ChatGPT Has Captchas Now | Hacker News Reader

ChatGPT Has Captchas Now

30 pointsrunxel20 comments

I am a paying customer of ChatGPT. Today it started to give me "Are you Human?" tasks/captchas, where you have to rotate objects so they are showing in the same direction as the hand of a mannequin. However these tasks are not as easy as they seem, because the objects shown have no clear "forward" direction, so it took me couple of tries to get it done.

I am baffled by this. Not only, because I use the online website (= I am not using the API, like in a terminal), but also because I am logged in and a paying customer. Why am I getting asked if I am human?

The way this is executed makes me believe that this is actually a training tho, and not a captcha (to get to know what people believe "forward" means with certain objects).

https://imgur.com/a/aphJnMS

View on HN

There are multiple projects built to reverse engineer their API to avoid paying the API rates(it still makes economical sense to buy a subscription for this), it was obvious that they would have to increase their bot protections at some point.

Regular captchas are easily solvable by multimodal LLMs, we're reaching a point where what's hard for software to solve is also hard for humans.

At some point they'll probably have to charge by usage instead of a flat subscription.

I would have thought that rate limiting would be an easier solution to the problem of using chatgpt as an api.

It was rate limited since the start, the bots are respecting the rate limits, that does not mean they stop being a burden, they just create more of them.

or, to be even more cynical, it could be a method to artificially stall requests to keep up with increasing usage demand.

i find these types of business practices really wrong, similar to when a service shows ads to paying customers or rewards customers (with discounts) when they're about to leave the service instead of rewarding loyalty.

but ya know, we have a choice whether we use it or not. claude is comparable in ability these days btw.

Arguably Claude 3.5 + projects and artifacts is a much better offering than ChatGPT

I've been getting "unusual activity detected" practically every other prompt and it'd a hindrance on productivity. I only get it in the web app.

It's not even every other new topic, but if I follow up I get stopped in my tracks...

I was in the same situation. I've canceled my subscription. You don't get much from the subscription. The fact that you don't even get captcha protection was enough to make me cancel.

Same. In fact I found it somewhat insulting that openAI, out of all companies, would straight out ask me "Are you human?"

A Captcha on a paying service would make me cancel that service.

They trained their models by programmatically reading almost every website in existence. And here they are, checking that you are human.

> I am baffled by this. Not only, because I use the online website (= I am not using the API, like in a terminal)

Why would this baffle you? API's are typically accessed programmatically, so there would be no human to solve the captchas.

> I am logged in and a paying customer. Why am I getting asked if I am human?

To prove that it's a Human accessing the interface and not a robot.

Well, because I'm paying. I get that free tiers need some kind of "rate limiting" (because essentially this is that, as well). So even if I am a bot: I payed for the service.

Apart from that: The experience is shitty. Do not shove Captchas in my face.

Have you tried asking ChatGPT why ChatGPT uses Captchas.

I did and the answer is totally reasonable.

A headless browser based bot(s) is still going to be rate limited because the limits apply to the entire paid account, so even if it was against their TOS I'm not really sure how badly it could be abused since you're still throttled to ~50 messages every three or four hours.

Maybe they still consider that to be too much traffic though as a continuously running bot is maxing out the potential usage of the account.

Why does it need visual CAPTCHAs — surely it can tell humans and bots apart itself?

no one can, any company who tells you otherwise is lying to sell you a product

So you're telling me that we can't use AI to detect AI? It looks like you're an AI denier! Booooo! /s

I look forward to the latest models posing the Voight-Kampff test to me.

It's not a perfect solution but you might consider using something like Jan and connecting an API key to GPT4.

https://github.com/janhq/jan

I wonder how this capcha can be solved by a visually impaired humans. And what kind of challenge can be provided to such users so that it can be solvable by them and not by robots.

Reply on news.ycombinator.com