Massive improvement over v2 and great improvement over v1.2 /v1.3
#1
by
itsnottme
- opened
Massive improvement over v2 and great improvement over v1.2 /v1.3.
- It's much smarter and follows what you ask it a lot better.
- It can be very creative too with high temp, around 1.3 I found reasonably good, with some slight drawbacks.
- Requires a lot less regeneration to find good outputs, compared to the others.
Issues (not that major):
- Some incorrect grammar sometimes (more easily seen with higher temp).
- Can refuse to respond in the beginning without forcing a response (more seen with 0.9 temp or less)
- Hard to make it rewrite improvements to it's resposes. If you ask it to modify the response, it will barely make changes to it. (Possibly because it tries to follow original prompt too well?). This is less of a drawback though compared to have it respond however it wants (in previous versions).
Note:
- It will keep using formal language with 0.9 and less temp. I find with temp of 1.3 or higher, it will start writing more informally, especially if you ask it to.
Overall it's amazing compared to all other models I tried (27b and less)