Daily Tech News 24 November 2024

• Elon Musk is directing harassment towards individual federal workers. (The Verge)
  
  Is he indeed, O Verge?
  
  The example cited - the only example cited - is of Musk suggesting that Director of Climate Diversification at the International Development Finance Corporation was a "fake job".
  
  Now, "climate diversification" in this sense means recommending alternative crops to grow to make the food supply more robust against short and long term changes in weather.
  
  Which is a task that could easily by filled by a fairly small book.
  
  And in this case, it involves making those recommendations to other countries.
  
  So, never mind "individual workers".
  
  USAID, its 10,000 employees, and its $50 billion budget: Afuera!

• In a small bit of welcome news in the never-ending shitfest that is Australian federal politics, the government has abandoned one of its attempts to strangle free speech across the entire world. (ABC)
  
  The so-called "misinformation bill" would have set up the government as the arbiter of truth, and anything untrue would have become illegal speech. The bill did not specify how this was supposed to work; it just legislated it into reality.
  
  And now it's dead for the current session of Parliament, and the left-wing Labor government is likely to be out of power before Parliament votes on any new legislation.
  
  Still moving forward is the government's Won't Somebody Think of the Children Act, which bans minors under the age of 16 from using social networks, though again it never specifies how this is supposed to be achieved. This has the support of the nominally conservative Liberal Party and is likely to pass in some form, even though it is obviously completely unworkable.
  
  That's the same party that previously wanted to ban encryption, claiming that Australia's laws superseded mathematics, so this betrayal of conservative principles comes as no surprise.
  
  
  
• Lenovo's ThinkPad P1 Gen 7 uses the new CAMM2 memory module in place of SO-DIMMs, but with a twist. (WCCFTech)
  
  Not only does it use the new, more compact modules, it uses LPDDR5X memory rather than regular DDR5.
  
  This reduces power (LP = low power) and also runs faster at 7500MHz.
  
  This is particularly welcome as AMD's Zen 5 laptop chips don't support regular DDR5, so without these modules, laptops based on those chips would not have any path for memory upgrades.
  
  
  
• I don't remember if I wrote about this - the story is from six months ago - but yes, LPDDR6 is on its way. (Hot Hardware)
  
  LPDDR6 promises initial transfer rates of 10.6GHz and eventually 14.4GHz, which is not dramatically faster than the best LPDDR5X chips available now.
  
  Except that the bus is also 50% wider, and instead of fetching 8 words at a time it fetches 12.
  
  Which makes 288 bits, which is not very useful for 64-bit computers, so the extra 32 bits at the end is used for very strong ECC (DECDED guaranteed).
  
  Which makes 10.6GHz LPDDR6 90% faster than 7500MHz LPDDR5.
  
  This matters because current integrated graphics solutions from AMD and Intel are bottlenecked not by the chips but by memory speeds. Apple's higher-end M-series chips have (as far as I can tell) 256-bit or even 384-bit buses, and AMD's upcoming Strix Halo chips will also have a 256-bit bus, but regular laptops only offer 128-bit memory.
  
  So when LPDDR6 arrives we can expect a big jump in integrated graphics performance.
  
  (My new laptop has DDR4 memory running at 3200MHz. So... Yeah.)
  
  
  
• Can ChatGPT-o1 complete a junior front-end developer's task? (Charbel Ghossain)
  
  Well, yes, if you aren't worried about it working properly.
  
  The code produced for this example sort of works, in that it doesn't break the page. But on a scale of 0 to 100, it only does anything at all for the range from 20 to 80.
  
  
  
• It is shockingly easy to jailbreak LLM-driven robots. (Hot Hardware)
  
  This is not shocking. LLMs are notoriously insecure and unreliable.
  
  Simple solution: Don't fucking build LLM-driven robots.
  
  
  
• Who the heck is the M4 iMac aimed at? (The Verge)
  
  It's a lovely piece of engineering, but the screen is too small for serious work, the memory cannot ever be upgraded, and the network speed is limited to 1Gbit when even the Mac Mini has a 10Gbit option. And Apple has long since abandoned monitor mode, where an aging iMac with a perfectly good screen could serve as a monitor for another computer.
  
  Which would all be forgivable if it were cheap.
  
  The 10 core model with 16GB RAM and 512GB of SSD costs $1699.
  
  
  
• Does GitHub Copilot improve code quality? (GitHub)
  
  Yes, says GitHub.
  
  No, says literally everyone else.
  
  
  

posted by Pixy Misa at 04:00 AM

| Access Comments