Plus: Nvidia's new tiny AI supercomputer; Veo 2 and Imagen 3; Google and Microsoft release reasoning models; Waymo to begin testing in Tokyo; Apptronik partners with DeepMind; and more!
The costs of those o3 runs are worrying. Not just from a resource consumption standpoint, but from an accessibility one too. The cost to access o3's general-purpose, apparently huge power will, of course, reflect its requirements. At this sort of scale, I can imagine wealth gaps widening further as those with endless money to chuck at o3 will do so wherever it can provide to them a competitive advantage, in order to stay at the top - and it looks like it could in a lot of areas. Unless OpenAI plays it carefully, this is exactly what they wanted to avoid to begin with - only the rich having access to the tech that matters.
I get your point and I agree with it. The required computing power to do the 88% ARC-AGI run is prohibitive for many, but this should not stay for long.
Although the cost of compute should go down, it likely won’t be dirt cheap, but it should become more accessible to more people and organisations over time.
Also, I wouldn’t discount the open source or open community. There might be something interesting coming from those people, like sharing GPUs or more efficient architectures. They may not compete with o3 yet, but they could come close to o1, and that might be enough to get this technology into the hands of more people.
The costs of those o3 runs are worrying. Not just from a resource consumption standpoint, but from an accessibility one too. The cost to access o3's general-purpose, apparently huge power will, of course, reflect its requirements. At this sort of scale, I can imagine wealth gaps widening further as those with endless money to chuck at o3 will do so wherever it can provide to them a competitive advantage, in order to stay at the top - and it looks like it could in a lot of areas. Unless OpenAI plays it carefully, this is exactly what they wanted to avoid to begin with - only the rich having access to the tech that matters.
I get your point and I agree with it. The required computing power to do the 88% ARC-AGI run is prohibitive for many, but this should not stay for long.
Although the cost of compute should go down, it likely won’t be dirt cheap, but it should become more accessible to more people and organisations over time.
Also, I wouldn’t discount the open source or open community. There might be something interesting coming from those people, like sharing GPUs or more efficient architectures. They may not compete with o3 yet, but they could come close to o1, and that might be enough to get this technology into the hands of more people.