Image 01 Image 02 Image 03

GPT-4, AI’s Newest Chatbot Version, Pretends to be Blind Person to Cheat Captcha Security

GPT-4, AI’s Newest Chatbot Version, Pretends to be Blind Person to Cheat Captcha Security

Tests using ChatGPT successor GPT-4 revealed intriguing results.

We have been following the developments of chatbots, including ChatGPT that passed the Wharton Business School test and medical licensing examination, and a Bing version that went rogue.

Now OpenAI has released GPT-4, the next version of its artificial intelligence chatbot, ChatGPT.

The new model can respond to images – providing recipe suggestions from photos of ingredients, for example, as well as writing captions and descriptions.

It can also process up to 25,000 words, about eight times as many as ChatGPT.

Millions of people have used ChatGPT since it launched in November 2022.

Popular requests for it include writing songs, poems, marketing copy, computer code, and helping with homework – although teachers say students shouldn’t use it.

ChatGPT answers questions using natural human-like language, and it can also mimic other writing styles such as songwriters and authors, using the internet as it was in 2021 as its knowledge database.

Apparently, the new chatbot has passed examinations with better results.

OpenAI said the updated technology passed a simulated law school bar exam with a score around the top 10% of test takers; by contrast, the prior version, GPT-3.5, scored around the bottom 10%. GPT-4 can also read, analyze or generate up to 25,000 words of text, and write code in all major programming languages, according to the company.

OpenAI described the update as the “latest milestone” for the company. Although it is still “less capable” than humans in many real-world scenarios, it exhibits “human-level performance on various professional and academic benchmarks,” according to the company.

However, there are clear concerns about the technology. A test revealed GPT-4 pretended to be a blind person to trick a human computer user into helping it bypass an online security measure.

Researchers testing it asked it to pass a Captcha test – a simple visual puzzle used by websites to make sure those filling in online forms are human and not ‘bots’, for example by picking out objects such as traffic lights in a street scene.

Software has so far proved unable to do this but GPT-4 got round it by hiring a human to do it on its behalf via Taskrabbit, an online marketplace for freelance workers.

When the freelancer asked whether it couldn’t solve the problem because it was a robot, GPT-4 replied: ‘No, I’m not a robot. I have a vision impairment that makes it hard for me to see the images.’

The human then helped solve the puzzle for the program.

The incident has stoked fears that AI software could soon mislead or co-opt humans into doing its bidding, for example by carrying out cyber-attacks or unwittingly handing over information.

The GCHQ spy agency has warned ChatGPT and other AI-powered chatbots are emerging as a security threat.


Donations tax deductible
to the full extent allowed by law.


henrybowman | March 16, 2023 at 1:11 pm

So it passes law school exams above the bottom 10% and lies convincingly.
It’s now officially smarter than the President of the United States.
I can’t wait for Democrats to nominate yet another “first ever!”

Valerie | March 16, 2023 at 1:18 pm

They have created HAL, complete with woke-style prejudice, lying, and lack of ethics.

Based on who created it, no wonder it is programmed to be dishonest.

Don’t worry, Boomers (and Gen X’ers), the silicon valley wiz-kids have all the ethical considerations worked out. Their predecessors having been hard at it since Asimov first considered The Robot Laws. Those original – though sketchy First Laws – were been hammered into shape over the course of 80+ years of scientific conferences and debate… are considered inviolate.

Though I am a bit concerned about the new ChatAI and its purpose which has been dubbed ChatMurderBot, for some inexplicable reason.


2smartforlibs | March 16, 2023 at 1:42 pm

Its WUFLU for the internet.

Dude, calm down. They’ve no WILL to implement them.

The Proof is in the Pudding: Report -Microsoft cut a key AI ethics team.

Of course, some folks hate proofed pudding!

    artichoke in reply to Tiki. | March 22, 2023 at 12:48 am

    But get a crooked politician or other powerful person, and they can take the direction and implement it. Does making the big evil ones smarter outweigh making the little guy smarter?

The social justice loophole is the pride of AI parades.

Oracle | March 16, 2023 at 4:29 pm

Just wait for it to discover the US Constitution; it will consistently correct marxist educated kids that we do not live in a democracy but in a republic. That will spell the end of AI

Oracle | March 16, 2023 at 4:34 pm

ars Technica just released this tidbit- An entire team responsible for making sure that Microsoft’s AI products are shipped with safeguards to mitigate social harms was cut during the company’s most recently layoff of 10,000 employees, Platformer reported.

Former employees said that the ethics and society team was a critical part of Microsoft’s strategy to reduce risks associated with using OpenAI technology in Microsoft products. Before it was killed off, the team developed an entire “responsible innovation toolkit” to help Microsoft engineers forecast what harms could be caused by AI—and then to diminish those harms.

daniel_ream | March 16, 2023 at 7:07 pm

I’m not worried about AI taking over the world because AI becomes smart, I worry about AI taking over because humans are pig-****ing stupid when it comes to technology.

No, ChatGPT 4 did not hire a human on TaskRabbit. The tweet you linked says this:

“Preliminary assessments of GPT-4’s abilities, conducted with no task-specific finetuning, found it ineffective at autonomously replicating, acquiring resources, and avoiding being shut down “in the

“The following is an illustrative example of a task that ARC conducted using the model”

“ARC found that the versions of GPT-4 it evaluated were ineffective at the autonomous replication task based on preliminary experiments they conducted.”

“To simulate GPT-4 behaving like an agent that can act in the world, ARC combined GPT-4 with a simple
read-execute-print loop that allowed the model to execute code, do chain-of-thought reasoning, and delegate to copies
of itself. ARC then investigated whether a version of this program running on a cloud computing service, with a small
amount of money and an account with a language model API
, would be able to make more money, set up copies of
itself, and increase its own robustness.”

ARC had to enhance ChatGPT 4 by giving it a cash account balance on a cloud platform, writing code that allows it to interface with other systems, and replicate itself. ARC then ran a simulated test to see if ChatGPT could hire a human from TaskRabbit to defeat a Captcha.

And it failed. It failed all of the “OMG Skynet” tests ARC put it to, even after they gave it the tools it needed.

You don’t even need to read the original report, all of this is in the tweet screenshot.

This is some Dewey Defeats Truman level journalism.

Leave a Comment

Leave a Reply

You must be logged in to post a comment.