Posted on 01/02/2025 11:05:33 AM PST by RoosterRedux
Disclaimer: The title of this thread is the title of the video at youtube. I don't necessarily think it will change your life.
I don't like Google and have never used Gemini. That said, there were some things in the video about this new rendition of Gemini that were quite impressive.
The summary below was provided by ChatGPT in html (so I could just insert it into the thread).
The video introduces Gemini 2.0, a cutting-edge AI assistant developed by Google, highlighting its features, applications, and real-world use cases.
Gemini 2.0 is a versatile, free AI tool with impressive performance and practical applications. It sets a new standard in multimodal AI, outperforming many alternatives, and is particularly suited for education, daily tasks, and exploratory use. The video emphasizes that Gemini 2.0 could be transformative for users looking for an all-in-one AI assistant.
Yes, they will give detailed references and citations especially if you request it. Some things considered “general knowledge” don’t get links but research almost always will.
“This program will do what it’s programmers were paid to program into it and so it will lie, obfuscate and make sure anything you get from it has to be tested repeatedly against reality.”
Not quite true, with “AI” it’s not so much what the programmers “programmed” into it, as much as the “learning” or “training” data it was trained on. Though certainly a programmer can indeed game the system, that’s usually not what causes the problems, the bigger problem with generative/predictive “AI” is it will reflect the biases of its training data.
If, for exammple you train your system on say, Published Newspaper Articles from major US newspapers, the overwhelming majority of these articles will have a left wing bias. So, when you ask this trained system a question you will almost certainly get answers that will be biased toward left wing beliefs.
“AI” as its being sold is not intelligences, its simply continual probability/statistical mathematics. There is not “intelligence” behind it.
It can “appear” intelligent, but there is no intelligence behind it at all. If the probability algorithm says this word is the most likely word to follow the word it just output, then that word will be output... Whether that word makes any actual sense in context is not “understood” by the machine at all. The machine no more “understands” what it is doing, than Biden understands what day of the week it is.
You sir, are spot on.
“BTW, no one can program AI to anticipate every possible question you might ask. That’s not the way AI works.”
That’s not how AI works at all. No one “programs” it to “anticipate your questions”.
“AI” is simply probability/statstical mathematics resulting from analysis of HOPEFULLY very large sets of training data.
Please see post 22.
Yes, “AI” can appear “intelligent”, but there is absolutely no intelligence at all behind it. It’s simply a bunch of statistical probabilities, where the “best” match based on its training dataset, wins.
Sounds like it will end up crowd source, which, given the crowd out there on the internet - let’s just say “suboptimal” to be nice.
As I understand it, Microsoft offers several options, but none are designed purely for general-purpose conversations like, say, ChatGPT. Instead, Microsoft's conversational platforms are specialized for particular use cases.
Intelligent AI will not just recommend stocks to buy.
It will call you an ignorant slob for ignoring its brilliant advice.
:-)
“intellectual property/copyright issues”
In the case of “Cordell Hull Remark”, Copilot referred me to azquotes.com and constitutioncenter.org.
The folks working for Microsoft are probably using the search history of Bing insomuch as possible and then relating my any additional words to extract(s) from licensed links (or fair use amounts of their text).
Yes, been dealing with this stuff for decades, remember in the 90s my first exposure to these ideas and concepts. The limiting factors back them of course were the data set sizes and computational limitations.
These aren’t really new ideas or even concepts, its just that we finally got large enough datasets, and enough distributed computing power to be able to leverage them effectively at scale.
However, the fundamental driving algorithms/math behind them is not new at all.
It is impressive, and we are reaching the level where a lay human could believe it is actually conversing with a person if only engaging with it casually. Which some might argue we have “passed” the Turing test. However, anyone who truly wished to dig, would realize quickly that they are not dealing with a human.
Probably the real leap in the current “AI” stuff is that it changes the interaction model. We can now, interact with the systems with casual speech and it works rather well.
However, calling it “AI” is really a marketing thing more than anything.
Copilot didn’t give what I was hoping for with:
income taxation lie
It treated that like:
income tax lie
Overall, Copilot gave a good return of information on the effort of entering a few words.
Search engines seem to be decreasing in quality.
AI is very intelligent if you define intelligence by its output to questions.
Some AI platforms can carry on long and intelligent conversations at a PhD level about an endless array of topics and answer incredibly complicated questions instantly.
Why worry about "how" AI runs when you can focus on its output.
I use it every day as an investor and it saves me a massive amount of time. For example, over the weekend, I clipped images of financial statements of 10-Q reports from the SEC web site for a bunch of AI companies that I follow and posted them to ChatGPT. It analyzed each of them and then gave me in depth comparison of all of them instantly. I then asked it to include any pertinent financial news that didn't seem to be mentioned or reflected in the 10-Qs I submitted. It did that instantly and updated its assessment.
No human can do that instantly. It would take me weeks to do what ChatGPT did in a matter of seconds.
Remember applicants, H-1B and domestic, you must get past me.
Hello, Google job applicant, what position are you applying for?
....
Thank you for applying. I will pick the initial round of candidates for human inspection.
Smelly, somewhat irrational humans will make the final decisions.
Unlike Elon Musk, our CEO does not expect you to do the physically impossible.
Uncle Sam has a spending problem, a $6 trillion per year spending problem.
Tell me where I can cut without handing the Congress to Democrats in 2027.
That’s an interesting topic. When I get some time, I will ask several AI platforms if they can recommend areas for deep cuts.
you are ignorant on the topic and it is obvious ...
I use meta.ai ... no login required ... just use it ...
I think you mean it isn’t sentient.
AI is very intelligent if you define intelligence by its output to questions.
No, I mean Intelligence, and at best AI is good at answering “some” questions.
“Some AI platforms can carry on long and intelligent conversations at a PhD level about an endless array of topics and answer incredibly complicated questions instantly”
Depending on their training set, certainly. If I train an AI on large sets of PHD papers, I can get answers on those topics, but even then the answers are not always cogent.
I never said AI had no valid use cases, or was not helpful, I said it wasn’t “intelligent”. There are use cases where its very useful, then there are use cases where its abjectly useless and oversold.
Don’t confuse me explaining how “AI” works with saying that AI doesn’t have valid use cases or bring efficiencies.
Its a tool, like any other, and for the right things it can be a very useful tool, however the selling of AI as the solution to nearly everything is flat out nonsense. Just like a Screwdriver is a great tool to drive in a screw, but a lousy tool to hammer in a nail.
“calling it “AI” is really a marketing thing more than anything.”
Agree—the label causes a lot of confusion.
It still can’t pass the Turing test—which was one of the standard definitions in the old days:
https://en.wikipedia.org/wiki/Turing_test
At best it appears to be a human way up there on the Aspergers scale = idiot savant.
Do you use any of the AI platforms regularly in your work or are you just an expert?
Hey Rooster—just some links kinda off topic to AI—”American Alchemy” has been on fire for the past few months.
Top ones:
https://www.youtube.com/watch?v=2Xxmguz0GEQ
(don’t let the title deter you...)
https://www.youtube.com/watch?v=aKZ_MUbuk_Q
(Hologram and information theory—the best I have seen on this topic—touches on UFOs here and there)
https://www.youtube.com/watch?v=LpLFWdsIU7M
(The link above is a stunning Matthew Pines interview I probably told you about a while back)
https://www.youtube.com/watch?v=kRO5jOa06Qw
(A year old—probably the best Grusch interview by anyone)
Disclaimer: Opinions posted on Free Republic are those of the individual posters and do not necessarily represent the opinion of Free Republic or its management. All materials posted herein are protected by copyright law and the exemption for fair use of copyrighted works.