How could Artificial Intelligence become dangerous?

[JasperFrank]

LLM models see cat pics and pornography. Its no wonder that they hallucinate once and awhile.

 

[rontaki]

LLM models see cat pics and pornography. Its no wonder that they hallucinate once and awhile.

Wait... aren't both of those describing the same thing?

 

[CalG]

If ai is used to "synthesize documents" that would be a bad thing.
Think about "photo image check cashing"

 

[kenmbz]

UK government findings

• Microsoft Copilot trial: A trial of Microsoft 365 Copilot by the UK government's Department for Business and Trade (DBT) found no discernible productivity gain, despite high staff satisfaction. The study concluded that any time saved on simple tasks was effectively cancelled out by extra work required to fix lower-quality AI outputs, such as for data analysis in Excel.

 

[kenmbz]

How AI can go so very, very wrong:

AI Overview

Raiders of the Lost Ark takes place in
1936,
the same year it was released in some parts of the world. It follows the events of Indiana Jones and the Temple of Doom, which is set one year earlier in 1935, and precedes Indiana Jones and the Last Crusade, set in 1938

Imagine that the movie was released in 1936 ?

 

[MossRoad]

It was a 'talkie'

 

[JeepHead]

I work for a major university and have been told by my leadership that I have to go through all of this training to become knowledgeable on supporting AI. For the past two days I have done nothing but either do training modules or sit in meetings with faculty who are experts in AI. It's been interesting to say the least. I have learned that AI will lie (not just give inaccurate info but instead actually make things up) which is called hallucinations. I have also learned how to force it into giving more credible information by not making things up. Also been learning a lot about role prompting which makes if even more useful. They give me challenges that I have to do and then submit for them to review. It's yet to be seen how they are gonna feel about all my challenges and data I have submitted pertains to firearms and ballistics

 

[kenmbz]

When I encounter AI, I try to use Alien and food prompts to give it something bizarre to work with.
It even tries to figure out what they eat based on having 0 knowledge.

 

[MossRoad]

When I encounter AI, I try to use Alien and food prompts to give it something bizarre to work with.
It even tries to figure out what they eat based on having 0 knowledge.

Remember Twilight Zone episode To Serve Man?

 

[BackRoad]

News release today about hackers using AI (Claude) to probe and identify corporations with weak cyber security, then used Claude again to hack and steal data, and then used AI again to review the stolen data for high profit information...either by holding the data ransom, or using the data for gain.

We're standing on a cliff edge...and it's crumbling...