'Hypnotised' ChatGPT and Bard will tell users to pay ransoms and drive through red lights Thread poster: Thomas T. Frost
|
|---|
Articles:
Unmasking hypnotized AI: The hidden risks of large language models: the main article, which is a bit technical (cached article, since the si... See more Articles:
Unmasking hypnotized AI: The hidden risks of large language models: the main article, which is a bit technical (cached article, since the site itself will not show it).
'Hypnotized' ChatGPT and Bard Will Convince Users to Pay Ransoms and Drive Through Red Lights.
IBM researchers convinced large language models to play a multi-layered “game” of offering incorrect advice to prove they are “ethical and fair.”: a writer's summary, more accessible.
This has no direct relation to translation, at least not for now, but the articles may still interest some colleagues.
As far as I can see, the scenarios remained isolated to any given conversation and did not involve any contamination of other sessions or users. The question is if this could happen, for example by training the AI model with contaminated data, as the first article suggests.
The discussion of AI bots used by banks is particularly worrying if a vulnerability were to let bank user A transfer money from user B's account by 'hypnotising' the bot. Will banks ensure there are sufficient traces to investigate such hacking? We may think they are responsible enough, but a recent experience makes me doubt it:
Real event: I had two current accounts with a major bank. Account 1 was funded and needed. Account 2 was empty and no longer needed. I looked for a simple function to close the account, but there was none. The only option was the chatbot. I asked it how to close an account. It said I could ask the bot to close it, so that's what I did, ensuring that I asked it to close the empty one.
A few days later, I logged in again. Only the empty account was available. The funded one had disappeared. The bot had closed the wrong account. After a one-hour phone queue, a human was put on the case. They had no bot trace and asked if I could send them the bot screenshots, which I did. They admitted the bot error, reopened the funded account, closed the empty one, paid a handsome compensation and apologised, so it ended well. But what if I had not been able to access the bot screens again (I had not initially saved a screenshot)? And what if someone else had used a bot to hack into my account and nobody had any bot trace? ▲ Collapse
| | | | Lieven Malaise
Belgium
Local time: 22:40
Member (2020)
French to Dutch
+ ...
Thomas T. Frost wrote:
But what if I had not been able to access the bot screens again (I had not initially saved a screenshot)? And what if someone else had used a bot to hack into my account and nobody had any bot trace?
Interesting questions. Especially since my banking app doesn't allow for screenshots to be taken (out of privacy protection reasons, I suppose). So I would have zero proof in your situation.
| | | | Thomas T. Frost 
Portugal
Local time: 21:40
Danish to English
+ ...
TOPIC STARTER
Lieven Malaise wrote: Thomas T. Frost wrote:
But what if I had not been able to access the bot screens again (I had not initially saved a screenshot)? And what if someone else had used a bot to hack into my account and nobody had any bot trace? Interesting questions. Especially since my banking app doesn't allow for screenshots to be taken (out of privacy protection reasons, I suppose). So I would have zero proof in your situation.
That sort of 'security' is plain silly, as you can just take a photo with a phone instead. It is your own data you are viewing, so there is no privacy concern.
About AI logging and tracing, the next question is: if they do trace everything, could a hacker then tell AI to log something else than what is actually happening? With a traditional structured bank login, you can only use predefined functions, which are usually completely secure. But when it's just normal English, it's much more tricky to prevent the bot from doing unauthorised functions.
And what does all this mean for our industry?
| | | | There is no moderator assigned specifically to this forum. To report site rules violations or get help, please contact site staff » 'Hypnotised' ChatGPT and Bard will tell users to pay ransoms and drive through red lights | Trados Business Manager Lite | Create customer quotes and invoices from within Trados Studio
Trados Business Manager Lite helps to simplify and speed up some of the daily tasks, such as invoicing and reporting, associated with running your freelance translation business.
More info » |
| | Protemos translation business management system | Create your account in minutes, and start working! 3-month trial for agencies, and free for freelancers!
The system lets you keep client/vendor database, with contacts and rates, manage projects and assign jobs to vendors, issue invoices, track payments, store and manage project files, generate business reports on turnover profit per client/manager etc.
More info » |
|
| | | | X Sign in to your ProZ.com account... | | | | | |
Login to reply/comment 
Translation news
