I first wrote about this on a linkedin post at: o3 - ASI (Super Intelligence)
o3 - ASI (Super Intelligence)
Following AI capabilities? o3 due to be released by OpenAI (o3-mini at least ) is approaching ASI . AI is outperforming most people by a factor of 2 or more on average. And even starting to out reason super human's, people that are deeply smart in a particular subject. The details of o3 are summarized by Alan D Thompson at o3.
Check out the various test, test designed to test the intelligence of AI models compared to humans. My personal use of AI doesn't quite match up with this analysis. I find more wrong answers than right. Taking a quantum computing online course I find that copilot, meta.ai often get the questions wrong and I have to prompt several times to get a correct answer. And one of the most frustrating things about Microsoft Office Copilot is that it can only help you write content. I want to be able to say something like, make sure all the fonts are the same across the whole document, insert a 3x5 table and set a nice looking style to it. IE. have it the AI actual do complex stuff that you have to look up how to do. Or in email find all emails related project X over the last 90 days and summarize the conversations. We need to get beyond just creating single page content and get to where AI can actually do things.
Back to ASI, one of the most interesting charts to me from lifearchitect.ai is the chip count, cost chart, see it here, the cost of GPT-5 estimated at $612M US and 120 days of compute time to build with the NVIDA H100 chip. That's a pretty massive amount of compute power.
Thinking about the power of quantum computing, if we can achieve super intelligence with classical computing imagine what AI might look like with quantum computing when we will have compute power that is going to be exponentially faster, I mean like super exponentially faster.
While quantum computing is still in what I call the public awareness phase it's not going to be long before we see quantum adoption, the race is on, just like computers of yesteryear the effort to shrink the size of CPU's is going to be the same trajectory for quantum computers. What now takes up massive amounts of space will one day be the equivalent of our smart phone size for a quantum computer (likely beyond our lifetime). While that size level is likely a long ways off we will see quantum computers working on AI problems. Some analysis indicate 4million qubits by 2035 on the aggressive side. That's enough to tackle some of the toughest chemistry problems.
We are on the precipice of an inflection point with AI and it's going to be interesting to see how it affects humanity.
*This article written without the help of AI. Well other than the header image, that is AI generated.
Add comment
Comments