Furthermore,
Ai becoming more more crazy::
As it progresses, AI is becoming more and more crazy. Meanwhile, She is often wrong, and sometimes she invents things. Nevertheless, So, can we do something there?
If you’ve already tested an AI, you probably know what we’re going to talk about. Similarly, Sometimes she gives answers next to the plate. Furthermore, The worst part is that this phenomenon increases even though technology evolves at high speed. In addition,
Yes, the new models are more powerful, more effective. Similarly, But also, they are strangely unstable. Furthermore, So much so that some experts talk about crazy AI. However, So, can we still prevent this innovation from losing your head?
IA who goes crazy: is this a problem? – Ai becoming more more crazy:
It is no longer a simple impression. Similarly, Openai has conducted his own research, and the results are final. In addition, Its latest models, ai becoming more more crazy: O3 and O4-Mini, produce hallucinations in 33 % at 48 % of cases. For example, Twice as much as the old O1 model.
We are talking about Hallucinations of AI when it invents facts or events. And she does it with a fluid and logical tone. As if everything was true. Obviously, this situation raises concerns about the accuracy and reliability of large languages of languages.
But is this really a problem? From a certain point of view, these ” hallucinations »Open interesting doors. They fuel creativityallow AI to invent instead of simply copying reality. Sometimes that’s exactly what we expect from her.
Except that everything changes when it comes to the facts. A crazy AI can be a problem ai becoming more more crazy: When it comes to providing reliable information. “” This is particularly problematic in the fields where decisions depend on factual precision. such as medicine, law or finance “Said Eleanor Waston, member of the Institute of Electrical and Electronics Engineers (IEEE) and Ethics engineer from AI to singularity University.
The more the AI gains in power, the more its answers become difficult to question. For what ? Because they are often well turned. They seem logical and convincing. Thus, we do not always realize that there is an error.
Reduce AI hallucinations: Mission Impossible? Not quite
Certainly, completely eliminating AI hallucinations seems complicated. Watson also confirmed it in an interview with Live Science. According to her, this phenomenon may well be an integral part of the current operation of the models. Same story with Sohrob Kazerounian, another researcher in the estate at Vectra AI.
This does not mean, ai becoming more more crazy: however, that we can do nothing. On the contrary, we can limit the damage. Watson recommends the generation increased by recovery. The principle is that instead of letting AI improvise its answers. it is pushed to rely on reliable and already existing sources.
You can also push AI to Show a little more self-control. Invite her to check her own results herself. Offer him to compare different perspectives or unroll reasoning step by step before concluding. Yes, it takes time. But it’s worth it to avoid mistakes.
Finally, Kazerounian advises interacting with AI with the ” same skepticism that we reserve for our human counterparts ». Anyway, a human verification remains essential To ensure that the answers are well ai becoming more more crazy: founded.
- Share the article:
Our blog is powered by readers. When you buy via links on our site, we can receive an affiliation commission.
Further reading: Good plan-The Hi-Fi (passive) Focal Aria Evo XN ° 1 “5” 5-star “speaker at € 999.99 (-23%) – Can Vantablack painting make the satellites invisible? – decades of reduced spatial research – Antarctic ice sends cosmic messages that challenge physics – The first steps of Shady Elnahas judoka in the anteroom of WWE.
