Andrej Karpathy – State of GPT ( At MSBuild 2023) | Hacker News Reader

Andrej Karpathy – State of GPT ( At MSBuild 2023)

youtube.com

53 pointsbeauzero15 comments

Open articleView on HN

Video available here:

https://build.microsoft.com/en-US/sessions/db3f4859-cd30-444...

This is the only option now since it looks like the original Youtube link (the Build livestream) has been made private.

This link would be better: https://www.youtube.com/watch?v=B4WAdtlSsK8&t=25890s (starting of his part)

The video seems to have been made private unfortunately.

Video is here: https://www.youtube.com/watch?v=bZQun8Y4L2A

First half is an overview, second is more practical and current.

I wish HN had a filter so I could hide all posts with "AI", "GPT", "LLM" and the like. It's so tiresome.

Maybe you could use an llm to classify them :)

Indeed.

I'd like something like this for media, but for whatever campaign the media is currently cycling at the moment. right now on a daily basis I'm reading about "bud light" and I don't know or care anything about it.

I'll save the sarcasm, but these advances in AI are arguably among the most astounding advances we've ever made as a species. I don't sympathize at all with the perspective that these topics are tiresome. To me, it comes off as jaded and obstinate. So much get-off-my-lawn around these subjects...

Well, good for you then. I really don't find it that impressive. Yeah, some recent advances are somewhat cool and all but I am just tired of people making absurd claims and predictions about it and hearing about "AI" every single day almost everywhere.

Sounds like an idea for gpt-based startup. You should raise some money for that

Sounds like a good violent monkey [0] script for you do this weekend. :)

[0] https://violentmonkey.github.io/

This isn't a sarcastic comment: you could ask ChatGPT to help you write a browser plugin to filter out such stories, if one doesn't exist already. (Not sure if the plugin would work on mobile, though.)

The more I use GPT it becomes very obvious that this is a dead end. ChatGPT constantly becomes more stupid and it does not matter if you use version 3.5 or 4. Had version 4 repeating the same very incorrect answer 4 times in the same chat today. I even told it explicitly to not give that answer and it did it anyway. Repeatedly. The chat part simply does not work any more. If it ever did or if it was just an illusion.

When you actually think about it. LLMs can never work for anything advanced. It is basically a dude that is good at guessing.

Reply on news.ycombinator.com