Daily Tech News 9 July 2024

• Goldman Sachs released a report on generative AI - the hot new thing pushing stock valuations of key tech companies into the trillions. In short: It's trash. (Where's Your Ed)
  
  The article linked above is not short but it is a good read. The full report is available for download and it doesn't pull any punches either:

  The promise of generative AI technology to transform companies, industries, and societies continues to be touted, leading tech giants, other companies, and utilities to spend an estimated ~$1tn on capex in coming years, including significant investments in data centers, chips, other AI infrastructure, and the power grid. But this spending has little to show for it so far beyond reports of efficiency gains among developers. And even the stock of the company reaping the most benefits to date - Nvidia - has sharply corrected.

  From the article:

  In essence, on top of generative AI not having any killer apps, not meaningfully increasing productivity or GDP, not generating any revenue, not creating new jobs or massively changing existing industries, it also requires America to totally rebuild its power grid, which Janous regrettably adds the US has kind of forgotten how to do.

  There is that, yes.

  Generative AI is not going to become AGI, nor will it become the kind of artificial intelligence you've seen in science fiction. Ultra-smart assistants like Jarvis from Iron Man would require a form of consciousness that no technology currently - or may ever - have - which is the ability to both process and understand information flawlessly and make decisions based on experience, which, if I haven't been clear enough, are all entirely distinct things.

  Right. Although the understanding doesn't have to be flawless, merely good enough for the task at hand, and cheap enough that it's not simpler to just train a human and pay them to do it.
  
  Generative AI doesn't understand anything - it is a language model, not a fact model; doesn't gain experience, at least not in its current form, which is trained once at enormous expense and then left to rot; and doesn't make decisions.
  
  It's not AGI and has no path to become AGI. Terry Winograd's SHRDLU from 1968 is in important respects more sophisticated than ChatGPT, even though it was written by a single grad student on an 18-bit computer more than fifty years ago.
  
  
  

• Leaked benchmarks of AMD's Ryzen 9900X pin it as the fastest CPU available for single threaded tasks. (Tom's Hardware)
  
  Intel had taken that crown, mostly by cranking up the power consumption of some of its chips to 400W, but the 9900X beats that.
  
  That's roughly in line with AMD's own promises. But this twelve core chip also beats AMD's sixteen core 7950X3D on the multi-threaded version of the same benchmark, which is about double what AMD promised.
  
  The eight core 9700X also beats Intel's 14900KS on single-threaded tasks though with only eight cores vs. 24 on the Intel chip, it's 33% slower on the multi-threaded version.
  
  We'll see real numbers soon enough; the chips are expected at retail next week.
  
  
  
• HP has cancelled its cheapest inkjet and laser printer models. (Tom's Hardware)
  
  Cheap because they locked you into a perpetual ink/toner subscription plan, without which they wouldn't function at all. Just need the scanner on your multi-function device? Sorry, you haven't paid for the ink you don't need.
  
  So abandoning this is actually good.
  
  
  
• You can now link 10,000 Moore Threads GPUs together to build a super computer. (Tom's Hardware)
  
  I wonder how viable this is. The Chinese-made Moore Threads GPUs have been tested in Windows PCs for gaming and they are just abominable, performing far below their theoretical capabilities. But supercomputers are tightly controlled environments, where you modify the code to fit the hardware, and they might actually deliver in such an environment.
  
  
  
• There are too many Python package managers. (DuBlog)
  
  Well, sure, but with Node and NPM you can install a single package and get 25 known critical vulnerabilities bundled in. Python can't compare with that.

Ancient Library in Tibet Creating Digital Archive of Its 84,000 Scriptures -

Tibet's Sakya Monastery is home to many wonders. Founded in 1073 CE, its collection includes some of the oldest Tibetan artwork, as well as 84,000 ancient manuscripts and books. Given its remote… pic.twitter.com/DYNL1cxBxI

— Archaeo - Histories (@archeohistories) July 6, 2024

posted by Pixy Misa at 04:30 AM

| Access Comments