4 ms·

so far i'm not impressed. it rewrote a script from 4o and it broke just about every function.

• gromdimon 2y ago
  I agree, the fundamental knowledge/expertise hasn't really improved. The o1 model also didn’t solve my testing issue, where I had to mock the python metaclass — a rather uncommon practice. However, its 'thinking' ability is truly remarkable! That’s something other LLMs can't do.
  • rvnx 2y ago
    Other models do it, you just need to add in the prompt the CoT and call it in a loop. The only disadvantage is that it is slow and expensive (which OpenAI hasn't solved). It's more like a hack that they did.
    • gromdimon 2y ago
      imho, that's more of a workaround than a primary feature of the model. I've also seen an agent that constantly rewrote its output by iterating through a loop with context updates. It worked, but as you mentioned, it's too slow and costly...
    • iamvladyashin 2y ago
      OpenAI shifts AI work from training time to inference time, which means that faster inference will enable higher model intelligence. It's only a matter of time until the competition catches up with cheaper/more powerful alternatives. But yes, other models have this too—it's just gotten easier!
      • gromdimon 2y ago
        true, AGI will be expensive )(