Hey everyone, I just wanted to clear something up: OpenAI hasn't actually achieved Artificial General Intelligence (AGI) with their new o3 model. There are a few reasons why that's not the case.
1. The whole idea of what AGI means is pretty fuzzy. It's like we renamed Machine Learning to Artificial Intelligence, then invented "Artificial General Intelligence" to cover what we'd left out. People toss around the term AGI without really thinking, which isn't very useful.
2. So building on the last point, if you're a bit loose with what you mean by AGI, you can pretty much say anything you want about it! It's easy to just change the definition to fit whatever you're trying to argue.
3. A lot of emphasis has been placed on o3's ARC prize results, and they are impressive. But here is a quote directly from ARC, "it is important to note that ARC-AGI is not an acid test for AGI."
4. And another quote, "Passing ARC-AGI does not equate to achieving AGI."
5. François Chollet of ARC says, "I don't think o3 is AGI yet. o3 still fails on some very easy tasks, indicating fundamental differences with human intelligence."
6. And yet another, "Furthermore, early data points suggest that the upcoming ARC-AGI-2 benchmark will still pose a significant challenge to o3, potentially reducing its score to under 30% even at high compute (while a smart human would still be able to score over 95% with no training)."
So, o3 is a big step up, that's for sure! But let's be clear, it's not Artificial General Intelligence – not even close. And the compute costs? Wow, that's a story for another time!
Gary Explains
Hey everyone, I just wanted to clear something up: OpenAI hasn't actually achieved Artificial General Intelligence (AGI) with their new o3 model. There are a few reasons why that's not the case.
1. The whole idea of what AGI means is pretty fuzzy. It's like we renamed Machine Learning to Artificial Intelligence, then invented "Artificial General Intelligence" to cover what we'd left out. People toss around the term AGI without really thinking, which isn't very useful.
2. So building on the last point, if you're a bit loose with what you mean by AGI, you can pretty much say anything you want about it! It's easy to just change the definition to fit whatever you're trying to argue.
3. A lot of emphasis has been placed on o3's ARC prize results, and they are impressive. But here is a quote directly from ARC, "it is important to note that ARC-AGI is not an acid test for AGI."
4. And another quote, "Passing ARC-AGI does not equate to achieving AGI."
5. François Chollet of ARC says, "I don't think o3 is AGI yet. o3 still fails on some very easy tasks, indicating fundamental differences with human intelligence."
6. And yet another, "Furthermore, early data points suggest that the upcoming ARC-AGI-2 benchmark will still pose a significant challenge to o3, potentially reducing its score to under 30% even at high compute (while a smart human would still be able to score over 95% with no training)."
You can find all the quotes from ARC here: arcprize.org/blog/oai-o3-pub-breakthrough
So, o3 is a big step up, that's for sure! But let's be clear, it's not Artificial General Intelligence – not even close. And the compute costs? Wow, that's a story for another time!
4 months ago | [YT] | 128