Thoughts on current chatty AI
-
The reports of totally made up answers that are so confident and well-written have me thinking. These AI seem to have no concept of the difference between fact and fiction. Being asked to produce copy is the same thing to them as being asked for factual content. They regularly plagiarize, so taking bits from the questions and bits from other writings, they assemble responses as if they were just copy, even when asked to answer with simple facts, whole snips of historical documents/scientific studies, or calculations. Perhaps what is needed is a sort-of 'scholar:' tag. So when asking for answers, it won't make things up.
Cpichols wrote:
These AI seem to have no concept of the difference between fact and fiction.
That is totally incorrect -- "AI" have no concept of anything. There is no "intelligence" in "artificial intelligence". In simplest terms, any "AI" is just a huge, nested if-then-else. When programming, it's up to the programmer to ensure that an if-then-else is testing the correct things, and is testing them correctly. If there is any point of failure, the results will be wrong at least some of the time. Machine learning works by feeding it massive amounts of data, and later indicating which is correct and which is not. As has been pointed out, it gets better with training. The problem is that it will never be 100% correct, yet people are already trusting these systems as being so. There is no discrimination, just a lot of tests that must be correct, yet can't be.
-
Cpichols wrote:
These AI seem to have no concept of the difference between fact and fiction.
That is totally incorrect -- "AI" have no concept of anything. There is no "intelligence" in "artificial intelligence". In simplest terms, any "AI" is just a huge, nested if-then-else. When programming, it's up to the programmer to ensure that an if-then-else is testing the correct things, and is testing them correctly. If there is any point of failure, the results will be wrong at least some of the time. Machine learning works by feeding it massive amounts of data, and later indicating which is correct and which is not. As has been pointed out, it gets better with training. The problem is that it will never be 100% correct, yet people are already trusting these systems as being so. There is no discrimination, just a lot of tests that must be correct, yet can't be.
The Lounge[^] This is the point right here. They can't be trusted. Specifically, they can't be trusted to "discern" (if/else or otherwise come to a conclusion about) the difference between fact sources (historical documents, scientific studies, current events) and fictional ones.
-
Cpichols wrote:
These AI seem to have no concept of the difference between fact and fiction.
The key word in the name is "Artificial". Anyone with half a brain knows that these machines have nothing anywhere close to intelligence.
Can you pass the bar exam?
-
Common sense and baseline IQ will tell anyone that AI and AI "Chat" is still very new and is in constant development and progression. Eventually, some day (soon?) it will be perfected. To judge it now is premature at best.
-
Common sense and baseline IQ will tell anyone that AI and AI "Chat" is still very new and is in constant development and progression. Eventually, some day (soon?) it will be perfected. To judge it now is premature at best.
Yes, it is new. And it will change (significantly?) over time. The problem, though, is an old one: people are lazy. Given a new tool to help them do a job, they'll quickly use it to *do* the job, without oversight or a critical eye. So what difference does it make if you're getting misinformation from an automated device, or a politician, a newscaster, or your neighbor Paul. If you don't take time to verify, then what does it matter.
Time is the differentiation of eternity devised by man to measure the passage of human events. - Manly P. Hall Mark Just another cog in the wheel
-
Common sense and baseline IQ will tell anyone that AI and AI "Chat" is still very new and is in constant development and progression. Eventually, some day (soon?) it will be perfected. To judge it now is premature at best.
They started AI research in the 50s. Lisp is a programming language created about 1960 specifically for doing research in AI. ChatGPT comes from OpenAI which was started in 2015 as a non-profit specifically to use existing (prior) knowledge of AI to research it. ChatGPT is actually the third generation of something based on what they were working on. So not really sure how any of this counts as "very new".
-
Can you pass the bar exam?
Myself I haven't tried. But in quite a few place in the US one does not need to be a lawyer to be a judge. One does not need to go to law school to take the bar exam. There are practicing lawyers that have not passed the bar exam. A lawyer was sanctioned because they submitted case law extracted from ChatGPT which was entirely made up. Myself I can drive a car in the snow and maneuver around construction cones without running into the side of truck. Hopefully you can do the same. Hopefully you do not rely on your self driving car to do that.
-
Common sense and baseline IQ will tell anyone that AI and AI "Chat" is still very new and is in constant development and progression. Eventually, some day (soon?) it will be perfected. To judge it now is premature at best.
-
Common sense and baseline IQ will tell anyone that AI and AI "Chat" is still very new and is in constant development and progression. Eventually, some day (soon?) it will be perfected. To judge it now is premature at best.
There's no intelligence in an LLM. It strings phrases together partially based on the frequency of those phrases appearing together in its training data. Here's an in-depth description: https://writings.stephenwolfram.com/2023/02/what-is-chatgpt-doing-and-why-does-it-work/. Yeah, they will add heuristics around it to make it seem less dumb, but how far is that going to take it? Have you seen the ChatGPT response where it's supposed to subtract 3 from 70, and it confidently answers "73"? And here is Donald Knuth's experience, Donald Knuth Asked ChatGPT 20 Questions. What Did We Learn? - The New Stack[^].
-
Can you pass the bar exam?
Can lawyers write specifications? designs? computer programs?
Freedom is the freedom to say that two plus two make four. If that is granted, all else follows. -- 6079 Smith W.
-
There's no intelligence in an LLM. It strings phrases together partially based on the frequency of those phrases appearing together in its training data. Here's an in-depth description: https://writings.stephenwolfram.com/2023/02/what-is-chatgpt-doing-and-why-does-it-work/. Yeah, they will add heuristics around it to make it seem less dumb, but how far is that going to take it? Have you seen the ChatGPT response where it's supposed to subtract 3 from 70, and it confidently answers "73"? And here is Donald Knuth's experience, Donald Knuth Asked ChatGPT 20 Questions. What Did We Learn? - The New Stack[^].
-
Maybe I can ask the AI for docs on what it means. Oh snap :laugh: I would also love to just give an AI a plot and ask it to make a game of it in the style of Witcher 3 :)
-
Al Fargnoli wrote:
strings phrases together partially based on the frequency of those phrases
Which describes many click bait sites also. So are they intelligent?
Almost as "intelligent" as a LLM-based chat bot!