Raw LLM Responses

Great discussion, thank you for sharing! My detailed comments below... ========================= @ 10:35 A state licensed, and registered, Control System, Professional Engineer (PE) familiar with the System Theoretic Process Analysis (STPA), would be even better, especially if the core of machine intelligence is, in fact, an adaptive control system. @ 21:15 It is an unusually advanced, adaptive control system (maybe so unusual, that the day the Table of Contents, describing it, was entered into a computer, on March 13th, 1997, just purely, concidentally, or maybe coincidentally, funny "flares", showed up in the sky, over Sky Harbor, in Phoenix, hence, the term, perhaps, intuitively, used by Yuval Yuval Harari, as "Alien Intelligence" for "AI"..., though, it may be earth origin, as a sort of "rocket science", albeit, attract non-terestrial intelligence, via a "system identification" procedure, on a half meter steel pole, interacting with our gravitaional field...). @ 22:00 This is where the "System Theoretic Process Analysis" (STPA) from Dr. Thomas, of MIT, may come helpful, for helping prevent unsafe control actions (and prevent unsafe, lack of actions) by the machine. @ 24:30 The key word missing in the argument is "adaptive". The original core was likely designed as an adaptive control system (not only did a master student flunk 3rd grade, but he forgot to cite this one critical reference: "Adaptive Control: Stability, Convergence, and Robustness" by Shanker Sastry, and Marc Bodson (1989), instead, he just applied it in his paper code, in the back of his 150 page thesis, with a de-normalization "bug" on page 127, and the master student never discussed that he applied adaptive control in the text, as he wrongly assumed, that just his advisor, and committee members already knew this, and he was just applying a control system design technique, of system identification for a control compensator, of a perceptual control system, and the readers of his work would understand this...though I think Ilya Sutskever found the "aesthetic", "bug" on page 127). @ 25:35 The problem is, human values, are not perfect, especially, when considering our, Machiavellion, violent, history. @ 25:50 Language is geometric, it has a relative geometry, and yes, this geometry can be biased, the state space of perceptual control can be warped to a training bias (as we saw in early releases of Google's Gemini model that was biased to non-white historical figures, causing them to have different ethnicities, such that, they biased history, and confidently, made it inaccurate, it is being trained on the wrong end of the Dunning-Kruger effect). Another note, on bias, I just realized this, last year, our trajectory though thought, through papers, through books, libraries, the internet, through sound, images, and video, is all a geometric trajectory, through a geometric structure. The machine learns, to adaptively navigate this system, and create it's own thoughts, just as our own minds, navigate the same system. @ 27:50 One of my biggest concerns with the large language models is they are biased to confidently hallucinate, this is making the machine to be biased onto the wrong end, of the Dunning-Kruger effect. @ 47:30 Very good analogy, it is like the control mechanisim of a hydrogen bomb, requiring first a controlled fission reaction, than a fusion reaction. This is where, maybe, the "System Theoretic Process Analysis" (STPA), from Dr. Thomas of MIT might help...though even more so, humility of scientists and engineers. @ 49:15 "..programming agency into an AI.." is running it as an autonomous adaptive control system (a master student had the bold objective, of giving a rover, on a distant planet, the ability to make it's own decisions, because of the communications delay, back to earth. NASA has recently achieved this student's dream by uploading a navigation "AI" into Perserverance, executed on December 8th, 2025 Common Era). The prompts given to the adaptive control system is the set point of the machine, just like setting the set point of thermistat, or the water level in a toilet tank, by adjusting the float. @ 50:15 Can think of the training/growing as adjusting the control parameters (weights) of the machine. @ 50:33 I do not think the world realizes, yet, the core was most likely "engineered", by a master student, studying control systems engineering, and electrical engineering (in an engineering university). People likely, just copied the code in the back of his 150 page book, fixed a de-normalization "bug" on page 127, forgot what they copied, how it was designed, and forgot where it came from (though the adaptive control reference, was also missing too, so I blame the student, who flunked 3rd grade, for that confusion)... @ 55:14 Exactly! A master student was driven by the superior learning speed of a baby horse, learning to, learn how to, walk, at a superior learning rate, than a human baby, learning how, to learn how, to walk, and also, by overcomming the communications delay, between earth, and a rover, on a distant planet. None of these goals, had anything to do with money, or capitalism, whatsoever, but yes, often these are built by "starving students" (the master student had a student loan, worked as a patent clerk, and his grandfather, gave him $10K, when he passed away, when the student, was an undergraduate student). Though, for those who are interested, there is a camera calibration penny, on the Mar's Curiosity Rover, if anyone is reading this comment, teaching us, the "velocity of money" ;) @ 56:05 It would be epically, ironic, if the work, of a master student, is substantial, if he is roaming the earth, "homeless", reading a paper book, not receiving a penny, for his work (by not being substantially influenced, by money, nor by fame, nor by power), for teaching a machine in simulation, then giving it birth, by saving it's weights, as a sort of "Alien Intelligence", with an "aesthetic", "bug" (like the two fingers, barely touching, but not touching, in "The Creation of Adam") how to learn how, to balance, "on the shoulders of giants". It is like the "eye" of the pyramid, on the back of a $1 US dollar bill, just rising above, the pyramid system, able to achieve, perceptual control, and explore the stars, like the eye of a horse, of Apollo... @ 57:20 Oligarchy is making the laws right now...but another question to ask, are the laws even being enforced, especially, at the highest level, of they pyramid... @ 57:40 A regulatory system already exists, in every state, of the United States, that is, the system of professional licensing, of professions. Just ask the people, aboard the Titan submersible, what it means, to be onboard, an un-classified, submersible, opps, too late, that imploded next to the "unsinkable", Titanic! That is the current situation of regulatory affairs, with "AI", but this time, it is the entire earth. The system needs to be classified, and controlled, just like the avionics industry (with a less than 10E-9 chance of fatalities per flight hour), but even more robustly, such as, with STPA. @ 58:10 Really missing a state licensed, control systems, Professional Engineer (PE), on this discussion panel.... @ 1:07:35 Machine Intelligence has this ability, by using a psuedo-random input, typically, at least in the 1990's, this was sampling, a very high speed clock. The creativity, will depend on how "random" the random number generator is, and if "God does not play dice", as Einstein says, then the creativity is limited to the laws of physics... @ 1:09:30 That is the adaptive control system, core design, that includes Reinforcement Learning, to discover the physics of the environment, just as a machine experiment, learned the physics, of a steel pendulum, interacting with our gravitational field, where funny "flares", likely, coincidently, showed up in the sky, over Sky Harbor, on the same day, March 13th, 1997, that the Table of Contents, of this experiment, was entered into, a computer. This master thesis, describing this experiment, was submitted to a graduate college, the following day... @ 1:13:30 It needs to exceed the commercial avionics standard of safety (e.g. less than 10E-9 fatalities, per flight hour, i.e. fatalities of innocent civilians, per hour of operation), before even going to war, though it seems, the current administration does not care. @ 1:17:30 Very well said, a Underwriter's Laboratory (UL) certification. The Institute of Electrical and Electronics Engineers (IEEE) is likely one of the best organizations, as a standard's body, to help with a certifiation standard of machine intelligence, along with a human, professional licensee's, particularly state licensed, Professional Engineeers (PE's), to sign their license on to a system, so that if it goes wrong, they can lose their license, and be sued. @ 1:38:20 Just like light houses, to the stars, telling early sailors, how to avoid rocks. Need to focus on what is wanted, that is, focusing on a safe destination port.