AI fails

[briantrumpet]

I think it's pure maths, not applied maths.

I applied myself to pure maths, but it didn't do me much good either.

 

[Psamathe]

Quite scary about what AI Chatbots are capable of and have done. Podcast 30 mins but I found it interesting.
BBC The Global Story: The AI chatbot users falling into delusional spirals https://www.bbc.com/audio/play/w3ct8m8p

...
In rare cases though, AI chatbots have sent users down a dark path. In a new BBC investigation, population correspondent Stephanie Hegarty speaks to people who have experienced delusions after talking to chatbots – including one man who grabbed a hammer and prepared for war after his chatbot told him it was sentient.

 

[The Crofted Crest]

£1m also gets treated as £1 metre (or rather meter as it's US).

 

[briantrumpet]

Quite scary about what AI Chatbots are capable of and have done. Podcast 30 mins but I found it interesting.
BBC The Global Story: The AI chatbot users falling into delusional spirals https://www.bbc.com/audio/play/w3ct8m8p

Also including Richard Dawkins losing his mind in response to flattery.

 

[briantrumpet]

Oops, only by a factor of five! How did that happen?

I think we probably know why.

 

[Psamathe]

Oops, only by a factor of five! How did that happen?

I think we probably know why.

View attachment 14999

Strange coincidence that these types of error are always in the direction of favouring the company making the "error". Never overestimating damage.

 

[briantrumpet]

£1m also gets treated as £1 metre (or rather meter as it's US).

I've switched to using £1mn, £1bn etc, for consistency and clarity, though no idea how AI would interpret that.. metre newtons?

 

[briantrumpet]

That's it. If they are doing this to mums, it's time to close the whole lot down. Bastards.

 

[briantrumpet]

 

[Pblakeney]

View attachment 15024

Tbf, it isn't incorrect.

 

[briantrumpet]

Tbf, it isn't incorrect.

A different kind of fail, I suppose. One of those moral questions such as asking whether it is ever justified to tell a lie.

 

[Pblakeney]

A different kind of fail, I suppose. One of those moral questions such as asking whether it is ever justified to tell a lie.

That is going to get complicated as two people will always have a difference on morals.
Just look at any thread on tax.

 

[Psamathe]

A different kind of fail, I suppose. One of those moral questions such as asking whether it is ever justified to tell a lie.

Whilst AI LLMs seem to have no problem lying, facing such questions they are not obligated to answer.

 

[briantrumpet]

Not exactly the kind of 'fail' I was thinking of, but I might be a little cheesed off if suddenly an AI centre took my electricity supply.

 

[Pblakeney]

Similar news from the US of AI becoming lord and master.
“

• OpenAI has announced plans to integrate ChatGPT with users’ bank accounts through Plaid, a financial data network that securely connects apps to banking information. The goal is to allow ChatGPT to analyze users’ financial data and deliver more personalized, curated financial advice. By accessing transaction history, balances, and spending patterns, the assistant could potentially help with budgeting, savings strategies, and broader financial planning. OpenAI says the connection would be secure, though the announcement raises ongoing questions about privacy, data protection, and how sensitive financial information would be handled.”