• Two things to clarify from yesterday.  First, that quote at the top of the post explaining the DRAM Apocalypse was from Jatin Malik, an engineer at Atlassian.
  
  Second, brains are computers.
  
  
  
• Agent psychosis: Are we going insane?  (Armin Ronacher)
  
  Apparently, yes:

  You can use Polecats without the Refinery and even without the Witness or Deacon.  Just tell the Mayor to shut down the rig and sling work to the polecats with the message that they are to merge to main directly.  Or the polecats can submit MRs and then the Mayor can merge them manually.  It's really up to you.  The Refineries are useful if you have done a LOT of up-front specification work, and you have huge piles of Beads to churn through with long convoys.

  That's from the Gas Town Emergency User Manual which would be a great name for a work of surrealist speculative fiction but is quite literally a user manual.

  Looking at Gas Town (and Beads) from the outside, it looks like a Mad Max cult. What are polecats, refineries, mayors, beads, convoys doing in an agentic coding system? If the maintainer is in the loop, and the whole community is in on this mad ride, then everyone and their dæmons just throw more slop up.  As an external observer the whole project looks like an insane psychosis or a complete mad art project.  Except, it's real?  Or is it not?  Apparently a reason for slowdown in Gas Town is contention on figuring out the version of Beads, which takes 7 subprocess spawns.  Or using the doctor command times out completely. Beads keeps growing and growing in complexity and people who are using it, are realizing that it's almost impossible to uninstall.  And they might not even work well together even though one apparently depends on the other.

  What is Beads?
  
  Beads is a quarter of a million lines of code to manage Markdown files in Git repositories.
  
  I have written entire enterprise systems with paying customers and decade-long track records that are no larger than that.
  
  But I didn't have agentic AI to help me, so they actually worked.
  
  There's a term in programming called technical debt, which measures the cost of a quick fix that you know you will have to rip out and fix properly one day.
  
  Vibe coding is the technical debt singularity.

• Minecraft-meets-Torchlight game Hytale supports modding out of the box.  So someone modded in Hytale.  (Tom's Hardware)
  
  You can play the entire game inside the game.
  
  You can also run Windows 95 as a mod inside Hytale.
  
  And Doom.
  
  The game has been out for a week.
  
  
  
• Dumbphone owners have lost their minds: The deranged rantings of a smartphone owner.  (Wired)
  
  It's hard to know where to even start with this.  The author - self-reportedly Gen Z - has a crippling codependency with her iPhone, and her friends who have abandoned such devices for dumbphones are crippled without them.
  
  They're like walking, talking, quarter-million line codebases that can only do a single task, and fail even at athat.
  
  
  
• Minisforum is planning to introduce the DeskMini BD395i Max - a mini-ITX motherboard with a Ryzen 395 CPU, up to 128GB of soldered RAM, and a PCIe slot.  (Notebook Check)
  
  Okay.  Sure.
  
  Let me know when you have something out in a Zen 6, okay?
  
  
  

Musical Interlude

Posted by: PatBuckman at Wednesday, January 21 2026 01:16 AM (rcPLc)

Hide Comments | Add Comment

• The reason why RAM has become four times more expensive is that a huge amount of RAM that has not yet been produced was purchased with non-existent money to be installed in GPUs that also have not yet been produced, in order to place them in data centers that have not yet been built, powered by infrastructure that may never appear, to satisfy demand that does not actually exist and to obtain profit that is mathematically impossible.
  
  — Jatin Malik on Twitter
  
  
  
• The mythology of conscious AI.  (Noema Mag)
  
  This article gets one important thing right: LLMs are not conscious.

  In a 2022 interview with The Washington Post, Google engineer Blake Lemoine made a startling claim about the AI system he was working on, a chatbot called LaMDA. He claimed it was conscious, that it had feelings, and was, in an important sense, like a real person. Despite a flurry of media coverage, Lemoine wasn’t taken all that seriously. Google dismissed him for violating its confidentiality policies, and the AI bandwagon rolled on.

  I commented on that story at the time.  Lemoine is a crazy as a sack of rats on crazy pills.  And was also completely and very obviously wrong, which is not the same thing.

  As AI technologies continue to improve, questions about machine consciousness are increasingly being raised. David Chalmers, one of the foremost thinkers in this area, has suggested that conscious machines may be possible in the not-too-distant future. Geoffrey Hinton, a true AI pioneer and recent Nobel Prize winner, thinks they exist already.

  Wait.  David Chalmers said what?
  
  Huh. He did say that.  A mostly sensible article summarising the evidence on both sides, concluding that pure feed-forward LLMs are not conscious but extended LLMs with recurrent processing - feedback loops - could be.  (Boston Review)
  
  Back to Noema:

  Taken together, these biases [anthropocentrism, which is irrelevant, human exceptionalism, which is irrelevant, and anthropomorphism, which is actually the key here - Pixy] explain why it’s hardly surprising that when things exhibit abilities we think of as distinctively human, such as intelligence, we naturally imbue them with other qualities we feel are characteristically or even distinctively human: understanding, mindedness and consciousness, too.

  A little bit of nonsense thrown in at the start but an accurate description of the problem in the end.
  
  But then it all falls apart:

  The very idea of conscious AI rests on the assumption that consciousness is a matter of computation.

  Which is rather like assuming that water is a molecule made of one oxygen atom and two hydrogen atoms.

  More specifically, that implementing the right kind of computation, or information processing, is sufficient for consciousness to arise.

  Because it is.

  This assumption, which philosophers call computational functionalism, is so deeply ingrained that it can be difficult to recognize it as an assumption at all.

  As much as the molecular structure of water is an assumption.

  But that is what it is.

  Nope.

  And if it’s wrong, as I think it may be, then real artificial consciousness is fully off the table, at least for the kinds of AI we're familiar with.

  "Kinds of AI we're familiar with"?  Do you mean feed-forward models, which are definitely not conscious, or enhanced systems with feedback loops?

  Challenging computational functionalism means diving into some deep waters about what computation means and what it means to say that a physical system, like a computer or a brain, computes at all.  I'll summarize four related arguments that undermine the idea that computation, at least of the sort implemented in standard digital computers, is sufficient for consciousness.
  
  First, and most important, brains are not computers.

  And we're dead.
  
  Brains are obviously computers and it is trivially easy to prove this.
  
  Take a line of BASIC code, like:
  
  10 PRINT 3+7
  
  What does that do?
  
  It prints 10.
  
  How do you know?
  
  Because you can execute that code in your head.
  
  How can you do that?
  
  Because your brain is a computer.
  
  It may be more than a computer - though nobody has produce a coherent, let alone convincing argument for this - but it is unquestionably a computer.
  
  There follow dozens of paragraphs of irrelevancies I won't get into, but suffice to say that it all goes downhill from there.

• Quick reminder that Intel's B570 is still available at $200.  It's not the fastest graphics card - it's comparable with Nvidia's RTX 3060, a midrange card from five years ago - but it's cheap, in stock, and works.  And it has 10GB of RAM, a small upgrade over common 8GB cards.
  
  
  
• An Altair 8800 that has been broken since its owner made a mistake assembling it in 1974 has finally been fixed.  (Tom's Hardware)
  
  Never too late.
  
  
  
• Reasons to stop using MySQL.  (Optimized by Otto)
  
  Oh, yeah?  What would you recommend?
  
  MariaDB.
  
  Oh.  Yeah.  Good call.  Carry on.
  
  
  
• The Slimbook Executive ticks all my boxes.  Pretty much.  (Liliputing)
  
  I'm not really in the market for a new laptop right now, but this seems to get a lot right.
  
  It's a 14" model with an Intel 255H CPU (6P/8E/2LP cores), a 2880x1800 120Hz screen - LCD rather than OLED, but it covers 100% of sRGB so it should be fine unless you're a professional artist of video editor - and a 99Wh battery despite weighing a modest 1.2Kg.
  
  It has two SODIMM slots and two M.2 2280 slots - unusual and welcome in a 14" laptop - and one USB4 port, a 5Gb USB-C port with DP and PD (DisplayPort and Power Delivery), three 5Gb USB-A ports, HDMI, wired gigabit Ethernet, an audio jack, and a full-size SD card slot.
  
  And the Four Essential Keys.
  
  No dedicated graphics, but the CPU includes Intel's Arc 140T graphics which are quite competent and generally comparable to AMD's 780M.
  
  Oh, and the keyboard is backlit and there's a physical privacy shutter for the camera.
  
  They don't ship to Antarctica though.
  
  
  
• Why Silicon Valley is really talking about fleeing California (it's not the 5% theft wealth tax).  (Tech Crunch)
  
  Oh, do tell.

  Take Larry Page, who [owns] about 3% of Google but controls roughly 30% of its voting power through dual-class stock. Under this proposal, he’d owe taxes on that 30%.  For a company valued in the hundreds of billions, that’s a lot more than a rounding error.  The Post reports that one SpaceX alumni founder building grid technology would face a tax bill at the Series B stage of the company that would wipe out his entire holdings.

  Oh, right.  It's not the 5% wealth tax.  It's the 100%+ wealth tax.  I can see how that would be a problem.

  David Gamage, the University of Missouri communist kleptocrat retard law professor who helped craft the proposal, thinks Silicon Valley is overreacting.  "I don’t understand why the billionaires just aren’t calling good tax lawyers," he told The San Francisco Standard this week.

  That's just it.
  
  They did.
  
  Their tax lawyers advised them to flee.  Immediately.

Posted by: David Eastman at Tuesday, January 20 2026 03:28 AM (J69gL)

Posted by: PatBuckman at Tuesday, January 20 2026 04:16 AM (rcPLc)

Posted by: Robin Munn at Tuesday, January 20 2026 10:04 AM (plzaG)

Posted by: Pixy Misa at Tuesday, January 20 2026 10:56 AM (PiXy!)

Posted by: Pixy Misa at Tuesday, January 20 2026 10:57 AM (PiXy!)

Posted by: Robin Munn at Tuesday, January 20 2026 12:35 PM (u8IQw)

Posted by: Pixy Misa at Tuesday, January 20 2026 01:22 PM (PiXy!)

Posted by: Pixy Misa at Tuesday, January 20 2026 01:23 PM (PiXy!)

Posted by: Robin Munn at Tuesday, January 20 2026 02:33 PM (u8IQw)

Posted by: Pixy Misa at Tuesday, January 20 2026 05:15 PM (PiXy!)

Hide Comments | Add Comment

• Same: Elon Musk wants $134 billion in his OpenAI lawsuit.  (Tech Crunch)

  The figure comes from expert witness C. Paul Wazzan, a financial economist whose bio says he has been deposed nearly 100 times and testified at trial more than a dozen times in complex commercial litigation cases.

  I have acted as an expert witness once.  You couldn't pay me enough to do it again...  Though $134 billion would be tempting.

  Wazzan, who specializes in valuation and damages calculations in high-stakes disputes, determined that Musk is entitled to a hefty portion of OpenAI's current $500 billion valuation based on his $38 million seed donation when he co-founded the startup in 2015.

  Much of the article is devoted to ad-hominem attacks by the partisan lunatics at Tech Crunch, who think fraud is okay against people for whom they feel irrational hatred.
  
  
  
• Better be quick, Elon because an analysis of OpenAI's finances suggests they could run out of money as soon as 2027.  (Tom's Hardware)
  
  If you're wondering why the beancounters at the big memory manufacturers are being wary of rapid expansion in the face of unprecedented demand, well, it's because they are able to read a spreadsheet and they do not like what they see.
  
  
  

Tech News

• Also because they are. Well, not Samsung but Micron has broken ground on its new $100 billion, 1.2 million square foot factory in upstate New York, which is expected to be in production by 2030.  (The Register)
  
  They originally planned to start construction in 2024, but were delayed by bats.

  The megafab - which would displace 500 acres of woods and wetlands, as well as two endangered species of bats - is scheduled to begin producing DRAM chips by 2030.
  
  Micron said it will create 1,216 acres of off-site bat habitat including maternity roosts to mitigate the potential damage its fab will cause to the Indiana and northern long-eared bat populations, as well as 628 acres of land to offset impacts to the sedge wren, short-eared owl, and northern harrier bird populations.

  Part of the land the factory will be built on is swamp, so Micron has been obligated to build a new swamp to replace the old one.
  
  
  
• Hardware Unboxed reported that there were no Nvidia RTX 5070 Ti review cards to be had for a planned model roundup, that Asus told them the 5070 Ti was "end of life", and that Australian retailers reported that there was zero stock of any model at suppliers.
  
  Nvidia insisted this was all lies and slander and slanderous lies and every model of the RTX 5000 lineup was still in production.
  
  For the real story, let's check Newegg, since their site can easily be filtered to exclude second-hand and out-of-stock items and cards from strange marketplace sellers in Uzbekistan.
  
  Total 5090 models available at any price: Zero.
  5080 - one, at 80% over MSRP.
  5070 Ti - one, at 40% over MSRP.
  5070 - 8 models starting at close to MSRP.
  5060 Ti 16GB - I initially found none at all, but a second search dug out one card.
  5060 Ti 8GB - the model all the reviews told you not to buy is readily available starting at MSRP.
  5060 - 20+ models in stock starting at MSRP.
  5050 - 4 models in stock starting at MSRP.
  
  So if you're looking for a low-end card or for the mid-range 12GB 5070, you're in luck - they're in stock and selling for their listed prices. But the cards with 16GB or more VRAM are gone.
  
  On the AMD side things are a little better. The three 16GB cards - the 9060 XT, 9070, and 9070 XT - all show multiple models available, but all starting at 5% to 20% above MSRP, a problem that has persisted since they launched except for a few short weeks late last year.
  
  Advantage: Team Unboxed.
  
  
  
• Well, that's...  Something.  Intel's Bartlett Lake embedded processors will bring up to 12 P-cores (full-size performance cores) and no E (efficiency / low power cores) running at up to 5.9GHz to Socket FC-LGA16A for the embedded market.  (WCCFTech)
  
  No Intel consumer CPU has ever had 12 P cores; the largest number ever was with the 10-core 10900 introduced in 2020, and subsequent to that the maximum has been 8.
  
  But, you point out, these are embedded CPUs, not consumer models.
  
  True.  But Socket FC-LGA16A is better known as Socket 1700, the same one used by Intel's 12th, 13th, and 14th generation desktop CPUs, and for which inexpensive motherboards supporting both DDR4 and DDR5 are readily available.
  
  Might be worth keeping an eye out for these.
  
  
  
• There are no cheap PCIe 4.0 switches to expand the number of lanes available in PC motherboards...  Except motherboard chipsets like the AMD B650 which does exactly that so people are using it in unintended ways.  (WCCFTech)
  
  I mentioned this before but this is a nice roundup with pictures of expansion cards using the B650 to provide four additional M.2 slots and four SATA ports in a half-height form factor.
  
  It's an open source project but it would be interesting to see someone pick it up and run with it.
  
  
  
• Rebuilding New York...  In Minecraft.  (Tom's Hardware)
  
  A project born of the Wuflu lockdown and weaponised autism.  It's legitimately impressive.
  
  
  
  
  
• Speaking of which, I played some more Hytale.  Despite being in early access and its developers apologising for its unpolished state, it's a fun game - and it only costs $20.
  
  The best way to think of it is that it's not a pure Minecraft clone but a combination of the best features of Minecraft and older action RPGs like Torchlight.  The world can be taken apart and rebuilt block-by-block - mostly - but it's faster-paced and more action-oriented than Minecraft, and has story elements that Minecraft is mostly lacking.
  
  
  
• In praise of tyranny: Nearly 5 million social media accounts have been shut down in Australia, and the New York Times thinks that's a good thing.  (New York Times)
  
  What The Times largely avoids mentioning is that the teenagers affected immediately created new accounts and worked around the ban.  Traffic from the sites to Australia has not been reduced at all since the ban went into effect.

Posted by: Rick C at Monday, January 19 2026 04:20 AM (1zWbY)

Hide Comments | Add Comment

• Taiwan has reached an agreement with the US for a $500 billion investment project in return for a tariff reduction from 20% to 15% and tariff waivers during the construction of new factories.  (Tom's Hardware)
  
  That amount includes $250 billion of direct investment and $250 billion of credit guarantees by Taiwan's government, but it also includes $100 billion of already-planned investment by TSMC.
  
  If you're throwing literally hundreds of billions of dollars around for Taiwanese semiconductor manufacturers setting up in the US, maybe earmark a few billion for Nanya to expand consumer DRAM production.
  
  
  
• Because Samsung sure ain't doing it.  (WCCFTech)
  
  Samsung is expecting a 5% increase in DRAM production in 2026 - and that's in the face of unprecedented shortages and a projected additional 30% increase in demand over the course of this year.
  
  The Big Three are just begging for Taiwan or West Taiwan to eat their lunch.

• AMD's long-awaited dual-vcache CPU was a no-show at CES but increasingly looks real.  (WCCFTech)
  
  Dubbed the 9950X3D2, it keeps leaking in benchmark results.  It is exactly two 9850X3D CPU complexes in a single package - 16 cores and 192MB of cache running at up to 5.6GHz.
  
  
  
• If you're in the market for a four-port 400Gb Ethernet switch, Mikrotik has one.  (Serve the Home)
  
  $1295.  And considering it has the same bandwidth as a 640-port 2.5Gb switch - if such a thing existed - that's pretty darn cheap.
  
  
  
• ChatGPT Go is now available at $8 per month.  (Bleeping Computer)
  
  With ads.  But of course with ads, as OpenAI is bleeding cash like a US federal agency under a Democratic president's second term.
  
  
  
• OpenAI says its new ads won't influence answers.  (Bleeping Computer)
  
  It will lie to you exactly the same as the more expensive plans.
  
  
  
• TSMC says AI demand is endless, just like tulip bulbs.  (Ars Technica)
  
  But you can eat tulip bulbs.
  
  
  
• Naya Connect is a very keyboard.  (Liliputing)
  
  It has a fairly standard 85-key layout - a little cramped but it does include the Four Essential Keys - plus an optional 6-key macropad, plus an optional 24-key numeric keypad, plus an optional 24-key macropad with either blank keys or a assortment of 84 interchangeable icon keys to choose from, plus an optional trackball with four buttons and four macro keys, plus an optional trackpad with four macro keys, plus an optional haptic dial (that is, it gives programmable physical feedback) with an integrated trackpad and four macro keys, plus an optional 3D mouse with six degrees of freedom...  With four macro keys.
  
  With five choices of keyswitch, keycaps in either black or white, and the metal frame in either plain aluminium or black anodised aluminium.
  
  The basic keyboard costs $119 (with your choice of colours and keyswitches), while the full setup costs...  A lot.  Well over $1000.  They have an early-bird "All-in" bundle that isn't complete and that's already $917.
  
  
  
• Microplastics from washing clothes could be hurting your tomatoes.  (Washington Post)  (archive site)
  
  You heard them.  Stop putting tomatoes in with your laundry.
  
  So, how solid is this study?

  "Sometimes microplastics can have a positive impact on soil or plant properties, and sometimes it could be negative," Djuric said.

  So you're saying maybe I should put tomatoes in with my laundry?
  
  
  
• Fixing the HTTP rate limit header specification.  (Dotat.at)
  
  A 429 response means you're asking too many questions, go away.  The new spec tries to tell you how long you should go away for, but at the moment it assumes that everyone involved is equally stupid.  This is an attempt to fix that.
  
  
  
• Played some more Hytale today.  It runs at 30fps on low settings on my laptop, which means it should run faster on just about anything else, since my laptop has a 2880x1620 screen and five year old integrated Vega 8 graphics and I habitually run it in silent mode (with the fan speed turned all the way down).  It's a great system for doing work, lousy for playing games.
  
  Also, in Hytale low graphics settings look basically the same as "epic" graphics.  The main change there is the render distance, just as with Minecraft, so unless you spend all your time looking to the horizon it make little difference.  
  
  

Musical Interlude

Posted by: Joe Redfield at Sunday, January 18 2026 06:19 AM (KOtXO)

Posted by: normal at Sunday, January 18 2026 09:21 AM (e0fX0)

Hide Comments | Add Comment

• The best graphics cards for gaming in 2026.  (Tom's Hardware)
  
  At #2 is Nvidia's 5070 Ti, and at #5 is the 5060 Ti, both of which have ceased to exist.
  
  
  
• The best options for building a new PC with DDR4 RAM you already have.  (Tom's Hardware)
  
  Both systems include the 5060 Ti, which as we have just noticed, has gone to the great parts bin in the sky.
  
  
  
• Nvidia has rebutted claims that the 5070 Ti and 5060 Ti are "end-of-life", insisting that all cards are still shipping and are just shagged out after a long render.  (PC Magazine)
  
  The fact that you can't actually buy them is clearly a skill issue.  Git gud, scrub.
  
  Hardware Unboxed - an Aussie hardware review channel - was planning a market roundup of currently available 5070 Ti models, and reached out to their industry contacts for review cards.  There were no review cards to be had anywhere.  (YouTube)
  
  Their contacts in the Australian retail market confirmed that there was no stock of the 5070 Ti at distributors at all, echoing recent reports from Germany and Japan.  And Asus told them, and subsequently confirmed, that the 5070 Ti was "end-of-life", had ceased production, and would not be restocked.
  
  Then Nvidia... Clarified...  That all RTX 5000 series cards are still in production, and Asus walked back earlier comments and agreed that the 5070 Ti is still in production.
  
  Hardware Unboxed requested - again - a review sample.  There has been no response as yet.
  
  
  
• AMD meanwhile is working with its board partners to prevent GPU price hikes.  (Notebook Check)
  
  Good luck with that.  The 8GB 9060 XT seems to have bounced back up from its recent low price - it was briefly selling below MSRP - but it and the three 16GB cards (9060 XT, 9070, and 9070 XT) are all in stock.  As is the 32GB 9700, though that model is quite expensive.

• TSMC plans to spend $56 billion on new production facilities and equipment this year.  (WCCFTech)
  
  That's equal to three Nanyas.
  
  
  
• Reprompt is - or was, it's just been patched - a Copilot exploit that, once you clicked on the link, kept stealing your data even after you closed Copilot.  (Hot Hardware)
  
  It worked like most so-called "Prompt Injection" vulnerabilities: It asked the AI for your data, and the AI helpfully handed it over because it has less self-interest than the average mycoplasma.
  
  This particular instance has been patched, but it was active in the wild for four months before Microsoft took action.  This is why even people who use AI tools are up in arms about Microsoft integrating it directly into their operating system. 
  
  If you use AI through your browser, it stops the moment you close the tab.  Copilot never stops until you kill it with fire.  Hot Hardware has a detailed article on removing it that involves running unsigned code directly out of GitHub because that may actually be safer than leaving Windows untouched.
  
  
  
• US senators are demanding answers from X (Twitter), Meta (Facebook), and Alphabet (Google) on sexualised deepfakes.  (Tech Crunch)
  
  Answers to questions like "How do I do it?", "Does it have pictures of Sydney Sweeney?", and "You're not recording this are you?  Stop recording right now!"
  
  
  
• Italy's privacy watchdog is under investigation for corruption and embezzlement.  (Reuters)
  
  I am shocked, shocked, to see corruption and embezzlement going on in here!
  
  
  
• The Minisforum N5 Air is a $499 5-bay NAS.  (Liliputing)
  
  It's a cheaper version of the N5 with a plastic rather than metal case, but otherwise almost identical.  Apart from five 3.5" drive bays, it has room for three M.2 SSDs (with two running only at PCIe 4.0 x1 speed and one at x2), and includes a Ryzen 255 CPU, room for two DDR5 SODIMM modules (none as standard, so beware of that), and a whole lot of ports: OCuLink, two USB4 ports, 10Gb and 5Gb Ethernet, three USB3 10Gb ports, and HDMI.  One of those USB3 ports is inside the system to host a boot device so that you can dedicate all the drives to data.
  
  
  
• The best office mini-PCs review by Notebookcheck.  (Notebookcheck)
  
  Top spot goes to the Apple Mac Mini M4 Pro.  With 64GB of RAM and 1TB of SSD, though, it costs more than three times as much as the same configuration for the Minisforum AI X1 at #5.  (They list the 32GB model, but I bought the 64GB model.)
  
  
  
• The FDA has announced a recall of chocolate bars from Spring & Mulberry.  (Southern Living)
  
  Then we have number four: Crunchy Frog.
  
  We use only the finest baby frogs, dew-picked and flown from Iraq, cleansed in the finest quality spring water, lightly killed, and sealed in a succulent, Swiss, quintuple-smooth, treble-milk chocolate envelope, and lovingly frosted with glucose.
  
  And what is this one: Spring Surprise?
  
  Ah, that’s one of our specialities.  Covered in dark, velvety chocolate, when you pop it into your mouth, stainless steel bolts spring out and plunge straight through both cheeks.
  
  
  
• Everything I ordered during my definitely not drunken New Years Eve shopping spree has now shipped.
  
  The SSDs doubled in price and the memory tripled the next day.  I do need to pick up an AM4 CPU to use it all - the motherboard is on its way - but those are cheap and readily available.
  
  
  

Bananya Interlude

Musical Interlude

1 Sigh.  My son's got a 6800 that he's happy with and I am using a 4070 which, again, works fine...but if the market's going to be stagnant for a couple of years I wonder if we ought to upgrade to a 9070 and a 5000-series.  But the 5070 isn't a worthwhile upgrade based on benchmarks and the 5070 Ti is both a lot more than I want to spend, and also higher wattage than I want to use.
Speaking of graphics, I updated all the drivers and stuff on my X1, and then installed the Radeon drivers, and now if I watch YT videos and try to do something in another window, the whole display flickers.  I solved that problem by installing Arch Linux.

Posted by: Rick C at Saturday, January 17 2026 02:00 AM (1zWbY)

Posted by: Joe Redfield at Saturday, January 17 2026 05:00 AM (KOtXO)

Posted by: PatBuckman at Saturday, January 17 2026 06:02 AM (rcPLc)

Hide Comments | Add Comment

• A sad story with a happy ending.
  
  Riot games just acquired Hytale developer Hypixel Studios.  (April 2020)  (WCCFTech)
  
  Hytale is officially dead and its studio disbanded.  (June 2025)
  
  Hytale, the long-awaited Minecraft-like MMO, is resurrected and will launch in early access soon.  (November 2025)
  
  Hytale is saved says Hypixel founder as the studio's dev costs for the next two years have already been covered by pre-orders.  (January 2026)
  
  If you're interested in that kind of game - similar to Minecraft but with RPG elements and modding support built in - it's available in early access right now on the company's website.
  
  For $20.  (Engadget)
  
  Because, to quote the founder of Hypixel Sudios, "The game is unfinished and runs on a build from over four years ago. Charging more didn't feel right. I don't think the game is good yet."
  
  Honestly it seems to run pretty well, though I haven't spent much time in it.  And I was running on a laptop with integrated graphics, so far from high-end hardware.

• Cerebras has scored a $10 billion deal with OpenAI.  (CNBC)  (archive site)
  
  Cerebras is the company that makes the wafer-scale AI CPUs, the ones with 900,000 cores on a single chip the size of a dinner plate.
  
  
  
• SK Hynix is investing $13 billion on a new factory...  (Tom's Hardware)
  
  ... To assemble HBM chips for AI servers.  Not for you.
  
  
  
• Now that the memory drought has dried up supply of DDR5 and the older DDR4 modules alike, DDR3 is next to go.  (WCCFTech)
  
  I have 96GB of DDR3 - at least - sitting in older computers.  Much older.  Built between 2011 and 2015.
  
  
  
• AI is a 400-year confidence trick.  (Tom Renner)
  
  I don't think you could generate music videos on a 17th century mechanical calculator, though.
  
  
  
• The Minisforum MS-S1 Max is the best Ryzen AI Max mini PC yet.  (Serve the Home)
  
  It certainly has the best complement of I/O: Dual 40GBps USB4 ports, dual 80GBps USB4 ports, dual 10Gb Ethernet, one HDMI, five regular USB ports, and three M.2 slots, and a PCIe slot (though just half height, single slot).
  
  And it's 15% faster than the Framework Desktop apparently thanks to better cooling, because the two systems use the exact same CPU.
  
  
  
• Digg is back.  (Tech Crunch)
  
  Reddit ate Digg's lunch years ago after a the latter pushed out an extremely unpopular UI update.  Now the founder has bought back the company and is trying again,  It's a clean design and runs quickly, at least.  And Reddit is a toxic swamp these days.

Posted by: Rick C at Friday, January 16 2026 12:52 AM (Gpwen)

Posted by: PatBuckman at Friday, January 16 2026 01:29 AM (rcPLc)

Posted by: Pixy Misa at Friday, January 16 2026 07:47 AM (PiXy!)

Hide Comments | Add Comment

• Every part of human bodies from the brain to, um, other bits, is increasingly overrun with microscopic particles of plastic, unless somehow every study generated by the entire field is measuring contaminants or misinterpreting normal biological processes.  (The Guardian)
  
  But that couldn't possibly be true, could it?

  "I’m convinced we detected microplastics," she said.  "But I’ve always said that [the amount estimated] could be maybe twice lower, or 10 times higher."  In response to Kuhlman’s letter, Prof Lamoree and colleagues said he had "incorrectly interpreted" the data.

  Well, that doesn't sound convinced.  And Lamoree was the lead author on a 2022 paper that first detected microplastics in human blood.

  Prof Lamoree said: "I really think we should collaborate on a much nicer basis - with much more open communication - and don’t try to burn down other people's results.  We should all move forward instead of fighting each other."

  Don't ask questions.  Questions are mean!

  "We do have plastics in us - I think that is safe to assume," said Materić.

  Assume?

  "But real hard proof on how much is yet to come."

  And may never arrive.
  
  
  
  
  
  

Tech News

• There is no audience for Apple's Vision Pro because there is no content for Apple's Vision Pro because there is no audience for Apple's Vision Pro and this is all somehow your fault according to Apple.  (Stratechery)
  
  With the Apple II, and the Mac, and the iPhone, you had to fend off customers with pitchforks.
  
  With the Vision Pro that's a little less true.
  
  
  
• First they came for your DRAM.  Next they came for your air conditioning.  And lights.  (MSN)
  
  Power demand in the US is expected to rise 25% by 2030, largely due to AI datacenter buildouts, according PJM (Pennsylvania-New Jersey-Maryland Interconnection), the largest power grid operator in the US.
  
  Is there 25% more power to go around?  Uh, no.

  When Asthana took over as CEO of PJM in 2020, power plants within its service area were shutting down more quickly than they could be replaced.  Six years later, that trend has continued even though power demand has risen.

  In fact, worse than no.
  
  
  
• President Trump has said that regulatory changes are on the way to prevent AI datacenters from pushing the costs for expanding the power grid onto retail customers.  (Tom's Hardware)
  
  No word yet on the other problem.
  
  
  
• Where would we be without analysts?  (The Register)

  "We had assumed this would be primarily a 2026 story.  More and more, particularly at CES, we had conversations with folks and they're saying 'You won't find price stability in memory up until maybe late 2027.'  It means that prices will stop rising.  Not that they’re going to drop.  That’s what we’re hearing," he told us on Tuesday.  "This could be much longer than we had previously anticipated. Not just we at IDC, but we as a collective market."

  There are still some options for buying memory that aren't insanely inflated, but they're getting fewer by the day.  Most people don't eat memory like I do, fortunately, or need multiple computers in every room of the house...  Like I do.
  
  
  
• The Framework Desktop has finally increased in price, for example.  (Tom's Hardware)
  
  Though the 32GB model has only increased from $1099 to $1139 - which still makes it competitive with leaked pricing for the Steam Cube.
  
  The 128GB model has jumped from $1999 to $2459, a much steeper climb but less than the price increase on 128GB of RAM bought at retail.
  
  
  

Musical Interlude

Posted by: HL King at Wednesday, January 14 2026 11:04 PM (zKMEf)

Posted by: Pixy Misa at Thursday, January 15 2026 12:17 AM (PiXy!)

Posted by: Gopher at Thursday, January 15 2026 12:39 AM (h2BK4)

Posted by: DougO at Thursday, January 15 2026 02:58 AM (OwJfJ)

Posted by: Mauser at Sunday, January 18 2026 07:40 PM (XWgGM)

Hide Comments | Add Comment

• With support for Windows 10 ending last October, user numbers for Windows 11 climbed to an all time high.  Which was only to be expected.
  
  Less expected was that the numbers for Windows 11 have been dropping ever since.  (MSN)
  
  Over the last two months Windows 11 has fallen from 53.7% of the PC market to a very slim majority at 50.7%, with Windows 10 growing from 42.7% to 44.7%, and most of the remainder going to Windows 7, which came out all the way back in 2009.
  
  This is not particularly good news for Microsoft.

• Nvidia CEO Jensen Huang is deeply hurt by all the negative opinions of AI.  (Tom's Hardware)
  
  He has no mouth and he must whine.
  
  
  
• A quantitative analysis of the cost of overreliance on AI advice.  (Science Direct)
  
  A very dry paper, but in short decision makers tend to over-rely on AI advice even when it directly contradicts available empirical data.  And this was in the context of an iterated prisoner's dilemma scenario, where optimal strategies are simple and well-known.
  
  
  
• With contract prices for DRAM rising 70% and NAND flash by 100%, memory now accounts for 20% of the cost of a mobile phone.  (WCCFTech)
  
  Which is not all that much.
  
  
  
• Amazon bought a Bee.  (Tech Crunch)
  
  Bee makes a wearable AI thing, which is to say, useless overpriced junk.
  
  
  
• VMWare ESXi has a -365 day vulnerability.  (Bleeping Computer)
  
  If a zero-day vulnerability is one that must be fixed immediately because it is open to instant exploitation, what do you call one that has been actively exploited for a year before the developers notice?
  
  
  

Musical Interlude

Posted by: Rick C at Wednesday, January 14 2026 02:11 AM (tCA/g)

Posted by: normal at Wednesday, January 14 2026 06:28 AM (e0fX0)

Posted by: PatBuckman at Wednesday, January 14 2026 08:14 AM (rcPLc)

Posted by: Mauser at Wednesday, January 14 2026 02:54 PM (XWgGM)

Posted by: Frank at Wednesday, January 14 2026 03:55 PM (amxkj)

Posted by: PatBuckman at Thursday, January 15 2026 12:39 AM (rcPLc)

Posted by: normal at Thursday, January 15 2026 04:27 AM (LADmw)

Hide Comments | Add Comment

• Two of the co-founders of Twitter and Pinterest are working on a new social network cult.   (MSN)
  
  A cult with venture capital.

  I have no idea whether they will succeed.  But as a columnist writing about how to keep our humanity in the 21st century, I believe it's important to focus on people who are at least trying.  Social media, after its promising early days, took a dark turn and is now a deeply alienating force, replacing healthy human interaction and leading us to spend much of our lives looking at our phones.

  That's only partly the fault of the social media companies.
  
  Mostly it's just people.

  Many fear that AI, still in its adolescence, will take us down the same path because of the financial incentives to maximize eyeballs and time online.  But it could go either way, and we as a society still have a choice about how we use these powerful tools.

  Fair.  Let's see how West Co - the new company in question, is tackling the problem.

  Their bid for redemption is West Co. - the Workshop for Emotional and Spiritual Technology Corporation

  It's the Sirius Cybernetics Corporation all over again.

  - and the platform they’re testing is called Tangle, a "purpose discovery tool" that uses AI to help users define their life purposes, then encourages them to set intentions toward achieving those purposes, reminds them periodically and builds a community of supporters to encourage steps toward meeting those intentions.

  Danger, Will Robinson.

  "A lot of people, myself included, have been on autopilot," Stone said.  "If all goes well, we’ll introduce a lot of people to the concept of turning off autopilot."

  But will all go well?

  Each morning, staffers light the candle at the "ancestor table," with photographs of family members dead and alive and other sources of inspiration: Harvey Milk, Jane Goodall, Saint Francis.  The founders participate in a fortnightly "covenant" meeting in which they renew their support for each other.  Moments of silence precede meetings.  Ceremonies mark the changing of seasons.  Employees give one another regular updates on their personal triumphs and struggles.  Founders rate their weeks on a zero to 100 "poopy to pleasure" index.

  I have never seen a tech company more likely to earn its own 10-part miniseries on HBO Max after the comet arrives.

• It's the year of Linux on the desktop...  For tech journalists.  (The Verge)  (archive site)
  
  The author installed Linux - specifically CachyOS, an open source distribution with similar goals as SteamOS, to support modern hardware and make it easy to run games.
  
  Everything worked...  Except Minecraft.  But that was because he wanted to play Minecraft Bedrock, because that's the version his kids play, and it's not easy to make it work on Linux.
  
  
  
• AMD's Ryzen AI Halo mini PC is a Ryzen 395+ AI mini PC for AMD.  (Hot Hardware)
  
  No specs, no price, no schedule.  It has a bunch of USB-C ports at the back, so we'll see if it delivers a lot of USB4 or it's something less interesting.
  
  
  
• Twitter is planning to make its algorithm open source, with releases every month.  (MSN)
  
  What use this is I'm not sure, but it's better than not having it.
  
  
  
• My home fibre network disintegrated.  (Alienchow)
  
  Not mine - I have fibre internet, but copper networking inside - but the author's, who carefully laid conduit in the concrete slab when building his new home and ran fibre throughout and is now finding that this was not a good idea at all.
  
  
  
• Why consumers have gotten the memory shortage narrative right.  (WCCFTech)
  
  Self-serving spin from Micron trying to paint themselves as anything other than villains.
  
  
  
• Save $130 on 32GB of DDR5 memory by making your own.  (Tom's Hardware)
  
  All you need is 32GB of DDR5 memory and the ability to desolder and solder surface-mount components.
  
  

Musical Interlude

Posted by: Joe Redfield at Tuesday, January 13 2026 04:22 AM (KOtXO)

Hide Comments | Add Comment

• Meta has signed long-term contracts with three suppliers for a total of 6GW of nuclear power for its datacenters.  (Tech Crunch)
  
  Two of the partners are startups developing small modular reactors; the third, Vistra, is a major operator of traditional large non-modular reactors.
  
  
  
• Meanwhile researchers in China have managed to break the Greenwald limit on the number of alt accounts a single journalist can have agreeing with his posts before he screws up.  (The Independent)
  
  Wait, I'm told in this context it's an empirical limit on sustained plasma densities in Tokamak-style fusion reactors.
  
  With this breakthrough, fusion power is now only nineteen years and twelve months away.

• It's a water heater!  It's a Bitcoin miner!  It's both!  (Tom's Hardware)
  
  Given that the main cost of mining Bitcoin is the energy it takes - and disposing of the waste heat somehow - why not use it for, say, heating water?
  
  The Superheat H1 does exactly that.
  
  And if the price of Bitcoin collapses, it still heats your water.
  
  
  
• Can light move faster than the speed of light?  (Science Daily)
  
  Some theories of loop quantum gravity predict small fluctuations of the speed of light in a vacuum.  New experiments put an upper bound on how big those fluctuations could be, ruling out some of these theories.
  
  I call them theories rather than mere speculation because they actually made sufficiently robust predictions to be ruled out in the first place.  A theory can be wrong, but it's not a theory if you can't test it.
  
  
  
• Can AI do your job?  (MSN)
  
  If you're a computer programmer, or you otherwise work solely with words, possibly yes.
  
  If you do literally anything else, the best AI models currently succeed at 2.5% of human tasks.
  
  
  
• AMD will be launching socketed versions of the Ryzen 400 processor series later this year.  (WCCFTech)
  
  Too late, really.  Zen 6 will be out this year, and if you want integrated graphics it looks like Intel's Panther Lake is faster - though only if you are willing to go with a system with soldered memory.  If you need conventional DIMM or SODIMM memory, you get models cut down from 12 GPU cores to just 4, which isn't beating anything.
  
  So maybe not too late, but underwhelming.
  
  
  
• And if you want dedicated graphics, you might need to buy soon.  (Tom's Hardware)
  
  Nvidia GPUs are hardest hit so far, with reports of stock on all models running low in Germany and Japan.  Since that was the first direct warning we had of the memory crunch (though in retrospect the signs were there months earlier) I bought myself a 9060 XT while they were in stock and on sale.  Which as of time of writing, they still are.
  
  
  
• Asus and Gigabyte are putting 64MB of ROM on their new AM5 motherboards.  (Tom's Hardware)
  
  Earlier AM4 motherboards often had as little as 16MB of ROM for the BIOS, which became a problem as AMD kept releasing new chips for the platform - Socket AM4 first appeared in 2016, even before Zen 1, and the most recent new processor for that socket, the 5500X3D, was launched in June last year.  There were simply too many different models of compatible processors to fit all the necessary code into 16MB.
  
  With AM5 confirmed to support the upcoming Zen 6 chips later this year, and strongly hinted to support Zen 7 due in 2028, it will end up being an even longer-lived platform with more CPUs, so motherboard makers are fixing the problem before it arises.
  
  
  
• Amazon is using AI to sell your products on their store without you having to lift a finger.  (Modern Retail)
  
  Also, without Amazon ever asking permission to do so.
  
  Also also, without Amazon bothering to sync the data properly from your online store so that they don't, for example, still list long-discontinued products for sale.
  
  Store owners are not happy.
  
  
  
• NBC News is intensifying a collapse of trust online.  (NBC News)
  
  No, they didn't have a brief moment of self-awareness.  They're blaming three years of AI for thirty years of their own failings.
  
  
  

Musical Interlude

Posted by: PatBuckman at Monday, January 12 2026 02:29 AM (rcPLc)

Posted by: normal at Monday, January 12 2026 07:47 AM (Sbqr6)

Hide Comments | Add Comment