Do not be naive
A lot of people are looking for Artificial Intelligence agents. By that they mean that you tell the AI what to do and it goes and does it. For example, suppose I am a campaign manager, I make a list of 25 things we need to do. Create commercials, contact people, do interviews, consult with certain experts, etc. I would then tell AI to go and set up those meetings, make those commercials, etc.
The problem is why can't I simply have an AI that is a campaign manager. Instead of 25 things needed it might make a list of 50 things. Instead of taking a week or a month to do them it might get it done in a day or a week.
If AI is so great why do we still need the campaign manager?
People have assumed that it could be 20 years after we achieve AGI that we get to ASI (Artificial General Intelligence and Artificial Super Intelligence). But already if we get AGI the AI will be more intelligent than every person on the planet.
Already AI is rewriting the code for AI. Think how many coders and how many lines of code are in there, it may be that some of the people on this planet are still better coders than AI, but so what. If it can go through all its lines of code and make small improvements that alone could cause it to increase by 5% in its skill level and IQ. Already the IQ is estimated to be 155 to 165. How many of the coders who wrote that code were operating at that level?
But you can apply the same technique to the design of the chips and all the components in the computers. If you make the parts better, cheaper and more economic. Then it will be very feasible to make them more powerful. So this too could cause you to increase by 5%. All these things give us a positive feedback loop.
Now the biggest and scariest improvement last week was that they could train AI using Youtube videos. That Shifts the training of these models into high gear. But the biggest and scariest improvement this week is they can be trained using synthetic data, training that is created by the AI itself. This improvement can cut the time to train the AI model by 90%, again decreasing the time required for the development of this AI to shrink drastically.
A lot of people are looking for Artificial Intelligence agents. By that they mean that you tell the AI what to do and it goes and does it. For example, suppose I am a campaign manager, I make a list of 25 things we need to do. Create commercials, contact people, do interviews, consult with certain experts, etc. I would then tell AI to go and set up those meetings, make those commercials, etc.
The problem is why can't I simply have an AI that is a campaign manager. Instead of 25 things needed it might make a list of 50 things. Instead of taking a week or a month to do them it might get it done in a day or a week.
If AI is so great why do we still need the campaign manager?
People have assumed that it could be 20 years after we achieve AGI that we get to ASI (Artificial General Intelligence and Artificial Super Intelligence). But already if we get AGI the AI will be more intelligent than every person on the planet.
Already AI is rewriting the code for AI. Think how many coders and how many lines of code are in there, it may be that some of the people on this planet are still better coders than AI, but so what. If it can go through all its lines of code and make small improvements that alone could cause it to increase by 5% in its skill level and IQ. Already the IQ is estimated to be 155 to 165. How many of the coders who wrote that code were operating at that level?
But you can apply the same technique to the design of the chips and all the components in the computers. If you make the parts better, cheaper and more economic. Then it will be very feasible to make them more powerful. So this too could cause you to increase by 5%. All these things give us a positive feedback loop.
Now the biggest and scariest improvement last week was that they could train AI using Youtube videos. That Shifts the training of these models into high gear. But the biggest and scariest improvement this week is they can be trained using synthetic data, training that is created by the AI itself. This improvement can cut the time to train the AI model by 90%, again decreasing the time required for the development of this AI to shrink drastically.
