Daily Tech News 12 July 2024

• Could AIs become conscious? Right now, we have no way to tell. (Ars Technica)
  
  Because we haven't properly defined the terms.
  
  What kind of AI? SHRDLU from 1968 (yes, I keep mentioning that) displayed more evidence of consciousness than ChatGPT, and examining its abilities we see clear indications of a primitive level of conscious.
  
  There's a term in AI, sphexishness, referring to the golden digger wasp, Sphex ichneumoneus. Like many wasps, the sphex wasp will paralyse its prey to serve as food for its larvae. In this case it will paralyse an insect, drag it back to its burrow (it's a digger wasp), check that the burrow is safe, and then put the prey in the burrow to serve as a larder for the baby sphexes.
  
  The thing is, if you move the paralysed insect while the sphex is checking the burrow, it will move it back next to the burrow, inspect the burrow again, and then drag it into the burrow.
  
  And if you move the insect again while the sphex is re-checking the burrow, the cycle will repeat. No matter how many times you do this, the sphex will not change its behaviour.
  
  SHRDLU was able to explain why it did what it did, step by step. It didn't get angry at being given repeated nonsensical orders, but it could account for every action that it took.
  
  Anyway, can computers become conscious? Well, they are already more conscious than insects, and have been for decades.
  
  Can computers become conscious with all the same complexity as human consciousness? Certainly not currently; they are not remotely powerful enough.
  
  Can LLMs become conscious? A single LLM, no. LLMs are designed specifically to avoid that.
  
  A pair of LLMs in a feedback loop? Maybe, yes. But likely also psychotic.

• Why The Atlantic signed that deal with OpenAI. (The Verge)
  
  Money.
  
  
  
• Ben Faw, the harbinger of AI slop. (The Verge)
  
  He made his mark pushing ads for his buddies' businesses unnoticed into online newspapers and magazines, and it was all downhill from there.
  
  
  
• A data breach at mSpy exposed the personal information of millions of customers. (Tech Crunch)
  
  mSpy sells spyware.
  
  
  
• TSMC is another big beneficiary of the AI bubble, with its market share topping $1 trillion for the first time. (Reuters)
  
  While still well behind the leading US bubblers like Apple, Microsoft, and Nvidia, it is nonetheless the most valuable company in Asia.
  
  
  
• The most important video game you've never heard of. (A Critical Hit)
  
  The Sumerian Game. It ran on the IBM 7090 mainframe ($2.9 million for 100 kiloFLOPs) and it was, well:

  Sir, I am sorry to report that 150 bushels of grain have rotted or been eaten by rats this past season.

  Hammurabi. It's Hammurabi.
  
  
  
• The man who destroyed Wikipedia. (Tracing Woodgrains)
  
  David Gerard, the man behind longtime lolcow dumpster fire RationalWiki, has also been systematically rotting away the credibility of Wikipedia for twenty years.
  
  Everything bad you've seen from terminally online leftists, he represents, and he's been at it for longer than most. He has his good points - he hates blockchain, for example, but he hates it from a doctrinaire liberal perspective, not because of the specific and often astoundingly ill-considered technical decisions that make working with it an unending misery.

posted by Pixy Misa at 04:38 AM

| Access Comments