To make it even simpler, he just took their probabilities instead of class labels to determine the "reliability" of predictions. That is, my code already implements partially the logic of conformal predictions, but without probabilities.
Too weak an improvement, often works worse than the original.
Again AI didn't make my life better :) although in the comments he keeps saying that the second approach is the most state-of-the-art and no better has been invented yet
But in terms of coding quality Gemini 2.5 is now definitely better than ChatGPT models and somewhere on Claude's level or higher. Probably this is the most correct and useful conclusion from everything that happened :)
Too weak an improvement, often works worse than the original.
Again AI didn't make my life better :) although in the comments he keeps saying that the second approach is the most state-of-the-art and no better has been invented yet
But in terms of coding quality Gemini 2.5 is now definitely better than ChatGPT models and somewhere on the level of Claude or higher. Probably this is the most correct and useful conclusion from everything that happened :)
I use a Chinese - deepseek. Very often happy, but did not try to improve something of my own. Maybe you can try and post the result?
PS. By the way, Chinese is much superior to alice and chatgpt.
I use a Chinese guy - deepseek. I am very often happy with it, but I have not tried to improve something of my own. Can you try it and post the result?
PS. By the way, Chinese is much superior to alice and chatgpt.
Chinaman is somewhere on the level between chatgpt and Claude/Gemini. Well, or at the level of chatgpt. Sometimes I make him move his transistors. It is Claude and Gemini who are now showing more advanced answers in programming, in my opinion.
Zy- in general water, not independent research, but a kind of analysis of the results of other works (researchers, with.....). Let's say, non-controversial conclusions.... Crossing out previous statements of their own.... In general, the water is wet, the sky is blue.
Allow the use of cookies to log in to the MQL5.com website.
Please enable the necessary setting in your browser, otherwise you will not be able to log in.
You are signed in. Two-factor authentication has been disabled. Set it up again and scan a new QR code, or continue with two-factor authentication disabled.
To make it even simpler, he just took their probabilities instead of class labels to determine the "reliability" of predictions. That is, my code already implements partially the logic of conformal predictions, but without probabilities.
I'll compare the results now.
My code:
His code:
Too weak an improvement, often works worse than the original.
Again AI didn't make my life better :) although in the comments he keeps saying that the second approach is the most state-of-the-art and no better has been invented yet
But in terms of coding quality Gemini 2.5 is now definitely better than ChatGPT models and somewhere on Claude's level or higher. Probably this is the most correct and useful conclusion from everything that happened :)My code:
His code:
Too weak an improvement, often works worse than the original.
Again AI didn't make my life better :) although in the comments he keeps saying that the second approach is the most state-of-the-art and no better has been invented yet
But in terms of coding quality Gemini 2.5 is now definitely better than ChatGPT models and somewhere on the level of Claude or higher. Probably this is the most correct and useful conclusion from everything that happened :)I use a Chinese - deepseek. Very often happy, but did not try to improve something of my own. Maybe you can try and post the result?
PS. By the way, Chinese is much superior to alice and chatgpt.
I use a Chinese guy - deepseek. I am very often happy with it, but I have not tried to improve something of my own. Can you try it and post the result?
PS. By the way, Chinese is much superior to alice and chatgpt.
There is a strong book out
It's not coded, just theory?
No codes, more theories and formulas, no practical proof, and kinda wants to apply, with pictures... All in all, 603 pages of water