Quote from: edzieba on 05/16/2025 01:43 pmQuote from: Star One on 05/16/2025 11:35 amQuote from: sanman on 05/16/2025 08:49 amQuote from: Star One on 05/15/2025 06:58 pmQuote Arvind Narayanan at Princeton University says that the issue goes beyond hallucination. Models also sometimes make other mistakes, such as drawing upon unreliable sources or using outdated information. And simply throwing more training data and computing power at AI hasn’t necessarily helped.The upshot is, we may have to live with error-prone AI. Narayanan said in a social media post that it may be best in some cases to only use such models for tasks when fact-checking the AI answer would still be faster than doing the research yourself. But the best move may be to completely avoid relying on AI chatbots to provide factual information, says Bender.https://www.newscientist.com/article/2479545-ai-hallucinations-are-getting-worse-and-theyre-here-to-stay/So they're vulnerable to the same mistakes as human minds.(I think we're currently debating some of these things in our Entertainment board threads.)I cannot help wondering if LLMs are a dead end for achieving AGI. You certainly would question one making critical decisions in human spaceflight.A statistical sentence-completer (what an LLM fundamentally is) trained on massive volumes of internet text will asymptotically approach the output of an average internet text writer. If you can ask a question to the youtube comments section and get a useful answer, an LLM is a great fit for that task (e.g. "how many cats are in this photo?") - albeit probably a computationally inefficient one - but the further you stray from that the more likely you are to get nonsense rather than a useful answer. Even if you perform supplementary training (e.g. feed it physics textbooks) that supplementary text is so vastly smaller in volume than the main training set that you'll get youtube-commentor output bleeding through without warning. You can't just compensate by weighting your supplementary training higher, as that only works for responses that can match closely to the training set, and anything that strays from that will then be more likely to produce garbage output (and the problem gets worse as you increase the weight).I think it goes back to has been discussed on here that true AGI needs to be via a physical body that disembodied AI cannot achieve that leap.
Quote from: Star One on 05/16/2025 11:35 amQuote from: sanman on 05/16/2025 08:49 amQuote from: Star One on 05/15/2025 06:58 pmQuote Arvind Narayanan at Princeton University says that the issue goes beyond hallucination. Models also sometimes make other mistakes, such as drawing upon unreliable sources or using outdated information. And simply throwing more training data and computing power at AI hasn’t necessarily helped.The upshot is, we may have to live with error-prone AI. Narayanan said in a social media post that it may be best in some cases to only use such models for tasks when fact-checking the AI answer would still be faster than doing the research yourself. But the best move may be to completely avoid relying on AI chatbots to provide factual information, says Bender.https://www.newscientist.com/article/2479545-ai-hallucinations-are-getting-worse-and-theyre-here-to-stay/So they're vulnerable to the same mistakes as human minds.(I think we're currently debating some of these things in our Entertainment board threads.)I cannot help wondering if LLMs are a dead end for achieving AGI. You certainly would question one making critical decisions in human spaceflight.A statistical sentence-completer (what an LLM fundamentally is) trained on massive volumes of internet text will asymptotically approach the output of an average internet text writer. If you can ask a question to the youtube comments section and get a useful answer, an LLM is a great fit for that task (e.g. "how many cats are in this photo?") - albeit probably a computationally inefficient one - but the further you stray from that the more likely you are to get nonsense rather than a useful answer. Even if you perform supplementary training (e.g. feed it physics textbooks) that supplementary text is so vastly smaller in volume than the main training set that you'll get youtube-commentor output bleeding through without warning. You can't just compensate by weighting your supplementary training higher, as that only works for responses that can match closely to the training set, and anything that strays from that will then be more likely to produce garbage output (and the problem gets worse as you increase the weight).
Quote from: sanman on 05/16/2025 08:49 amQuote from: Star One on 05/15/2025 06:58 pmQuote Arvind Narayanan at Princeton University says that the issue goes beyond hallucination. Models also sometimes make other mistakes, such as drawing upon unreliable sources or using outdated information. And simply throwing more training data and computing power at AI hasn’t necessarily helped.The upshot is, we may have to live with error-prone AI. Narayanan said in a social media post that it may be best in some cases to only use such models for tasks when fact-checking the AI answer would still be faster than doing the research yourself. But the best move may be to completely avoid relying on AI chatbots to provide factual information, says Bender.https://www.newscientist.com/article/2479545-ai-hallucinations-are-getting-worse-and-theyre-here-to-stay/So they're vulnerable to the same mistakes as human minds.(I think we're currently debating some of these things in our Entertainment board threads.)I cannot help wondering if LLMs are a dead end for achieving AGI. You certainly would question one making critical decisions in human spaceflight.
Quote from: Star One on 05/15/2025 06:58 pmQuote Arvind Narayanan at Princeton University says that the issue goes beyond hallucination. Models also sometimes make other mistakes, such as drawing upon unreliable sources or using outdated information. And simply throwing more training data and computing power at AI hasn’t necessarily helped.The upshot is, we may have to live with error-prone AI. Narayanan said in a social media post that it may be best in some cases to only use such models for tasks when fact-checking the AI answer would still be faster than doing the research yourself. But the best move may be to completely avoid relying on AI chatbots to provide factual information, says Bender.https://www.newscientist.com/article/2479545-ai-hallucinations-are-getting-worse-and-theyre-here-to-stay/So they're vulnerable to the same mistakes as human minds.(I think we're currently debating some of these things in our Entertainment board threads.)
Quote Arvind Narayanan at Princeton University says that the issue goes beyond hallucination. Models also sometimes make other mistakes, such as drawing upon unreliable sources or using outdated information. And simply throwing more training data and computing power at AI hasn’t necessarily helped.The upshot is, we may have to live with error-prone AI. Narayanan said in a social media post that it may be best in some cases to only use such models for tasks when fact-checking the AI answer would still be faster than doing the research yourself. But the best move may be to completely avoid relying on AI chatbots to provide factual information, says Bender.https://www.newscientist.com/article/2479545-ai-hallucinations-are-getting-worse-and-theyre-here-to-stay/
Arvind Narayanan at Princeton University says that the issue goes beyond hallucination. Models also sometimes make other mistakes, such as drawing upon unreliable sources or using outdated information. And simply throwing more training data and computing power at AI hasn’t necessarily helped.The upshot is, we may have to live with error-prone AI. Narayanan said in a social media post that it may be best in some cases to only use such models for tasks when fact-checking the AI answer would still be faster than doing the research yourself. But the best move may be to completely avoid relying on AI chatbots to provide factual information, says Bender.
If you think LLMs have some deeper generalization capabilities beyond just fancy autocomplete then it might help to give it a physical body, but of course if you view LLMs that way then you probably didn't buy edzieba's argument anyway.
1) 'Embodiment' does not need to be physical, it can be entirely virtual (e.g. a maze-solver in a virtual maze that only receives information about the walls local to itself, rather than being provided the entirety of the maze layout and its exact coordinates within it, would constitute 'embodiment').2) Embodiment will not affect how an LLM functions.
I view this from the opposite direction: LLMs demonstrate that many behaviours we assume to be complex (or 'deeper') are in reality a lot simpler than we assume - 'emergent behaviour' is not just a property of swarm computation, but any highly parallel computation. Rather than evaluating LLMs as 'smarter' than assumed, we may need to evaluate some human (and higher animal) behaviours as being simpler than assumed. Than again, I come from a Cybernetics background, so a system-of-systems being capable of more complexity than the sum of its individual systems seems a natural way to observe things.
Quote from: edzieba on 05/17/2025 07:48 am1) 'Embodiment' does not need to be physical, it can be entirely virtual (e.g. a maze-solver in a virtual maze that only receives information about the walls local to itself, rather than being provided the entirety of the maze layout and its exact coordinates within it, would constitute 'embodiment').2) Embodiment will not affect how an LLM functions.Okay, but how do you most reliably ensure that without having the AI "living life" out in the "wider world" (experiencing that wider world entirely thru its local perspective, of course)?The biggest sample of the universe is the universe itself.Quote from: edzieba on 05/17/2025 07:48 amI view this from the opposite direction: LLMs demonstrate that many behaviours we assume to be complex (or 'deeper') are in reality a lot simpler than we assume - 'emergent behaviour' is not just a property of swarm computation, but any highly parallel computation. Rather than evaluating LLMs as 'smarter' than assumed, we may need to evaluate some human (and higher animal) behaviours as being simpler than assumed. Than again, I come from a Cybernetics background, so a system-of-systems being capable of more complexity than the sum of its individual systems seems a natural way to observe things.Fair enough, even our biologically evolved systems are like this. I guess that's why our relatively small number of genes have been able to produce such complex systems like ourselves.
"Luminous [algorithms] are we, not this crude matter."
The world is awash in predictions of when the singularity will occur or when artificial general intelligence (AGI) will arrive. Some experts predict it will never happen, while others are marking their calendars for 2026.A new macro analysis of surveys over the past 15 years shows where scientists and industry experts stand on the question and how their predictions have changed over time, especially after the arrival of large language models like ChatGPT.Although predictions vary across a span of almost a half-century, most agree than AGI will arrive before the end of the 21st century.
Researchers at Apple have released an eyebrow-raising paper that throws cold water on the "reasoning" capabilities of the latest, most powerful large language models.In the paper, a team of machine learning experts makes the case that the AI industry is grossly overstating the ability of its top AI models, including OpenAI's o3, Anthropic's Claude 3.7, and Google's Gemini.In particular, the researchers assail the claims of companies like OpenAI that their most advanced models can now "reason" — a supposed capability that the Sam Altman-led company has increasingly leaned on over the past year for marketing purposes — which the Apple team characterizes as merely an "illusion of thinking."
because apparently the term 'UFO' has been retired? Was this done for some sort of make-work reason, so that more people can be hired to to update all the documentation?)
AI might be better able to make that determination based on some quantifiable assessment, having trained on lots of footage that could have included instances of bug-on-lens or other conventionally explainable phenomena.
QuoteResearchers at Apple have released an eyebrow-raising paper that throws cold water on the "reasoning" capabilities of the latest, most powerful large language models.In the paper, a team of machine learning experts makes the case that the AI industry is grossly overstating the ability of its top AI models, including OpenAI's o3, Anthropic's Claude 3.7, and Google's Gemini.In particular, the researchers assail the claims of companies like OpenAI that their most advanced models can now "reason" — a supposed capability that the Sam Altman-led company has increasingly leaned on over the past year for marketing purposes — which the Apple team characterizes as merely an "illusion of thinking."Paper is linked to in the article.https://futurism.com/apple-damning-paper-ai-reasoning
Quote from: sanman on 06/11/2025 10:34 ambecause apparently the term 'UFO' has been retired? Was this done for some sort of make-work reason, so that more people can be hired to to update all the documentation?)During the start of the current UFO flap, someone thought the term "UFO" carried too much baggage that prevented people (especially military people) from talking about their experiences. Essentially, we've come to use the term "UFO" as a synonym for "Alien space ship". So they invented a term that was meant to be less pre-determined about what they saw (just as "UFO" was meant to be when it was invented by Capt. Ruppelt.)
I'm fairly sure CSICOP introduced "UAP" in the 1980's to replace "UFO".
It’s official. Using artificial intelligence tools makes us stupid.That’s according to researchers at the Massachusetts Institute of Technology, who have found using popular chatbots like OpenAI’s ChatGPT can “erode critical thinking and problem-solving skills”.
