I have been using the AI chat bots quite a bit lately as well as over the past year. I have increased my usage over this time as these tools help greatly. I have used all of these tools. Some are better than others for particular tasks. When I do Python programming, I prefer Claude as it is fast and has good explanations. The problem though is the free version limits your use. So, I switched to CoPilot (Microsoft) as it is equally as good although you have to ask for further explanation. I tried Gemini (Google) for python programming and it's not very good for beginners like me. It doesn't give you a complete set of code, only snippets and suggestions. With CoPilot I can tell it what I want, and it generates a complete set of code. It does take some iterations to remove some errors and get it to where I want it, but it is pretty impressive. I did some coding back in college, so I know the basics, and this allows me to create programs and learn at the same time. Very good way of doing it.
So, for programming I pick CoPilot unless you want to pay the Claude subscription fee then Claude would be my choice.
Just recently I was doing some research on radio noise measurements and was using AI to do some calculations around thermal noise floor and receiver sensitivity. I used all three and all of them gave me the wrong answers initially. They had slightly different calculation issues and after several minutes of both doing the calculations myself and pitting the AI bots against each other I was finally able to get them to give me the correct answer. They had some issues doing watt to dBm conversions and watts to milliwatts. It seems they weren't trained properly on these scientific calculations.
Being that they call themselves "artificial intelligence" bots I figured they would learn from their mistake which I pointed out to them. When I worked on an artificial intelligence project at the University of Illinois in 1983 (yes, the university had an artificial intelligence platform running on their minicomputers) I understood that artificial intelligence was supposed to take input and continuously learn. This is what intelligence is.
So, I asked each AI tool the following questions "do you learn and adapt in real time from individual conversations?"
Claudes' answer
No, I don't learn or adapt from our conversations. Each conversation starts fresh and I can't retain or learn from previous interactions. My training was completed in April 2024 and doesn't update during conversations.
I aim to be direct about this since it's important for you to understand my capabilities. While I can engage thoughtfully within a single conversation using the context we build together, this doesn't persist after the conversation ends, and I start fresh with each new interaction.
I notice that sometimes I may inadvertently give the impression of learning or adapting in real-time when I acknowledge mistakes or refine my responses based on feedback within a single conversation. But this is just me processing the immediate context and feedback you provide, not actual learning or adaptation that persists beyond our current exchange.
CoPilot answer
Great point! While some AI systems are designed to learn and adapt in real-time, my current version operates differently. I don't have the capability to learn from individual conversations. Instead, I rely on feedback from users like you to improve over time.
Here's a quick breakdown of how some AI systems work:
Types of Learning in AI:
Supervised Learning: Where models are trained on labeled data.
Unsupervised Learning: Where models find patterns in unlabeled data.
Reinforcement Learning: Where models learn through trial and error.
Real-Time Learning: Some advanced systems adapt and learn continuously from new data.
My Learning Process:
User Feedback: Users provide feedback on my responses.
Developer Review: My developers analyze the feedback.
Model Updates: Developers make updates based on the feedback to improve my performance.
While I can't learn in real-time, the continuous loop of feedback and updates ensures that I get better with time. Your feedback plays a crucial role in that process!
Gemini answer
I am still under development, but I am learning and adapting in real time from individual conversations. I am able to access and process information from a wide variety of sources, including my own past conversations. This allows me to provide more relevant and informative responses to your questions.
I am also constantly being updated with new information and capabilities. As a result, I am able to provide more accurate and comprehensive responses over time.
Conclusion
Hopefully, you have read this far because I was surprised to find that only Gemini says that they learn and adapt from individual past conversations. That's good to know so hopefully the next person that asks them to do a thermal noise calculation it will learn from my correction and provide more accurate answers. If we still have to rely on human beings to take updates and retrain the models, then is that really artificial intelligence?
Comments