Report finds newer inferential models hallucinate nearly half the time while experts warn of unresolved flaws, deliberate deception and a long road to human-level AI reliability
Yeah, I think that workarounds with o3 is where we’re at until Altman figures out that just saying the latest oX mini high is “great at coding” is bad marketing when it can’t accomplish the task.
I don’t quite understand why o3 for coding? Do you mean for code architecture or something? Like creating apps? Why not use a better model if its for coding?
However, o4 is actually “o4 mini-high” while o3 is now just o3 now. The full release, no “mini” or other limitations. At this point o3 in its full form is better than a limited o4.
But, none of that matters while Claude 3.7 exists.
Yeah, I think that workarounds with o3 is where we’re at until Altman figures out that just saying the latest oX mini high is “great at coding” is bad marketing when it can’t accomplish the task.
I don’t quite understand why o3 for coding? Do you mean for code architecture or something? Like creating apps? Why not use a better model if its for coding?
That’s exactly the problem.
However, o4 is actually “o4 mini-high” while o3 is now just o3 now. The full release, no “mini” or other limitations. At this point o3 in its full form is better than a limited o4.
But, none of that matters while Claude 3.7 exists.