What hasn't changed since Sonnet 3.5:
- It can do small functions ok
- It can't do anything else
Give it a small problem with ~3 constraints and it will write code that looks sort of ok but upon inspection the constraints are not properly maintained throughout.
I haven't seen any improvement from sonnet 3.5 to sonnet 4.5
If anything, the model got worse. Does not respect my system prompt as much. Really likes to babble even when I tell it to keep it short.
All I know is that GPT-5-codex in Codex cli and I have been able to solve and complete development of projects that stranded from hitting the capability limits of the tools available just a few months ago.
This was created by someone typing a paragraph into an app on their cell phone. Tell me again how AI isn’t rapidly advancing faster than any of us can even begin to realize?
I was talking about LLMs
My guess is that video generation is also going to platue similarly, if it hasn't already.
The clip you posted is literally slop, and with current techniques it probably will not turn from slop to not-slop
Ngl, they likely aren't. If I know a system very well I can ask ChatGPT to do the tedious work of putting my thoughts into code.
I then review it, accept/reject it, and stitch it with the code it made it write earlier. Continue until completed.
I have a feeling that Sonnet 3.5 was much more useful than 4.5 but I can't really prove it because they don't provide access to 3.5 anymore .. otherwise I would have like to compare responses