100 player to choose each move as you say. What I'm saying is that there will be no wisdom of the crowd effect.

This only true if you select the most likely move instead of sampling from the probability distribution over possible moves. In the latter case there is no reason for there to be a wisdom of the crowd effect.

If you sample from the probability distribution you are modeling, there is no reason it shouldn't play like a 1100 player.

I have no idea why you would find the existence of this machine disappointing. If you wanted to work out something for yourself then you still can.

It would be more efficient to use drones but precommit to shooting a soldier back in the US whenever a drone gets shot down.

I don't really understand your point, you don't have to use it to generate the next sentence or paragraph of your story. You can it directly to generate ideas for what comes next by asking it to complete a summarization of the plot.

I must agree reading those copypasta, "300 red scares" and "300 confirmed scoots" had me close to tears.

Imo these are researchers. Their job was to validate their algorithm for doing generative super-resolution on a dataset, they chose the largest and most well-known dataset, it worked reasonably well on their dataset. The model itself is not productive ready for at least the reason that the dataset is not representative. This is ubiquitous in ML papers, they validate their idea on not completely realistic but widely available datasets. The outcome is a piece of knowledge about the behavior on that dataset not a product.

Maybe it would be helpful if you gave an example of the simplest python function it won't be able to synthesize, and if/when they release the code GPT into the API we can test your prediction.

Who are you quoting here?

This article seems pointless without detailing any of the results, why not give some numbers on their performance on benchmarks?

I doubt a human could learn to become even remotely competitive with only self-play within a human lifetime. Go has improved via a distributed effort, so we should try to estimate the number of go games played by humanity (as an upper bound).

Who here can beat 4700 https://www.retrogames.cz/play_124-Atari2600.php?language=EN ?

This approach will be thwarted if the time is shown in the observation space. If this is so then every kth state will give a novelty reward and every other state will give zero, it doesn't matter what the agent does.