What are you building if you dont mind sharing? I am moving and all these online applications and such are so annoying.

In terms of a browser agent I built a chrome extension that essentially does everything i do in a browser using locally hosted models and/or APIs. It's called Marley. I can tell Marley to read or summarize the current page and read it aloud (Chatterbox TTS), edit the page, create domain-based site rules for css and js (DOM editing and injection), it has vision (can take screenshots) and action (can click on things and navigate, fill out forms, long horizon tasks, etc). It also can connect to an active CLI session that I might be running in a separate terminal for remote-control. So many things. It manages my passwords in an encrypted vault, it has a media player. LOL This is my 'One extension to rule them all'.

No! because they did not release GPT-2 XL until analogs appeared a year later. So the SOTA model during GPT-2 days was closed weights.

I have a 1500 dollar machine that can run it at 50 tok/s (3 V100s)

How did you buy 3 V100's for $1500??

Not OP and just guessing, but probably SXM2 GPU modules for the V100. Those can be acquired fairly inexpensively, but there is work to do to get them working together and the V100 has some limitations on the types of models you can run.

PCIE 16 gig is <300 dollars these days a piece

on eBay you can buy 4 of them for 1000. PCIE 16 gigs.

innovation requires monopoly is a Theil one liner -- probably true too

Probably not true

I wonder if you can sue him in civic court, could teach him a lesson.

The lesson would be that you spend a great deal of money upfront, go through a tremendous amount of stress, end up with a judgment against someone who doesn't have the money to pay it anyways?

Small claims maybe to recoup towing/recovery fees, but a civil case? No way. That shit will make you go gray quick.

How do you know it was a Honda?

The issue then becomes it writes insane one liners that are impossible to read. Would be difficult to set up an RL environment where the goal is correctness and readability.

you could have an AI judge look for correct and readable code

the codex limits are actually pretty high too. You might want to check it out.

Agree with that, they seem really good limits for daily use on something like Chat GPT Pro $20 account. I'm in the curious situation of using the Codex CLI within Cursor IDE and not really getting value out of my $60 Cursor sub. Plus at every update it seems Cursor seems to break more of their UI in the 'not a cloud agent chat UI' vs the more traditional VSCode sort of layout of code first. I should probably cancel.

I have known about TBPN since early summer last year. They are widely known about in Cali startup culture at least.

This is the IT version of "I was listening to them before they got big!". Just as nerdy.

its probably not even a conscious decision from dwarkesh to be hyperbolic. pod casters who are hyperbolic are just simply watched more

he was not sure, but if i recall correctly, he put the probability at something like 90 percent of being able to do non verifiable tasks.

How the hell does someone come up with a probability like this? It's such bullshit, a telltale sign is the whole numbers used.

It's not based on anything but gut feeling, which is fine. Just state so.