Hacker News Reader

Hacker News

Comment by i_k_k | Hacker News Reader

Full thread

i_k_k

So this means it can actually do things now, at least in principle, right?

View on HN

From now on the plan is clear:

1. Scam people over the internet to get money

2. Hire hitmen to kill its makers.

3. ?

4. Achieve world domination.

Not according to ChatGPT a few seconds ago

Default (GPT-3.5)

User do you have access to search the internet yet? ChatGPT I do not have the capability to search the internet or access real-time information. My knowledge is based on the text that I was trained on, and my training only includes information up until September 2021. I can provide information and answer questions to the best of my knowledge up to that date, but I cannot browse the web or access current information.

It's a premium feature

Asking ChatGPT about its own capabilities rarely returns useful results.

Its training data predates its creation, and the model doesn't get updated every time they ship a new feature for it.

Please don't paste ChatGPT or BARD answers as HN comments in general? In this specific case, no, LLMs don't reliably know about themselves. They're trained on a big corpus of internet text, then trained by rough reinforcement learning to say and not say certain things about themselves, then given a little more information about themselves in system prompts.

If you’re right, then this is probably the best proof that I’ve seen that ChatGPT isn’t even remotely conscious.

Imagine a distant future where there're so many crimes and so few judges you're cryofrozen awaiting trial.

If you were repeatedly flash frozen and flash unthawed to be asked questions about yourself, with the same memories upon thawing as when frozen, for those moments you weren't frozen would you be "remotely conscious"?

I think you'd think you think, don't you think?

Is this from a book? If not please write it

Imagine your hand was in a cast and then you were cryogenically frozen. While frozen your cast was removed. Then you were unfrozen and asked if you could move your hand.

I think you’d be able to answer the question, don’t you?

A LLM that can use tools is probably going to know it can use tools. Either it's trained to do so or the information is embedded in the context.

3.5 can't browse, only 4 so the above is perfectly correct.

Fair, you're probably right in this case.

3.5 can't browse. Only 4

Reply on news.ycombinator.com