3 ms·

> ... 4o and o1, which are great step changes but nothing the competition can't quickly replicate. Where did that assertion come from? Has anyone come close to

> ... 4o and o1, which are great step changes but nothing the competition can't quickly replicate. Where did that assertion come from? Has anyone come close to replicating either of these yet (other than possibly Google, who hasn't fully released their thing yet either), let alone "quickly"? I wouldn't be surprised if it's these "sideways" architectural changes actually give OpenAI a deeper moat than just working on larger models.

dcre 2y ago
Claude 3.5 Sonnet is at least as good as 4o.
- stavros 2y ago
  Much better than 4o, in my experience. I've stopped using 4o almost completely, except for cases where I need good tooling more than I need good performance (PoCs and the like).
  - disiplus 2y ago
    it's better at coding for me, and it looks like because it was trained on more data. i have this problem with a library that i use, and it was trained on python2 version but python3 version is completely different. Python3 versions is relatively new, but out from at leaset 2020 and you would mostly find examples with Python2 version if you googled. They both produce garbage code in this solution. Claude versions is just 20% less garbage, but still useless. The code mixes those two, even if i specify i want the python3 version or directly specifying a version.
- falcor84 2y ago
  Oh, sorry I wasn't clear, I was referring to the advanced (speech-to-speech) voice mode, which for me is the highlight of this "omni" thing, rather than to the "strength" of the model.
  - dcre 2y ago
    I thought that might be the case. On that: even though other big players haven't copied that particular feature, I doubt (though I'm not by any means an expert) that it's anywhere near as hard to replicate as the fundamental model. I do agree that at least product-wise, they're trying to differentiate through these sort of model-adjacent features, but I think these are classic product problems requiring product solutions rather than deep tech innovation. Claude Artifacts, GPT Canvas, speech, editor integration, GitHub integration, live video integration, that kind of stuff.

dcre 2y ago

Claude 3.5 Sonnet is at least as good as 4o.

stavros 2y ago
Much better than 4o, in my experience. I've stopped using 4o almost completely, except for cases where I need good tooling more than I need good performance (PoCs and the like).
- disiplus 2y ago
  it's better at coding for me, and it looks like because it was trained on more data. i have this problem with a library that i use, and it was trained on python2 version but python3 version is completely different. Python3 versions is relatively new, but out from at leaset 2020 and you would mostly find examples with Python2 version if you googled. They both produce garbage code in this solution. Claude versions is just 20% less garbage, but still useless. The code mixes those two, even if i specify i want the python3 version or directly specifying a version.
falcor84 2y ago
Oh, sorry I wasn't clear, I was referring to the advanced (speech-to-speech) voice mode, which for me is the highlight of this "omni" thing, rather than to the "strength" of the model.
- dcre 2y ago
  I thought that might be the case. On that: even though other big players haven't copied that particular feature, I doubt (though I'm not by any means an expert) that it's anywhere near as hard to replicate as the fundamental model. I do agree that at least product-wise, they're trying to differentiate through these sort of model-adjacent features, but I think these are classic product problems requiring product solutions rather than deep tech innovation. Claude Artifacts, GPT Canvas, speech, editor integration, GitHub integration, live video integration, that kind of stuff.