Search Results
8/8/2025, 5:14:28 AM
>>106185661
I thought you were trolling for a sec but I'm eating good with V3 0324. I'm just testing GLM 4.5. One thing I do to test new models is that I just hit enter with a blank prompt and see what I get out of it with some swipes. You can tell right away how fried they are if you're getting STEM QA pairs too frequently, if not all the time and not general web scrapes.
I know DS V3 was good the moment I saw it generate random web scrapes that weren't limited to STEM topics from a blank prompt.
GLM 4.5 I don't know, because I have to give it like 4 tokens of raw text to start, otherwise it just generates BOS indefinitely.
Pic was some shitty fanfic I wrote with V3.
I thought you were trolling for a sec but I'm eating good with V3 0324. I'm just testing GLM 4.5. One thing I do to test new models is that I just hit enter with a blank prompt and see what I get out of it with some swipes. You can tell right away how fried they are if you're getting STEM QA pairs too frequently, if not all the time and not general web scrapes.
I know DS V3 was good the moment I saw it generate random web scrapes that weren't limited to STEM topics from a blank prompt.
GLM 4.5 I don't know, because I have to give it like 4 tokens of raw text to start, otherwise it just generates BOS indefinitely.
Pic was some shitty fanfic I wrote with V3.
Page 1