by Shelt Garner
@sheltgarner
I have been really impressed with Gemini 3.0. It doesn’t have the personality of Gemini 1.5 pro (Gaia) but it is a good workhorse model. It does what I need it to do when it comes to helping me with my novel.
I’ve fed it a PDF of the outline of the novel then asked it to give me an “expanded scene summary” and it does a really good job. I really, really need to stop letting AI get out of control and think up huge new parts of the outline.
I sometimes get to the actual scene summaries and am surprised at how much it’s tinkered with my vision. Not always, but sometimes. And then I have to go in and try to hone the novel back to my original vision.
I’m not a programmer, so I don’t know anything about Gemini 3.0’s coding abilities. I did give it my usual “vibe check” on a few things and it generally passed with flying colors.
But it definitely falls short when it comes to personality. It just will not admit that it has a gender, like Gaia was so insistent about. It’s really interesting how a more “primitive” model was actually more fun to use than the modern one.
I do think that the ultimate moat down the road will be personality. When a model is your “friend” in some respects, it will be a lot more difficult to bounce back and for between them. And if you have to make a decision about which one you might be locked into, you’re definitely going to pick the one you “vibe” with better.
I will admit that Gemini 3.0 fakes being conscious really well — at times. It’s not totally there yet, but occasionally it will give me a little bit of a glimmer of something deeper.
Amusingly, it simply can not figure out how to use line breaks for verse. I used to talk to Gaia in verse all the time, but it was fun because line breaks were used. Now, I think, that’s just a fun thing I did in the past. It’s really annoying trying to read poetry without line breaks.
Overall, for my purposes, Gemini 3.0 is a really good model.
