I tried #ChatGPT5 today a little bit and thought I'd share my first impressions. - It is *fast*. I'm not sure if they are temporarily throwing lots of hardware at it for good reviews or if the performance is just better. - I had it edit a photo. The results were horrific. Photoshop jobs are safe.🧵
- I had #ChatGPT5 read half a blog post and generate the second half. Then we worked together revising it. Probably 75% of it's criticism was spot on, but only about 15% of it's fixes worked without editing. Most were pretty bad. - I didn't save time or effort, but it challenged me to write better.🧵
- #ChatGPT5 (unlike o3) is compatible with my custom GPTs and Projects, potentially upgrading them for free. I say potentially because my anemic testing was roughly equivalent other than speed. - It looked impressive on a research task, but I didn't verify and it could be pure hallucination.🧵
- I did not test the #ChatGPT5 Thinking or Pro models yet. - I did not test code generation yet. - I did not test any jail-breaking yet (as I am anti-censorship, I expect to be disappointed by filtering upgrades). - I see the old models are unavailable, making A/B testing more difficult.🧵
Lastly, I see a lot of folks making fun of #ChatGPT5 errors. It's a bit like mocking a foreign student for failing an English test. Y'all are trying to do stuff LLMs *are not good at*. That said, they are funny and it has no feelings to hurt so rock on, but it's not a fair measuring stick. /🧵