Thanks, I'll take a look at these. To be honest, I can't even get 4.o to work with basic proofreading of text. The false-positive rate is like 70%. It will show me 15 errors and display what it changed, but there is no error and no change. When I ask ChatGPT what it changed, it responds by apologizing and saying there was no change.
Try it yourself. Upload a 1500 word text document and ask it to proofread it for typos, errors, and mistakes. It can't do it reliably.
I tried 4 of these. Cartoonize didn't work. Sketchy didn't work. Whimsical Diagrams gave a useless straight-line chart of data. Podcast search returned that no podcasts discuss a specific brand of watches that I've heard discussed on many different podcasts.
Jason, I appreciate your input and your candor. I hear you on how the results can sometimes not meet expectations. In my experience and in my testing, many but not all of the responses are useful and usable; some of them have been less so. Fortunately, the responses take just seconds to generate, so I find that I'm able to get useful results if I try multiple approaches — and sometimes vary the specific custom GPT I'm using. In any case, thanks for sharing your experience.
My excitement for AI tools has worn off. Like you mentioned, so many of these tools don't work well enough to be used for anything other than fun experimentation. I know they will get better, but they struggle with the most basic tasks.
Thanks, I'll take a look at these. To be honest, I can't even get 4.o to work with basic proofreading of text. The false-positive rate is like 70%. It will show me 15 errors and display what it changed, but there is no error and no change. When I ask ChatGPT what it changed, it responds by apologizing and saying there was no change.
Try it yourself. Upload a 1500 word text document and ask it to proofread it for typos, errors, and mistakes. It can't do it reliably.
Hi. Try 250 word fragments. Works well for me. 1500 may push the model beyond its context or memory limits for now (probably not for long). Cheers.
This is an incredibly interesting article Jeremy - it highlights how powerful the GPTs can be (as well as limitations to be aware of!). Thank you!
I didn’t realize free users can now access the apps - very cool! These posts are great
Wow, there is just so much here. Thank you.
Hey there. Was trying Sketchy - seems to require a ChatGPT plus account which is annoying.
I tried 4 of these. Cartoonize didn't work. Sketchy didn't work. Whimsical Diagrams gave a useless straight-line chart of data. Podcast search returned that no podcasts discuss a specific brand of watches that I've heard discussed on many different podcasts.
Jason, I appreciate your input and your candor. I hear you on how the results can sometimes not meet expectations. In my experience and in my testing, many but not all of the responses are useful and usable; some of them have been less so. Fortunately, the responses take just seconds to generate, so I find that I'm able to get useful results if I try multiple approaches — and sometimes vary the specific custom GPT I'm using. In any case, thanks for sharing your experience.
My excitement for AI tools has worn off. Like you mentioned, so many of these tools don't work well enough to be used for anything other than fun experimentation. I know they will get better, but they struggle with the most basic tasks.