HN Explorer

Tell HN: ChatGPT have added time-consuming CAPTCHAs between every few messages

3 points by cmcaleer ↗ HN
Happens on my Plus account, have been seeing other reports on Twitter. Every 3-5 messages I have to solve 5 puzzles which require on average 20 clicks per challenge (each puzzle takes avg 3 clicks + 1 submit button) before I can continue my question (between browsing pictures and submitting the answer), and if I rush through and get one wrong, it'll tell me at the end and make me do all 5 all over again. Repeat every few messages.

Massively flow-breaking.

9 comments

[ 3.7 ms ] story [ 67.7 ms ] thread
senttoschool
Wow.

It's quite clear that the bottleneck is in Nvidia GPUs being limited in supply. Better buy up Nvidia stock.

3cats-in-a-coat
Wouldn't make sense to give users CAPTCHA rather than just limit their rate more.
version_five
If true it will be the last time I use it. Hardly seems sustainable for them, it's going to destroy interest. Web scale businesses do they best to select for customers they can treat like shit, but this goes too far I suspect. Or maybe they're just trying to get out of the consumer chat business?
3cats-in-a-coat
It'd be odd considering:

1. Plus accounts pay $20+/mo (you can't have bots use it for free)

2. No point proving you're human, you already provided a CC.

I haven't seen this. Needless to say, if I do see it, I'd just stop being a customer and deal with Bing & other services.

But you didn't even provide a screenshot :P ?

cmcaleer
Here's how it looks (using DarkReader)[0], occasionally I get a 'put the person in the seat' question, but it's much the same. I don't think my account is particularly special, I don't hit the rate limit for GPT4 all that often and my IP isn't treated as suspicious anywhere, so I just assumed it was something everyone was getting.

Annoyingly, I'm making a plugin for ChatGPT so I'm now faced with the question of whether I want to continue making a plugin when the dev experience is now straight-up adversarial to me, or whether I want to use an inferior local model which is less reliable for what I want to make.

[0]: https://i.imgur.com/ktBVLds.png

senttoschool
The assumption is that people are buying Plus accounts and then abuse it with a bot - maybe a browser extension.
3cats-in-a-coat
It's like saying I buy a sandwich and then abuse it by cutting it with a knife before i eat it. Why does it matter how one uses his account? It has a very low limit anyway. And now GPT-4 API is public which is supposed to be used through a bot... and it's cheaper.
senttoschool
I don't know. How would you explain their action?
3cats-in-a-coat
I also don't know. I'm not arguing with you, more like thinking out loud if this makes sense for OpenAI to do, and I'm not sure it makes sense TBH. But... this doesn't mean they're NOT doing it. People do stupid things sometimes.