Comments by "Michael Nurse" (@michaelnurse9089) on "Sabine Hossenfelder" channel.

  1. "The learning database already includes virtually 100% of all text written by humans, " No, before starting training they run all the text through AI inference of the previous model. This improves quality by a significant percentage. In reality, there is always going to be another layer of AI between the current one being trained and the data.
    24
  2. She summarized it with ChatGPT to 2 pages.
    4
  3. Somewhat strawman. These SV guys would say they dream in S-curves, only that this particular S-curve will destroy us all before its inflection point.
    3
  4. Yup, that is the thing. Nobody, not Sabine or the funny looking guy, or anyone else knows for sure where this S-curve ends. A little humility all around would go a long way.
    3
  5. "My understanding is that AI in mining has not reduced the number of people" The thing is, you don't see the full picture because your vision ends at your borders. Here in South Africa we used be the biggest mining country in the World, sending millions underground into hellish conditions. But the rest of World moved onto machines, in particular Australia, while such things were politically unpalatable here. One-by-one the mines closed down here, not because people were suffering or the ore had run out, but because automated mining is 2x more cost efficient. Today South Africa has the highest unemployment rate in the World - a true rate of about 60% (official 40%) and most of those were once miners. Machines (and AI) can definitely eat up jobs, it is just not always in your backyard.
    2
  6. "but you can of course argue that our so called intelligence is not much more than that either." You cannot argue that, nowhere in our brains have we found a specific statistical method underlying the thinking process. This is just armchair speculation.
    2
  7. In her defense she probably has ChatGPT write the script.
    2
  8. No. Model quality improves linearly with compute. Right now they are compute constrained and data constrained.
    2
  9. It is a dead technology with because it costs roughly 10x solar and wind + batteries.
    1
  10. No idea. ChatGPT went to 100m in two months. Cars took roughly 60 years to achieve the same.
    1
  11. Hyperloops would work fine if there was a strong market for them, but 99% of people hate trains and their inconveniences, so Elon built self driving EVs instead.
    1
  12. Several of his predictions have not panned out.
    1
  13. At least in Terminator they kill you. In the Matrix they put you in a snot bath and stick cables up your orifices.
    1
  14. FFS - it said edible glue. Edible glue is wheat and water mixed together, the same stuff pizza is made from.
    1
  15. Everyone knows this. The questions is whether the curve dies out before AI intelligence exceeds our intelligence or not. If it is the latter there will be serious problems. I suspect the former.
    1
  16. Also f'ed up the Ray Kurtzweil date.
    1
  17. Autonomous driving is doing really well. Tesla will do 9/10 commute runs with no safety issues. Waymo are far ahead of that.
    1
  18. No, because you are superhumanly intelligent. Nor am I.
    1
  19. You just generated some extra text data with your post. It is not the best quality but every little bit helps.
    1