given that 50 weekly prompts limit for o1, your o1 and 4o strategy makes a lot of sense.
I recently moved from Claude 3.5 Sonnet to Gemini 2.0 Experimental Advanced, a reasoning model, and talking to it like a chatbot has been fine. I have alternated between Gemini 2.0 Flash and Gemini 2.0 Exp. but that ended up the bifurcating my workflow which was a little annoying.
Are you paying for the Gemini Advanced plan to get access to the 2.0 Experimental Advanced reasoning model? I've been pretty happy simply using Gemini Experimental 1206 in the Google AI Studio up to now.
Yes paying for the advanced plan. The extremely context window for Gemini is a huge perk as it reduces the anxiety of a chat prematurely ending due to hitting that token limit.
For sure! It's also great at retrieval (the whole "needle in a haystack" thing). I recently fed Gemini in AI Studio my entire history of posts for the past 2+ years (over 500K tokens) and it was able to answer specific questions very accurately.
Dang, this is almost something I could have written myself (though I would lose patience toward the end and rush the conclusion, then just ask the readers to weigh in and save the piece).
Seriously, this is exactly how I've started thinking about o1 and 4o (they are really trying to give physicists a run for that "worst at naming things" title, aren't they?). 4o and related LLMs are designed for conversations (or at least that's what they are best at), whereas you climb up to the top of the mountain to ask the guru a question once a year or whatever (o1). That's the right way to frame these.
I have not yet tried using a prompt designed from 4 for 1, but will make a conscious effort to try this out for tougher questions.
given that 50 weekly prompts limit for o1, your o1 and 4o strategy makes a lot of sense.
I recently moved from Claude 3.5 Sonnet to Gemini 2.0 Experimental Advanced, a reasoning model, and talking to it like a chatbot has been fine. I have alternated between Gemini 2.0 Flash and Gemini 2.0 Exp. but that ended up the bifurcating my workflow which was a little annoying.
Nice!
Are you paying for the Gemini Advanced plan to get access to the 2.0 Experimental Advanced reasoning model? I've been pretty happy simply using Gemini Experimental 1206 in the Google AI Studio up to now.
Yes paying for the advanced plan. The extremely context window for Gemini is a huge perk as it reduces the anxiety of a chat prematurely ending due to hitting that token limit.
For sure! It's also great at retrieval (the whole "needle in a haystack" thing). I recently fed Gemini in AI Studio my entire history of posts for the past 2+ years (over 500K tokens) and it was able to answer specific questions very accurately.
Dang, this is almost something I could have written myself (though I would lose patience toward the end and rush the conclusion, then just ask the readers to weigh in and save the piece).
Seriously, this is exactly how I've started thinking about o1 and 4o (they are really trying to give physicists a run for that "worst at naming things" title, aren't they?). 4o and related LLMs are designed for conversations (or at least that's what they are best at), whereas you climb up to the top of the mountain to ask the guru a question once a year or whatever (o1). That's the right way to frame these.
I have not yet tried using a prompt designed from 4 for 1, but will make a conscious effort to try this out for tougher questions.
I love the "Guru on the mountain" analogy, too!
Which reminds me: https://www.youtube.com/watch?v=Qf3xwmXXDH8
Well, that was a big bust.