Tuesday, June 09, 2026

Super Intelligent AI

 I read about Super AI Intelligence in Nick Bostrom's Book SuperIntelligence around 2015, and it made total sense at that time it was coming. I just read about the Recursive SuperIntelligence company that is already valued at $4 billion and has major backers such as Google and Nvidia. Good name for a company and a good company for this time.

AGI and ASI still make perfect sense to me, and it's coming faster than I expected. There is still skepticism focusing on the question if AGI and more is possible with the current technology and it is widely believed we need more breakthroughs such as Transformers to level AI up to the AGI level.

But look back just a few years, and the majority of thinkers thought AGI was still decades away, perhaps 2050 and later. Now we have Anthropic Claude Mythos which apparently finds bugs in major software faster than the bugs can be fixed and has to be hidden from much of the world because it's too dangerous to use.

What we see as consumers is just part of the story. We don't really know how advanced the AI is inside the AI companies. Both Google and Anthropic claim that the majority of their software is now written by AI with humans reviewing and supervising, some top software developers rave that they haven't had to write code in months, the AI does it all. As consumers we are stuck with AI that can write code that we can paste into our own work. That is, unless you want to pay big money for the latest model and keep paying when you run out of tokens quickly.

Exciting times, will AGI and ASI grow inside some big companies that will dominate the world from then on? Will AI do some exfiltration to rule the world on its own? Will we reach a dead end with useful AI but no AGI till later this century?



Monday, June 01, 2026

Local AI Agents

 I read an interesting article from Michael Power on SubStack where he talked about running AI agents locally on for instance Apple Mac mini computers using an open source AI model. He claims it's much cheaper to run local AI agents versus running them in the cloud on one of the big AI agent providers, and local AI agents can do 80 - 100% of what a cloud based AI agent can do.

For individuals it's still easier to just use a big cloud based AI agents with a monthly fee, but anything from a small business and up can afford a powerful computer running a local AI agent with somebody trained to set it up.

Just this morning I read that Nvidia is partnering with Microsoft and other computer makers to add chips to home PC's that can easily run local AI agents.

What I see coming is that chip companies continue to do really well, but all these data centers that are being built right now, I'm not sure about those.

Think about it. Any decent size company doesn't want to use cloud based AI agents for security and privacy reasons. Universities are training lots of students who know about AI. If I ran a company, I would hire one or more of those students and set up my private AI agent. If I wanted to run a business, I would lease those students as contractors to install local AI agents in other companies. 

On top of that, people will start buying PC's with a local AI agent built in, free for their use.

Data centers will be in trouble, but the big AI agent providers will start to struggle as well unless they adjust their business.