Daily Tech News 18 February 2023

• Microsoft has lobotomised Bing's AI chat search thing called Sydney. (The Verge)
  
  After a bit of poking around with an ice pick, Sydney will now only respond to five questions per session and fifty per day. Although ChatGPT isn't designed to learn after initial training is complete, it does modify its responses based on inputs and earlier responses within a session, leading to increasingly erratic behaviour because at its core it's a box of mindless junk.
  
  
  
• Speaking of boxes of mindless junk ChatGPT is scary good at my job but there's one reason I'm not panicking. (ZDNet)
  
  ChatGPT isn't capable of arranging underhanded deals to flog fake SSDs. Yet.
  
  Also, it's not union.

• The Auspicious Machine is like a chunky Blackberry. (Liliputing)
  
  It has a 640x480 LCD screen - rather low resolution in this day and age, a physical QWERTY keyboard, a D-pad for games, a tiny trackball, and no CPU, memory, or storage whatsoever (though there is a micro-SD slot).
  
  That's because it's designed as a carrier for a Raspberry Pi Compute Module - specifically the CM4 - or anything with a compatible connector and form factor. The CM4 is just 40x55 mm - 1.6x2.2 inches - so it easily fits within a phone-sized device, and is available with up to 8GB of RAM and 32GB of storage.
  
  Not a terrible idea, though the CM4 is much slower than current phone models, even budget ones.
  
  
  
• The 7950X3D appears to be slower than the 7950X in Geekbench and Blender. (Tom's Hardware)
  
  The 5800X3D was also slower than the regular 5800X in common benchmarks, while turning in amazing results in certain game titles. (TechSpot)
  
  Nothing seems to have changed there, so if you mostly play Apex you can give this one a pass, but if you enjoy racing sim Assetto Corsa you could see a 40% increase in frame rates (if your graphics card can keep up).
  
  
  
• Nvidia's RTX 40980 Ti and Titan RTX Ada: Everything we know. (Tom's Hardware)
  
  Which is basically nothing. It's all speculation, though we do know that the 4090 only enables 8 out of 9 of the shader clusters on the AD102 chip, leaving room for a new card to be all of 12.5% faster.

posted by Pixy Misa at 04:00 AM

| Access Comments