Mark sounds like hes drowning in his saliva Truly the pavlovian response is on display here.     See Less

So many assumptions…yet ignoring the obvious. The massive GPU power resources on such a ridiculous scale…will come from where? Last time I checked electrical power is insanely expensive.     See More that current GPU architecture is incredibly sloppy both in terms of power consumption and heat, regardless of the $value of Nvidia. The premise of LLM’s storing extremely large datasets…is a stone age problem definitely NOT ready global release for the masses. Ease up everyone, and think realistically.    See Less

If AI Agents become the backbone of business/activity and the Internet goes down, ala CrowdStrike outage, we are actively building more and more "systems" on top of the Internet an     See More ting ourselves up for failure. I would imagine UNTIL we can make the Internet bullet proof, building more and more systems is a fool's errand. Civilization is based on Electricity and the Internet and if either of these goes down, MANY have no idea how to grow food, handle day to day life without these systems. VERY WORRIED this "Super Cool Futuristic AI Agents" is the normal evolution of apps/services but sure seems like a high deck of cards being added to the top of our very susceptible system and AI Agent's will one day actively take our systems DOWN!    See Less

Thank you for such increible information and explanation and I share your thoughts about the future, my opinion would be even bigger that we expect     See Less

The world of AI framworks is facinating to me, since u can mix and match agents with different capabilities along the cost, speed and accuracy curve.
So u may have free AI such as LAMA 3.     See More e grunt general purpose work that is needed most of the time. Whilst u use specialist AI agents that are expensive but they are experts, highly accurate and fast as fuck. The teacher LAMA's are also needed for when the framework draws a conclusion from partial products produced or obtained by the specific agents.
If the framework has a negative or positive feeback or feedforward configuation then the desired result can be compared with the partial product. These itative comparison can drive the teacher AI ti say ok, these are the errors or difference between desired partial output what do need to do to remove or minimise the difference.
The teacher will then say, OK what kind of AI do need to train or how can I train an existing AI that I tought before, better.
This is very interesting since there is reaning and understanding that AI will get things wrong. Even the most expert of expert AI will get some things wrong relstive to what we want or logically wrong.
To be honest having a Formal Maths AI or classical Formal reasoning machine as part of thid framework would be great. This woulf ensure thay any step could be checked for jt logical behsvious was ensured.    See Less

that was a very long way to explain that there will be "templates"     See Less

Definitely not crazy; it all depends on the cost of running those product models. We simply can't have that trivia if the mean medium cost are to thousands of dollars a day!

Thank     See More video! ❤    See Less

Yep, I'll keep repeating it over and over. We can have a manageable, Zero-Loss AI transition through both sensible regulation and by the fact that AI is going to enable MORE jobs than i     See More ve replaced. We have years of work converting not just hearts n minds to AI use, but businesses themselves. Use cases to be discovered, data hoards unleashed from private companies. I know of two geologists who are building AI based on centuries of company mining data, and they're gonna be able to send teams out and pinpoint sample pattern areas. Success means more jobs especially if the company hits paydirt at multiple sites.    See Less

Mark is a most NPC looking person in the world.     See Less

Is this because of sun glasses, or because of the meta quest? I reference to the bright eye sockets.     See Less