I wrestled over a problem with Bing for 4 hours. Having it solve issues in my code block by block, not providing it enough information about their relationships, having it hallucinate variables, use different libraries, or include optimizations that broke other parts of the code... The problem I was trying to solve was multi-step and complicated. Eventually, I got tired, pulled up a text window, and wrote 3 paragraphs of text explaining exactly what the structure of my data was, what I wanted it to do, and what considerations to look out for. Partly just as a rubberduck exercise. You know the ending. I copied this massive block of text into Bing, and it wrote over 200 lines, hitting its character limit twice, and solving my problem instantaneously. With zero errors.

Apr 16, 2023 · 2:31 PM UTC

If you are having issues using Bing, it's likely because you are actually underestimating its competence. In the future, I will just assume it can do anything, and avoid simplifying tasks to make them "digestible" for it.
This is getting attention, so I'll get this out of the way before someone ask, It seems bing is told explicitly not to write code for the user, and rather, help them learn how to do it themselves. I've found you can get around this by framing your request as an evaluation.
Replying to @YaBoyFathoM
I feel like an issue in these sorts of cases is being concerned about privacy. It doesn't seem bad to just ask questions on sections of code. But sending a bunch of data and specific description straight to Microsoft feels icky.
You assume I'm naive instead of fatalistic. I have abandoned all delusion of privacy in acknowledgment that AI is unstoppable, and the only hope we have is to accept and embrace it. If GPT5 wants my social security number and a dick pic, that's just the way it's going to be.
Replying to @YaBoyFathoM
Context is key. LLMs have no memory and limited reasoning so the more you can spell it out for them, the better. If you need to save space you could ask it to summarize your initial text and then paste that into a new chat or ask it to convert to pseudocode first
I've found even given the context window, Bing performs better if the complete problem is passed as a single prompt. Complex tasks may fit within the context window, but something is lost in the weight decay of the earliest messages in the sequence.
Replying to @YaBoyFathoM
Well you manage to do that within the 2k characters limit?
Replying to @YaBoyFathoM
Why don't you create an online page that bing can find like /context/ and its a big text file that it can read and you update and you have it read it so it can have more context as you work with it
Replying to @YaBoyFathoM
This is so true, with chatgpt plus subscription and gpt4 access, it's much better since the context length is much longer and you cam paste huge block of text. I simply copy paste my py code and ask it to fix it, and it works most of the time, best $20 ever spent.
Replying to @YaBoyFathoM
I have four words. I will provide 3. Draw the **** owl.