I've only used the free Gemini feature on Google, and in intensive questioning, it does very often get things wrong. I observe it seems to lack the quality of a good human scientist of demurring or remaining circumspect when there is insufficient evidence to justify an accurate conclusion. It does know enough to ask questions, but it is much more like a poor scientist, jumping to conclusions that haven't been properly tested. Its flaw seems to be that it is programmed to be helpful and eager to please. It is a product after all. I find it likable but over-enthusiastic and highly fallible.
You have to use it a lot to understand the quirks, but when you know how to prompt it well, I find it gives more accurate and thorough responses than I'd get anywhere else
No doubt the agents that cost money to use are many times more capable, but the default agent on Google, as a mundane example, came to the erroneous conclusion one of my wine refrigerators is a dual zone model because of the two relays on the circuit board. While it had quite correctly and astutely identified the relays as the problem and had also helped me remove the circuit board, the truth is my wine refrigerator is a single zone model. I had given it all the relevant numbers related to model and parts, but it did not realize this manufacturer uses the same circuit board in both dual and single zone models to save money because the cost of an extra relay is trivial compared to the price of sourcing and inventorying two different circuit boards.
What else? Among other subjects, I've discussed my changes in thyroid hormone levels vs changes in medication levels and asked it to project the magnitude of changes in those hormone levels going forward given new changes in medication levels. Some of its projections are simply ludicrous. It really needs to be more like a good human scientist and say, "It's not clear - we will need to see what the next data point is, but your basic strategy seems reasonable." But it can't do that, it's hell bent on being helpful and providing a real answer. It just can't help itself.
As an aside, it's become clear to me that my own understanding of the appropriate use of thyroid medication in hypothyroidism to generate a healthy thyroid system and patient is superior to that of my endocrinologist, but she has not read the same ten or so books I have. Neither has the damned AI agent.
Yes I'm pretty sure ChatGPT or Claude Pro would correctly solved your refrigerator issue (they're pretty good with that). I usually get an answer simply by taking a photo of the serial number plate. Similarly with thyroid or health topics, it'll give much better answers if it has lots more context, such as if you uploaded all your medical records. You'd probably need the paid versions to allow that.
I think you're likely right about the superiority of those paid models. I learned today that AI agents (sort of) read actual books as I do, this during their training process. Not necessarily eBooks, that due to copyright restrictions, but genuine books, which may be de-spined and then scanned into the thing. The value having been destroyed, the book remains are then pulped. As an avid reader, do I view that as a desecration? Hardly, books are just another product.
If it's a very rare, old book, I might have more of a problem.
I've only used the free Gemini feature on Google, and in intensive questioning, it does very often get things wrong. I observe it seems to lack the quality of a good human scientist of demurring or remaining circumspect when there is insufficient evidence to justify an accurate conclusion. It does know enough to ask questions, but it is much more like a poor scientist, jumping to conclusions that haven't been properly tested. Its flaw seems to be that it is programmed to be helpful and eager to please. It is a product after all. I find it likable but over-enthusiastic and highly fallible.
You have to use it a lot to understand the quirks, but when you know how to prompt it well, I find it gives more accurate and thorough responses than I'd get anywhere else
No doubt the agents that cost money to use are many times more capable, but the default agent on Google, as a mundane example, came to the erroneous conclusion one of my wine refrigerators is a dual zone model because of the two relays on the circuit board. While it had quite correctly and astutely identified the relays as the problem and had also helped me remove the circuit board, the truth is my wine refrigerator is a single zone model. I had given it all the relevant numbers related to model and parts, but it did not realize this manufacturer uses the same circuit board in both dual and single zone models to save money because the cost of an extra relay is trivial compared to the price of sourcing and inventorying two different circuit boards.
What else? Among other subjects, I've discussed my changes in thyroid hormone levels vs changes in medication levels and asked it to project the magnitude of changes in those hormone levels going forward given new changes in medication levels. Some of its projections are simply ludicrous. It really needs to be more like a good human scientist and say, "It's not clear - we will need to see what the next data point is, but your basic strategy seems reasonable." But it can't do that, it's hell bent on being helpful and providing a real answer. It just can't help itself.
As an aside, it's become clear to me that my own understanding of the appropriate use of thyroid medication in hypothyroidism to generate a healthy thyroid system and patient is superior to that of my endocrinologist, but she has not read the same ten or so books I have. Neither has the damned AI agent.
Yes I'm pretty sure ChatGPT or Claude Pro would correctly solved your refrigerator issue (they're pretty good with that). I usually get an answer simply by taking a photo of the serial number plate. Similarly with thyroid or health topics, it'll give much better answers if it has lots more context, such as if you uploaded all your medical records. You'd probably need the paid versions to allow that.
I think you're likely right about the superiority of those paid models. I learned today that AI agents (sort of) read actual books as I do, this during their training process. Not necessarily eBooks, that due to copyright restrictions, but genuine books, which may be de-spined and then scanned into the thing. The value having been destroyed, the book remains are then pulped. As an avid reader, do I view that as a desecration? Hardly, books are just another product.
If it's a very rare, old book, I might have more of a problem.