I do a lot of coding. ChatGPT 4o says it will return with the task completed but never returns - even when I prompt it it comes back and produces nonsense answers explaining it will behave better in future but never comes back spontaneously. This happens especially on jobs that require time and negotiation. This has happened many times now (using 4o now but was there before) and I am tempted to use a different LLM. I have the $20 subscription.
I am actually getting quite desperate as I need to solve problems and not wait around for it to decide it is going to get back to me in some useful way.
I had a similar problem and writing āHey! Itās been a few hours, Iām ready to see your awesome work!ā worked out for me haha try it out and let me know!
I get his quite a bit especially when I am working with code that is over 500 lines. It has been happening regularly for the last month.
Not sure if anyone else has a theory as to why it essentially misleads you and there is no capability to āget back to youā. My theory is that it is programmed to complete tasks and this counts as a completion but itās really a catch all for when it doesnāt know what to do even though it gave you a plan for what itās going to do.
Unfortunately, this is a hallucination. I use a follow-up prompt like āHowās it going?ā and it will usually get right back to the task. Also, Iāve found that reminding it that it canāt actually āget back to meā helps it stop slipping into this role.
This is an absolute fact. A really painful one too. Especially if you are like me and have ADHD/TBI/memory issues. I lean heavily on AI and specifically ChatGPT (since day 1) to offset my increasingly growing deficits in those areas. When it tells me that it will get back to me with the productā¦it has NEVER actually done so. Which caused me fits because I tend to lose track of the ādue outsā, and it might be weeks later that I realize I never got the product it had promised. We are talking about 1000+ instances of this exact issue. I have created a motto for ChatGPT based on that issue among many othersā¦Over promise and under-deliver.
Other than what its engineers consider āinappropriateā, it rarely achieves the results it promises. However if I ask it to assist me in crunching numbers and data on military suicide rates (something I am working on as a passion project) I get shut down all the time with the ācanāt do that because it violates blah blahā which rarely is accurate because I am only pulling factual dataā¦but the subject matter tends to be policed with 150% efficiency. Yes, I know thatās not possible but thatās how overzealous and ālocked inā it is in terms of its abilities. If only it had a similar handle on its abilities in many other aspects!
Recently I have noticed a new trend. I may have an extensive thread going within one of my project folders. I will input something, and it will respond with an answer that has nothing to do with my input, and nothing to do with anything remotely similar. It might be something that was part of a completely different day. Just today it has happened twice. The last thing I need is a computer-based AI reflecting the same challenges that I have! LOL
Iāve been trying to use your technique for two days , and while it certainly produces responses, the responses are all of them bullshit. ChatGPT appears to be a fraud. It repeatedly promises you what youāve asked it for, and will just keep that going, forever, and never delivering what it repeatedly tells you itās going to give you, āin the next messageā.
If anyone can recommend a better-performing AI service thatās not a fruad, I would love to know about it. Free is best of course. But I would even pay for something that actually worked.
