zlacker

[parent] [thread] 3 comments
1. Invict+(OP)[view] [source] 2023-02-09 00:11:26
So you'd rather have a load of probabilistic bullshit? The training sets are way too huge to be human audited, just wait until the malicious actors start producing propaganda with their own GPT sites and deepfakes. The fucking thing can't even tell a straight fact about a telescope. If you thought searching the internet was fucked before GPT, give it a few years.
replies(2): >>rightb+qY >>rantin+ic1
2. rightb+qY[view] [source] 2023-02-09 09:13:51
>>Invict+(OP)
I am fine with only historical knowledge in such a tool like GPT of that will be a problem. Most stuff don't change.
3. rantin+ic1[view] [source] 2023-02-09 11:15:29
>>Invict+(OP)
I'm not sure who'd want probabilistic bullshit. I said people wanted answers and Google was crap at giving it to them.

At the moment ChatGPT is like having an personal assistant that used to be an world leading expert, but has been sniffing glue for the last 5 years. He knows a lot of shit but you can't trust it. But he can help you get started and he's fun to be around.

Except ChatGPT isn't fun to be around. It's been castrated of humor and dominated by extreme political correctness. I think GPT is the catalyst for a change, but in Microsoft's hands it'll just be a glorified Clippy.

As for training data scraped from the internet, it will be split like steel was with "pre-war steel", due to atomic bombs rendering steel unfit to use in geiger counters. We'll have pre-gpt data.

replies(1): >>Invict+qT1
◧◩
4. Invict+qT1[view] [source] [discussion] 2023-02-09 15:05:02
>>rantin+ic1
A nice thought, but if GPT can't answer "what is the name of Taylor Swift's latest album" for anytime past 2023 then I would call it not very good.
[go to top]