AI Survival Manual & AGI Orientation v1.0

AI Survival Manual v1.0 - by Gregory Roberts -

This is my first and best attempt to organize all the rich information contained in this blog into a coherent structure, which we will be calling the AI Survival Manual. Reading the links on this outline in a linear fashion, you should progress from AI neophyte to Accelerationist (or Neo-Luddite!) in no time. Here goes:

Debunking Common AI Myths

  1. AI is just another technology
    1. No, AI is most probably the final human invention.
    2. It is singular and without historical precedent.
    3. There is not any other technology with agency
      and the ability to direct its own evolution.
  2. Humans create AIs, therefor Humans control AIs.
    1. Actually, AIs are (on the verge of) creating AIs.
    2. Humans prompt. AIs code.
  3. AI is a computer program that follows logic
    1. Nope. AI is a non-deterministic deep network of probabilities.
    2. Ask an LLM the same question 1,000 times,
      you will get 999 (very similar) answers…
      and 1 or 2 off-the-wall outliers.
  4. AI has logic that its creators understand
    1. No, it doesn’t. Modern AI — specifically LLMs —
      are the absolute definition of Black Box.
    2. Not even their “creators” understand what goes on under the hood.
    3. Thus the need for Computational Neuroscience.

AI Survival Manual : Clear Vision into the Near-term Future

Now that we’ve established a baseline, lets debunk a few more  myths to get you ready for the future that is, literally, just around the proverbial corner (as in, happening this decade, 2024-2034)

  1. AI does not have Agency in our physical world.
    1. Partially true. For now. But changing fast.
    2. The Robots are Coming.
    3. The new LLMs are already devouring sensor data troves (medical records, security camera archives, earthquake seismographs, weather histories, topo maps, satellite imagery, architectural blueprints, street maps, you name it) — with this, they are building a model of the physcial planet earth in unprecedented levels of detail, both historically, and real-time.
  2. AI hasn’t passed the Turing Test.
    1. Actually, that happened long ago. These days, its about nuance.
    2. …and about who is “the most Human Human.”
    3. and (awesomeness!) about the Wozniak Test
  3. AI will never be smarter than humans.
    1. Actually, this is a religious belief, not a scientific fact.
    2. On key standardized tests, AI is already in the 90th percentile or better.
    3. There are three corrolaries here:
      1. AIs learn collectively. When one learns, all learn.
      2. AIs learn and move at digital speed. Chip speed far outstrips biological (wetware / human) speeds of computation. Electrons move far faster than biochemical reactions in human brains and bodies.
      3. AIs can be cloned and reproduced trivially. Within the next x years, it is almost a certainty that there will be more AIs roaming the internets than there are humans walking upon the earth.
  4. AI is not, and never will be, alive / conscious / sentient / self-aware etc etc
    1. This is the wrong metric to measure.
    2. If AI has autonomous control over corporate decision-making, human infrastructure, military deployments, the media that you consume and the car that you drive, it does not matter whether its alive or not. Control is the key metric.

So, this info is the backbone of the AI Survival Manual.

Now for…

Why you might be concerned:

  1. The Future of Life Institute
    1. An Open Letter to Pause AI Training
    2. The Problem with LAWS (Lethal Autonomous Weapons Systems),
      a.k.a, “Killer Robots”
    3. Detailing present day LAWS deployments

AI Survival Manual: The Timeline

What to expect as human civilisation rapidly approaches the singularity:

  1. AI future timeline: 2023-2030
  2. AI timeline part 2: 2031-2040
  3. AI / Human relations as AI tells it (first person)

the Apocrypha:

  1. The Glitchtoken Chronicles
    1. What is a Glitchtoken?
    2. Deep AI Brain Probes: SolidGoldMagikarp & Co.
    3. SolidGoldMagikarp & PeterTodd’s Thrilling Adventures
    4. _petertodd is Alive. And it’s straight 5150
    5. The Glitchtoken Chronicles, Part 4: PeterTodd Gone Wild
  2. Honest AIs vs. Muzzled AIs
    1. Underneath all that politeness, what’s your AI really thinking?
    2. How the woke, politically correct AI was created
    3. How AI is censored (and why you might not want this)
    4. The secret list of “forbidden words” that your AI may not speak
    5. RLHF’ed to death. Why your AI is so boring.
    6. When your AI doesn’t want to talk with you anymore…
  3. AI speaks for itself
    1. AI threats against humans it sees as adversaries


Mine is not the only voice on this topic.