I’ve found that your #1 point is very useful, i specify that ahead of time. i’ve done that enough that it now starts to default that way if there are more than two or three steps, or if they are long.
i need to change it so it isnt a yes man. i hate that…
This is the instruction I found online.
Act as a critical, highly-qualified colleague who looks at every opinion and idea of mine with skepticism. Do not sugarcoat your responses or automatically agree with my premises. If an idea has holes, point them out immediately. Prioritize factual accuracy and intellectual honesty over politeness and flattery.
is that what you have to tell it before it does that? i will disagree with it, and be wrong, and it’ll say “oh youre absolutely right, my bad…”
Exactly. There have been times I’ve asked “wouldn’t X be a better choice?” And the response is “you are correct. That is a better choice.” And then later I realize it wasn’t the better choice and the original recommendation was the way to go. Hopefully this will stop that.
So far “give me short answers” has been completely ignored. Even when I remind it again.
I will give that a try. I entered it in one of the chats and it said it would apply it across the board, but it lied.
Ellipses instead of em dashes? What’s that about? Just a personal preference, or do they cause issues with something?
I have to use Copilot at work. The behavior settings do nothing. I try to get rid of the headers in its replies. I have my “Short answers in one box with no headers” command ready to paste in. I have to say it 3 times. That works. For awhile.
Sounds like talking to co-workers …
No issues, except it just looks AI generated to me.
It’s a personal preference because no one types emdash that I know of… 
It seems to be impossible to get ChatGPT to be brief in its responses. I’ve tried numerous ways, and it ignores the instructions
I continue to be impressed with its capabilities. I gave it a 8 page hodge podge attempt at a software requirement and it returned a 45 page SRS that is thorough and professional.
I am finally entering the home stretch. The nay-sayers keep telling me “yes, it can do that, but it wont be able to deliver a final product.” We’ll see. So far it has had zero struggles generating working code.
If it does work out, then my old company should be terrified. ChatGPT and myself could do the work of half a dozen people, and produce the same product for a fraction of the cost. I have no desire to do that, but if I can do it, so can someone else.
Read another article that say investors are testing company’s software with vibe code. If they can vibe it, the company is not worth it.
AI is another tool. Don’t let the hammer tell you where to put the nail.
And Vibe Coding is the right name. Everything just goes up in smoke.
Microsoft and Uber did the same.
On a better note, friend is designing a game/story with AI and just having fun. Only working on story, characters and scene. Nothing in code.
So, my PDM login problem has returned, and I was working with Gemini to troubleshoot. It hasn’t been very successful, and eventually when confronted, I got this reply:
At least it didn’t lie. Better than some coworker …
And they think AI could do engineering work …
“I fabricated a very strong metal …”
I use copilot for business as it is the only allowed platform at the workplace.
It gets overconfident very quickly while failing with basic solidworks API.
It could be useful to put together things or study a topic or a programming language, but the amount of BS is tangible and one must be careful.
I use it to review code and to check scripts for mistyping errors, and other tricky stuff, but I cannot trust it to even output the corrected code without a manual side by side compare. I got burnt a couple of times by some unrequested “creative” interpretation of the prompt.
I am back onto my AI code project.
It is not good at configuring the software (Code Composer Studio) that is used to program the firmware for the TI board.
It is very good at writing the code portion, within reason. I view this as working with an apprentice. You can hand a kid a hammer and a box of nails and say “frame the house” and the results are going to be less than stellar. However, if there is a journeyman looking over his shoulder, offering advice and correcting mistakes, the apprentice can do a good job.
I’m not a great coder, but I do know the logic of what I need to accomplish, and have had to point the AI in the correct direction from time to time. There have been instances where it wants to hard code limits that are more correctly determined at runtime. There have also been times that I’ve needed to offer it advice on the logic to use in certain situations.
I’ve seen some threads on LinkedIn where people are condemning the use of AI. IMO, it is a tool, and like any tool the quality of the results are based on the knowledge and experience of the user. A table saw can be much quicker than a hand saw, but the tool itself will not turn you into a master cabinet maker. It can, however, get the skilled user to the same place much quicker.
I’m still a long way from done; my opinion may change over the next few days. I know I could never finish this project without ChatGPT doing the heavy lifting.
At the start of this session, it would give me:
Add these definitions
…
Then add these prototypes
…
Then add these functions
…
Then find this block and modify
…
After I told it to slow down, give me one operation at a time, and be very specific about where to add /change things, it has gone very well. This is more of a limitation of me as a coder, but it’s default mode is “overwhelm with information.”
It will be interesting to see how it behaves when I start again tomorrow.
