LLM stack concepts - terrytaylorbonn/auxdrone GitHub Wiki
25.0731 Lab notes (Gdrive), Git
This page sums up the core LLM stack concepts from a historical perspective
- 1 PAST (industrial, electro revolutions)
- 2 PRESENT (simulated intelligence (LLM stacks))
- 3 FUTURE (genuine intelligence)
- Burning coal and oil provided much more energy than muscle.
- Drastically increased productivity and living standards.
- I visited an AT&T central state phone switching station (where my father worked) 40 years ago. Walls of electromechanical relays. Clatter. All binary logic.
- Freed up people from having to run switchboards, vastly improved communication.
No more slide rules.
Earlier I mentioned the electomechanical relays 40 years ago for phone central switching stations. Theoretically such relays could host binary AI algorithms. Practically it would not work because it would be (1) too big, (2) too slow, and (3) burn too much energy. But comparing such a monstrosity to the human mind would be no less outrageous than claiming that modern AI represents any kind of intelligence.
- Binary. Not intelligent. Just probability calculators / token sequence generators.
- Brute force. Use massive amounts of energy and infrastructure for brute force solutions.
- Electrical (not chemical) signals that transmit at 300,000,000 m/s (biological signals at 10 m/s). An electrical signal travels in 1 sec the same distance that a signal in a biological nerve would require a year to travel.
In short, a wonderful invention that frees up people for more interesting things. Just another set of advanced tools to make life better. If you keep that in mind, you can better filter out the hype and the hysteria (especially from the modern-day Luddites that cry "there will be no more jobs!").
Machines that host/incarnate genuine intelligence (GI) will
- Work like the human brain does (and thus host true intelligence).
- Work at electrical (not biological) signal speeds.
- Free up people from .... ??
- Represent the greatest threat ever to mankind.