compare to chat gpt 4o, not even close. i think it even worse than 4o mini. other AI like claude 3.5 sonnet, o1-mini,preview did pretty good at coding, why dont gemini?
3 Likes
Bc Gemini’s main feature is a big context window (2 million tokens for Pro). And Google entered the LLM race much later than Openai and Anthropic. Openai were the first in this field in general, no wonder they are still the leaders and they have sota model o1. Though honestly I think o1 is just finetune on long thought data, no change in architecture, just training costs but not development costs. I assume google is also now either thinking or already working on a similar finetune.
2 Likes
By the way, if you need to perform some task related to the code, then you can clearly instruct the model, system instructions are needed for this. In the Gemini web version (not in ai studio), some kind of system prompt is also written first. And if you need the behavior of the model close to o1, then I made a prompt that makes the model think the same way o1 does. Someone wrote that the results have really improved, heh.
Here is the prompt
1 Like