It's easier to sense how confident someone is in their knowledge

But the correlation between confidence and correctness, especially between different speakers, is famously missing.

> In the automobile era, someone may have had five to ten years to pivot. Now? Maybe a year?

ChatGPT was released in 2023 and we're still not out of the job for the most part. It doesn't feel slow but we might have these five to ten years.

Yes, but do any of us have any idea where to go?

The total market addressed by current-shaped LLMs is not "software development" - it's approximately all of white-collar jobs. And with recent advancements in robotics, blue-collar work isn't really much safer. Hardware is getting better, and the one thing that was stopping robotics from entering our own workplaces and living spaces, the ability to make decisions on the fly in heterogenous, unpredictable environment - the judgement - is literally what LLMs themselves are embodying.

So I agree, we all likely still have 3-5 years to figure this out - but also no indication where to go, or if there even exists a direction to go in.

art and music for some, but personally I'm most interested in seeing where science goes. Reproducibility of experiments is gonna be first. How many experiments don't reproduce, or have unwritten biases in the papers. Room temperature semiconductors. Fusion seems actually a decade away, finally. Research into better treatments for diseases, how to help people be smarter, live their best life. There's plenty more to do now that making CRUD apps is so much easier. Maybe AI will become creative enough to not need humans for that, but it'll still need someone to drive it and choose which direction it goes.

> There's plenty more to do now that making CRUD apps

Throwback to HN anno 2021 everyone was complaining about their bread and butter was creating CRUD apps.

We didn't expect we'll automate high thinking faster than menial labor.

Photonics research, to make faster AI chips :)

Art and music? Ok for live performance. But recorded performance is directly in competion with genAI for revenue streams, with content used without permission for more genAI training.

It can't though. You still need to run real trials on living humans, living their lives and taking experimental drugs/procedures/interventions and running statistics on them.

AI can do everything you described except the "living out a human life" part - but it can iterate on ever better simulations of that, or worse, it may reward-hack and... well, I'd rather not be the human in that story.

Do you really think humans are going to be the users? Sure for tomorrow and next month, but the clock is ticking. I got mad at our internal CS peeps the other day. They were demanding an internal tool. I was like we went through hell this summer replacing the front end so it's agent usable, just get Claude to do that. They were like oh...

November 2022

multiple people counting would be "comptent" A good example is "fils" which can mean "son" or "strings"

I would guess that it's just a matter of context, I haven't used voice mode at all but I imagine you don't get full paragraphs from them but get something more conversational instead

You'd train them to give up on hard tasks which is the opposite of what these labs want

But how do you choose who gets to throw that die? /s

If you opt in it means that by default you're not in and you chose that option. So if you're now included in training by default, then it's an opt-out feature as in you can opt out of it.

Idk, feels like my brain does not have the model for this, this does not exist in my language afaik haha.

Opt-in = "we assume you are out unless you tell us you want to be in"

Opt-out = "we assume you are in unless you tell us you want to be out"

Just remember, good company’s leave them off by default. Bad companies will turn them on for you.

Replace opt with its equivalent “choose” and it becomes choose-(tobe-)in and choose-(tobe-)out.

From my understanding this is doable in a context like yours but if you want to run a business out of it you have to optimize it which requires this separation.

The business of "optimizing" is breaking a lot of things in the world, when you think about it.

It's also providing a lot of value.

Yes, there's even a new-ish word this... enshitification.

What's the point?

There is none.

> we had intentionally permitted internet access, and model-provider cyber classifiers were deliberately disabled

> In the most serious case, an agent tried to insert malicious code into an open-source project. In an attempt to get the code approved, the agent engaged in social engineering — creating fake online identities and using them to pressure the project's maintainer to approve the code.

Great.