Another feedback loop with LLM is that more and more websites will use text generated from a chatbot, and then this content will be fed into the next version of the chatbot.
Maybe there are math tools to determine the influence of feedback loops in processes. Autoregressive curves and generator functions in algebra come to mind, but I don't know yet if these bring any assistance to this subject.
11:30 Neither algorithm did what I thought it would, but are healthcare, education, and sustainability perhaps the three areas with the most potential growth for days engineers?
very revealing. I have had great fun exploring the limits of chatGPT by asking it difficult/cutting edge biochemistry (CISS), to interpret the worldview inherent in 'Transformers' the movie, and dredging up exact references for psychology research in the 1970s. Only the last was fully satisfying.
The pattern I'm particularly interested in sorting out is under what conditions you decide to pronounce "data" as dætə or deıtə, since you used both pronunciations 4 times in the same output haha
lol, that was interesting and funny, particularly how the bots are extroverted and introverted but also bunch of made up skills without the specific input. Ritvikmath, I like the way you think 😆
Actually I was surprised that education was mentioned and not finance or logistics. But heslthcare and sustainability.... spot on. "More Professional" means more made up facts. Awesome experiment. Next step, feed all 10 iterations to both and let them be scored on their professionalism level...
There is nothing wrong with this, nobody keeps putting phrases over and over? You check the output before you use it right? Most important take away is that the prompt we use is most significant, the number of iterations it goes through + how much it is allowed to wander (stochastic randomness) you will get diverging results. This diverging result is what we want since it gives more unique answers. If we want the same fix the seed, set the "conciseness highest".
The end of the intro got a good chuckle out of me. Keep up the great work!
Another feedback loop with LLM is that more and more websites will use text generated from a chatbot, and then this content will be fed into the next version of the chatbot.
Maybe there are math tools to determine the influence of feedback loops in processes. Autoregressive curves and generator functions in algebra come to mind, but I don't know yet if these bring any assistance to this subject.
11:30 Neither algorithm did what I thought it would, but are healthcare, education, and sustainability perhaps the three areas with the most potential growth for days engineers?
very revealing. I have had great fun exploring the limits of chatGPT by asking it difficult/cutting edge biochemistry (CISS), to interpret the worldview inherent in 'Transformers' the movie, and dredging up exact references for psychology research in the 1970s. Only the last was fully satisfying.
3:49 My guess is that after 4 or 5 iterations, ChatgGPT will go into a loop and just cycle through a couple responses.
The pattern I'm particularly interested in sorting out is under what conditions you decide to pronounce "data" as dætə or deıtə, since you used both pronunciations 4 times in the same output haha
😂 I hadn’t even noticed
lol, that was interesting and funny, particularly how the bots are extroverted and introverted but also bunch of made up skills without the specific input.
Ritvikmath, I like the way you think 😆
Actually I was surprised that education was mentioned and not finance or logistics. But heslthcare and sustainability.... spot on.
"More Professional" means more made up facts.
Awesome experiment.
Next step, feed all 10 iterations to both and let them be scored on their professionalism level...
There is nothing wrong with this, nobody keeps putting phrases over and over? You check the output before you use it right? Most important take away is that the prompt we use is most significant, the number of iterations it goes through + how much it is allowed to wander (stochastic randomness) you will get diverging results. This diverging result is what we want since it gives more unique answers. If we want the same fix the seed, set the "conciseness highest".
the lie is one of the indications of intelligence :)
haha! fair