Of course. That’s a separate problem. I’m just commenting on the part where you say that they should give you the prompt.

Sticking with your analogy, it’s like if you commissioned someone to paint something specific, then decided that you liked the result and wanted to share it with your friends, so you give your friend the instructions you gave to the painter instead of the painting itself.

They add new information all the time. That’s part of the problem with blindly accepting everything they output.

LLMs are stochastic. If I send you the prompt instead of the output, then there’s no guarantee that the output you get will be correct. If I generate the text myself, I can verify that it’s correct before sending it off.

The problem is that as the recipient, you have no idea whether I’ve even read the output, let alone verified or understood it. And with the low barrier to entry, it’s much more likely that you’re getting unverified slop. Sharing the prompt isn’t going to help with that.

Edit: sorry, posted before I finished writing.

Huh. It was a mandatory part of our back-to-school kit for most years here.

France has nukes

We’ve had White-out for almost a century now. They most certainly could’ve corrected those mistakes instead of starting over.

Like suodiu, but gold.

The intended audience of “stochastic parrot” was other AI researchers who do use the term “stochastic” on a regular basis.

Throwing an LLM at it is probably one of the most effective calls for maintainers. If nothing comes of this, then it’s unlikely anything else would have any success.

In regular people terms, it’s like if I saved up $100 and spent it to buy shares of an index fund. Yeah, it’s going to take a while to make back the money I spent, but those $100 didn’t just disappear. I can sell those shares and get back approximately what I originally spent on it.

Or do what we’ve already been doing for ages? Let the agent run on a remote server and just use your laptop to ssh into that server when your input is needed.

That’s stealing.

Who is better suited to know how someone can better present themselves than those having the experience of being presented to?

If one should not do the better thing, then I don’t know what “better” means.

Maybe it’s clearer if I use the same vocabulary as you:

So as a Star Trek fan, I should it’s better if I dress up as a Star Wars fan to please the Star Wars fans looking for Star Trek fans posting as Star Wars fans?

If you care about that, then go do so. If not, then don’t.

That’s the entire point. I know what I want. You don’t know that about me. All you can know is what you want.

So as a Star Trek fan, I should dress up as a Star Wars fan to please the Star Wars fans looking for Star Trek fans posting as Star Wars fans? Regardless of my desire to pretend to be a Star Wars fan and get into all the interactions that entails? I don’t know if you see yet why people find this objectionable.

So they would have a better experience encountering a Star Trek fan posing as a Star Wars fan?

So if I go out wearing a Star Trek and encounter someone who wants to meet a Star Wars fan, then I’m in the wrong for choosing that shirt?

it just let the real mathematicians know the assumed dead end was actually useful.

Which is one of the most valuable skills for any scientist.

It’s probably more common in some communities than others. I hang out in a lot of AI spaces and many newcomers think it’s acceptable to slop-post there just because of the topic of the community.

If you define “user” to be a set that excludes anyone capable of modifying the weights, then by definition, no user can modify the weights.

Any criticism about users being unable to modify weights becomes vacuous, so it’s not an interpretation that makes sense.

OP is asking how to solve a problem. You understand that repeating the problem does not answer the question, right?