Daily Tech News 7 January 2023

• High-end CPUs like AMD's Ryzen 7950X and Intel's Core i9-13900K are power hungry beasts, but you can tell them not to be. And here's what happens if you do that. (AnandTech)
  
  Reducing the 13900K to 125W cuts performance by 20%. The same thing happens with the 7950X if you reduce the power all the way down to 65W.
  
  On the other hand, if you reduce the power of the AMD chips down to 65W, they use more than 65W. So do the Intel chips actually, but the AMD chips exceed the setting by more than the Intel ones except at full power where the AMD chip behaves itself but the Intel chip blows out to over 300W.
  
  Anyway, cut the Intel chip to 125W, or the AMD chip to 65W, you'll lose 20% performance when running all cores simultaneously (single threaded performance isn't affected at all) but your system will run much cooler and hence much quieter.
  
  Also a good measure of how the new high-end notebooks will run, because they will use desktop chips running at reduce power envelopes just like this. And the answer is they'll be faster than the fastest desktop chips from the previous generation.
  
  
  
  

• If pressing a button will destroy the building, you should do more than just add a sign saying "please do not press this button again". (The Register)
  
  Maybe, I don't know, don't put the button there?
  
  
  
• It's just fifteen years to 1901. (Epochalypse)
  
  Time to start stocking up on canned goods.
  
  
  
• What happened to housing startup Pippin? (Spencer Burleigh)
  
  It was a dumb idea, executed by idiots, in California. Take it from there.
  
  
  
• If you're waiting for Nvidia's new low-end GPUs - the 4050, 4060, and 4070 - whether in a laptop or desktop get ready to be disappointed. (WCCFTech)
  
  Benchmarks of the laptop versions have leaked and they are very slightly faster than the previous generation, except for the 4050 which is actually a worthwhile improvement. (Though the laptop 3050 was honestly not very good.)
  
  Also, AMD's 7600M XT and 7700S are the exact same chip with the exact same settings. I don't know why.
  
  
  
• ChatGPT is a language model, not a world model. (New York Times)
  
  This makes it useless for anything beyond language.
  
  Some users have shown that it can translate simple JavaScript code to Python. That's plausible, since that's something a good language model should be able to do.
  
  But others have noted that similar AI models for assisted programming produce code with a significantly higher than normal rate of critical security bugs. And people have noted that ChatGPT will simply "lie" when it doesn't have access to facts that will support whatever you have told it to write.
  
  It doesn't know that it's lying because it doesn't have a model of the world, it only has a model of descriptions of the world.
  
  This works well enough for AI art, because good art is hard to create but bad art is easy to detect. If the hands are on the wrong end of the arms, you click the retry button.
  
  If there's a critical security flaw in the code your AI just wrote, on the other hand, you are screwed.

posted by Pixy Misa at 04:00 AM

| Access Comments