April 15, 2025
The GIST Editors' notes
This text has been reviewed in line with Science X's editorial course of and insurance policies. Editors have highlighted the next attributes whereas making certain the content material's credibility:
fact-checked
preprint
trusted supply
proofread
Examine cracks the code behind why AI behaves because it does
AI fashions like ChatGPT have amazed the world with their potential to put in writing poetry, clear up equations and even cross medical exams. However they’ll additionally churn out dangerous content material, or promote disinformation.
In a brand new examine, George Washington College researchers have used physics to dissect and clarify the eye mechanism on the core of AI programs. The analysis is revealed on the arXiv preprint server.
Researchers Neil Johnson and Frank Yingjie Huo appeared into why AI repeats itself, why it typically makes issues up and the place dangerous or biased content material comes from, even when the enter appears harmless.
The researchers discovered that the eye mechanism on the coronary heart of those programs behaves like two spinning tops working collectively to ship a response. AI's responses are formed not simply by the enter, however by how the enter interacts with every little thing the AI has ever realized.
This evaluation might result in options that might make AI safer, extra reliable and immune to manipulation.
Extra data: Frank Yingjie Huo et al, Capturing AI's Consideration: Physics of Repetition, Hallucination, Bias and Past, arXiv (2025). DOI: 10.48550/arxiv.2504.04600
Journal data: arXiv Offered by George Washington College Quotation: Examine cracks the code behind why AI behaves because it does (2025, April 15) retrieved 15 April 2025 from https://techxplore.com/information/2025-04-code-ai.html This doc is topic to copyright. Other than any honest dealing for the aim of personal examine or analysis, no half could also be reproduced with out the written permission. The content material is offered for data functions solely.
Discover additional
Novel 'pruning' method reveals promise for decreasing AI bias with out harming efficiency shares
Feedback to editors
