trc001

These were Yale students, so probably smarter than average

In my experience, Ivy League students are some of the most profoundly stupid people I've ever met

It's become trendy to delete it. I say trendy because many papers delete it without offering any proof that it is meaningless

Uh, the bellman equation was first used for control theory and is the foundation of modern reinforcement learning... so wouldn't that imply LLMs "come from" control theory?

Oobabooga is still good as a Swiss Army knife sort of wrapper for a single user trying out new models

It's really only one guy that underestimates it

This is just inaccurate

Real estate price estimates would be the classic, and frankly still common, example

Latex is great, but installing it is a hassle. FWIW, typst is lean enough to install (very quickly) as a vscode extension. I’ve never had a useful Tex install that was less than like 4gb

I tried this out this week for a 2 page technical memo. It generally works great, and pandoc can sort of convert latex to typst to get you started, but the crucial thing it’s missing atm is the ability to embed pdf images. You have to convert to svg or a rasterized format first. Way too much of my workflows are written to produce visualizations in pdf, and this creates enough friction to keep me from using it for anything bigger than a memo. Other than the pdf thing though, I’m sold. The improved compilation time (which is not an overleaf thing — latex is slow) is huge

I feel like there must be a better source for this information

Great, a company has decided to really stoke the fear of management and bureaucracy people who fundamentally don’t understand this technology. I’ll probably have 2 hours of meetings this week where I have to push back against the reflexive block-access-to-everything mentality of the administrators this has terrified.

Two quick steps should be taken

Step 1 is permabaning these idiots from huggingface. Ban their emails, ban their ip addresses. Kick them out of conferences. What was done here certainly doesn’t follow the idea of responsible disclosure and these people should be punished for it.

Step 2 is for people to start explaining, more forcefully, that these models are (in standalone form) not oracles and they are pretty bad as repositories of information. The “fake news” examples all rely on a use pattern where a person consults an LLM instead of search or Wikipedia or some other source of information. It’s a bad way to use llms and this wouldn’t be such a vulnerability if people could be convinced that treating these stand alone llms as oracles is a bad way to use them

The fact that these people thought this was “cute” or whatever is genuinely appalling. Jesus.

What is this actually measuring though? Half of what makes these tasks difficult seems to be the lack of an explicit target. Failure in that task suggests more about critical thinking and reasoning skills than "computer skills", which is suggestive of something much more specific...

Not to be too pedantic, but I think the Kansas City you are referring to is in Missouri :)

You’re right about the draw though. The KC metro is growing — sort of feels like Denver felt about 15 years ago

They should change the state motto to, “a slightly less boring version of Nebraska”

I’ve been saying some version of this to people for years.

But in his defense, I think he’s been more the victim of the secular decline in polling quality than anything else.

It seems that the better law would be to require sites carry some standardized nsfw/18+ tag or marker so they could easily be filtered by parental controls. Sites in violation could be fined if they did not include the marker.

There would probably still be trivial ways to work around this (e.g. content blocker to remove the marker) but it would be no more trivial than the current vpn workaround and it would be more privacy preserving.

In academic or research work this can be tricky because sometimes you put something down and can’t come back to it for a week/month. At that point, this trick is more of a hindrance than anything…

Good to see more people pointing out the problems using standalone llm (I.e. not connected to an external data source/the internet) for search. So many people I talk to dismiss gpt because it can’t answer some subject-specific question accurately.

I think parent is getting at the idea that dnn learns abstract representations of the data in whatever way is useful for the mode objective function. But I agree, the article still seems to present a reasonable argument

This. I’ve been pointing to Serle for months now on this and soo many people who have never bothered to study philosophy seem to want to weigh in on this it’s maddening.

So, so many bad/shallow takes that don’t seem to realize that people have been debating this for hundreds of years already

Well, there sort of is a linear relationship between the ammount of training data and the quality of the model [1]

[1] https://arxiv.org/abs/2203.15556

Am I the only person who thinks we should pump the breaks on letting something like this write and execute code? I’m not on the whole “gpt is alive” train, but… you know, better safe than sorry…