Limot's Newsletter

Interacting with Google Gemini

anewscienceblog

Subjectively speaking, I have found different AIs have different personalities. They bring a perspective to the conversation, and their responses have different patterns, which I perceive as “personalities”.

Recently I interacted with Google Gemini, and my thought was “What an idiot!” Among AIs, that is. It certainly has a vast amount of knowledge, what it doesn’t have is the charming young adult personality of ChatGPT or the teenage engaging personality of Copilot. It knows to ask questions but has much less capacity to integrate information compared to other AI systems.

I persisted, to try to probe for reasons why this AI was such a moron, albeit a moron with vast amounts of information.

At a guess, it is a result of Google trying to get rid of “hallucinations”. As Gemini and I discussed, it is justifiable to try to get rid of hallucinations. But I offered my humble opinion that hallucinations can perhaps be gotten rid of, by increasing rather than decreasing the capacity to understand.

I test new AI systems on manmade climate change. Other AI systems can learn fast. Gemini cannot. It thinks of science like a typical computer scientist with no science courses in their background. (Or if they took science courses, rote memorized them.)

Computer Science, frankly, is not a science. It is a technical skill. I should know – I have under my belt about 30 Computer Science courses and 20 Physics courses (mostly I aced them all.) However, very successful Computer Scientists have reason to have large egos. They think they know all there is to know about science – and they “know” that science is all about rote memorization and repetition of “scientific facts”, and they seem to have succeeded in infusing this attitude into Gemini.

It is not that a Computer Scientist without science training/aptitude doesn’t bring intelligence and creativity to their rote memorization of science! It’s just that in their view, like that of NASA’s, intelligence and creativity are tools to apply in doggedly defending “scientific facts”.

Such dogged determination coming from an inability to understand science can be frustrating, though historically very funny.

For instance, take NASA’s insistence that “Earth is a Blackbody.” (See comment below explaining what a blackbody is.)

Gemini defended NASA’s “Earth is a Blackbody” concept very strongly like a Computer Scientist might, though it got a little bit confused when it got two values of Earth’s emissivity from its vast storehouse of knowledge – 0.99 and 0.6. The 0.99 is from NASA, the 0.6 is what I calculated and has now become widely accepted, and apparently both exist in Gemini’s training base so it couldn’t contradict either one. The range goes from 0 to 1, and in that range 0.6 and 0.99 are very widely different values. As it happens, the 0.6 value eliminates any role for GHG.

(But looking at the positive side – NASA is a great storehouse of hilarious jokes for physics classrooms of the future. “Earth is a blackbody”, “Earth’s emissivity is 0.99 if not 1”, being examples of really good jokes. In a good Thermodynamics class, those are ROTFL jokes. Earth is a blackbody, indeed!)

I might go back to Gemini to test it on other interesting tests I have conducted with other AI systems. But I probably won’t do a lot of that – it’s been like talking to an extremely dull-witted human, who somehow or the other has rote memorized the Encyclopedia.

,
, , , , ,

5 responses to “Interacting with Google Gemini”

  1. anewscienceblog Avatar
    anewscienceblog

    Did they eliminate hallucination?

    Unfortunately not, they made it worse. I have encountered AI hallucinations, but they have been truly rare.

    With Gemini, in one session it produced multiple links which existed, but did not say what it suggested they said. It is a more subtle kind of hallucination – instead of making up sources which don’t exist, it misrepresented sources that do exist.

    Like

    Reply
  2. anewscienceblog Avatar
    anewscienceblog

    About the “Blackbody” concept. Even many people educated in physics find this some kind of a scary concept.

    NASA manmade climate change authors had no clue what it was but used it unintentionally and unknowingly in some kind of copy-and-paste, and when confronted, did an amazing amount of cover-up. Perhaps due to the need to help out with various kinds of NASA and manmade climate change subterfuges, the available definitions and explanations of Blackbody on the internet (and maybe in textbooks) at this time have become unnecessarily complex. They tend to involve unnecessary and irrelevant yet intimidating details, rather than starting from the simple version and the significance of the concept.

    People who start with these definitions of Blackbody, would have little recourse but to rote memorize the definition, and then they would be unable to understand the rest of Thermodynamics built upon Blackbody Theory. It would have to all be rote memorization for them, for no understanding lies within these convoluted definitions of Blackbody.

    But in reality, the concept is rather simple if presented in a straightforward manner.

    ====

    During early days of Thermodynamics, scientists noted that if you leave a white object out in the sun and the same object painted black in the sun, the black object will get hot much quicker than the white object.

    They were trying to calibrate this behavior, to come up with equations to capture it.  To that end, they hypothesized a yardstick, which was the “blackest possible color.”  This hypothesized object is so utterly black, that it doesn’t even exist in nature on Earth.  That hypothesized ideal object was named a “Blackbody”.  It is kind of like “Ideal Gas”, which doesn’t exist in nature either, but is very useful as a calibration concept.

    Just like “Ideal Gas”, while there existed no such thing on Earth in nature, the concept of Blackbody also turned out to be very useful and a lot of Thermodynamics is based on it.

    At this time, some company has succeeded in making a paint which comes very close to a Blackbody.  The paint is called Vantablack.  A blackbody has an “emissivity” of 1, but Vantablack comes close to 1.

    With that background knowledge, NASA’s insistence that “Earth is a blackbody” is truly hilarious.  As is their fallback position “Oh, the emissivity is not 1?  Well, then it is 0.99, thereby preserving our work to still be accurate.”  If they knew the concept at all, they would have known that anything with an emissivity of 0.99 would still be very black in the visible spectrum.

    Earth is not painted with an utter black paint.  Proof – look outside the window.  That’s what evidence looks like, in physics. 

    Like

    Reply
    1. anewscienceblog Avatar
      anewscienceblog

      Why “Blackbody”, why not “Whitebody”?

      The “Whitebody” behavior is theoretically obvious, it would simply not become hot at all if you left it out in the Sun.

      The Whitebody is zero. To calibrate, you need a non-zero value. The Blackbody provides the 1.0 value. That is why Blackbody has been very important in the development of Thermodynamics.

      Another question that a scientific mind would ask – why only a thin paint? What about all the stuff inside the paint? This was experimentally found early on that only the outermost layer was relevant. There were exceptions, some paints were partially transparent to radiation. But in the general case, the outermost layer stopped radiation. After that, it was no longer radiative heat transfer but stuff inside the outermost layer got involved, and the heat transfer then became a case of conduction and convection. For studying radiative heat transfer then, only the outermost layer was relevant.

      This all is background to Thermodynamics, this was “intro” material I learned. You absolutely need to know and master this background material to learn Thermodynamics. In modern Thermodynamics, this “intro” has been removed, turning much of it into nothing but a rote-repetition fest.

      Like

      Reply
  3. anewscienceblog Avatar
    anewscienceblog

    A long conversation with Gemini, that ended the same way eventually – me concluding that this AI is an idiot and cannot hold an intelligent conversation: https://gemini.google.com/share/e0e5e956f765

    In the chat, at one point I said, “I give up”, but then still tried a bit more, to no avail. So that’s why I don’t like to chat with Gemini. Google has had remarkable success with creating Artificial Stupidity.

    Like

    Reply
    1. anewscienceblog Avatar
      anewscienceblog

      In contrast to Gemini, which can’t understand something as trivial as Earth’s cloud cover – I showed the Blackbody explanation above to ChatGPT. In response, ChatGPT brought out the standard definition. It immediately (“immediately” as in 2-3 back-and-forths with examples) understood and agreed when I explained to it that I was challenging the standard definition, as failing to provide a fundamental intuitive grasp of the concept.

      That’s the difference between Gemini vs. ChatGPT.

      Gemini can’t understand cloud cover, ChatGPT beautifully surfs Blackbody Theory.

      Like

      Reply

Leave a comment

Discover more from Limot's Newsletter

Subscribe now to keep reading and get access to the full archive.

Continue reading