AKA I Am a Man!
Leading AI chatbots show dementia-like cognitive decline in tests, raising questions about their future in medicine
Dec 2024, phys.org
Something about anthropomorphism - saying they have dementia, like saying they hallucinate, is assigning human-like qualities to a robot:
Researchers assessed the cognitive abilities of the leading, publicly available LLMs — OpenAI's ChatGPT versions 4 and 4o, Anthropic's Claude 3.5 "Sonnet", and Google's Gemini versions 1 and 1.5 — using the Montreal Cognitive Assessment test, widely used to detect cognitive impairment and early signs of dementia, usually in older adults.ChatGPT 4o achieved the highest score (26 out of 30), followed by ChatGPT 4 and Claude (25), with Gemini 1.0 scoring lowest (16).The uniform failure of all large language models in tasks requiring visual abstraction and executive function highlights a significant area of weakness that could impede their use in clinical settings."Not only are neurologists unlikely to be replaced by large language models any time soon, but our findings suggest that they may soon find themselves treating new, virtual patients - artificial intelligence models presenting with cognitive impairment."
via Department of Neurology at Hadassah Medical Center Jerusalem, Hebrew University, Tel Aviv University: G Koplewitz: Age against the machine—susceptibility of large language models to cognitive impairment: cross sectional analysis, BMJ (2024). DOI: 10.1136/bmj-2024-081948
AI's next frontier: Selling your intentions before you know them
Dec 2024, phys.org
Sure it sounds scary, but isn't this what predictive analytics is all about? (And we've been doing that for years)
Forthcoming - "persuasive technologies" using "digital signals of intent" to predict your behavior in real time via Anthropomorphic AI agents."We caution that AI tools are already being developed to elicit, infer, collect, record, understand, forecast, and ultimately manipulate and commodify human plans and purposes."Again, "We caution that AI tools are already being developed to elicit, infer, collect, record, understand, forecast, and ultimately manipulate and commodify human plans and purposes."
via University of Cambridge's Leverhulme Center for the Future of Intelligence: Beware the Intention Economy: Collection and Commodification of Intent via Large Language Models, Harvard Data Science Review (2024). DOI: 10.1162/99608f92.21e6bbaa
Condé Nast, other news orgs say AI firm stole articles, spit out “hallucinations”
Feb 2025, Ars Technica
In February 2024, [generative AI company] Cohere announced that it would provide legal protection against intellectual property claims to its paying enterprise customers. This includes "full indemnification for any third party claims that the outputs generated by our models infringe on a third party's intellectual property rights," for Cohere "enterprise customers that comply with our guidelines and do not intentionally attempt to generate infringing content." (Note this is not a ruling like the Reuters case but just the beginning of the lawsuit.)
Release the copybot trolls!
New study identifies differences between human and AI-generated text
Feb 2025, phys.org
Just the stats ma'am:
They show how LLMs write by prompting them with extracts of writing from various genres, such as TV scripts and academic articles.LLMs used present participle clauses at two to five times the rate of human text, as demonstrated in this sentence written by GPT-4o: "Bryan, leaning on his agility, dances around the ring, evading Show's heavy blows."They also used nominalizations at 1.5 to two times the rate of humans, and GPT-4o uses the agentless passive voice at half the rate as humans. This suggests that LLMs are trained to write in an informationally dense, noun-heavy style, which limits their ability to mimic other writing styles.The researchers also found that instruction-tuned LLMs have distinctive vocabularies, using some words much more often than humans writing in the same genre. For example, versions of ChatGPT used "camaraderie" and "tapestry" about 150 times more often than humans do, while Llama variants used "unease" 60 to 100 times more often. Both models had strong preferences for "palpable" and "intricate."
Can we just pause for a minute and recognize that we're saying "more often than humans do" as if we knew what universal human speech was like. All this talk about bias in the algorithms, certainly important because it can be amplified, but what about the bias in the base sets? What are the base data we're using to say what a "human" is like?
Also, as a native English speaker, I do recognize that non-natives tend to overuse present participles (like words ending in -ing), which may or may not have anything to do with this and the joke that 'AI is just three [people from underdeveloped communities] in a trenchcoat'. If you're not sure what I'm talking about, go listen to an Excel tutorial for a few minutes. ...
via Carnegie Mellon University: Alex Reinhart et al, Do LLMs write like humans? Variation in grammatical and rhetorical styles, Proceedings of the National Academy of Sciences (2025). DOI: 10.1073/pnas.2422455122
No comments:
Post a Comment