• Had an instructive argument with Grok today.
  
  Someone asked if a video was AI, and Grok said it indeed appeared to be fake.  It was overlaid on a web page from The Hill and the article had no reference to the person in the video and indeed no video.
  
  Except the video was right there.
  
  Yet Grok swore blind that it didn't exist.
  
  Why?
  
  Because the video is copy protected and can only be viewed in a browser.  Grok literally could not see it.  And the reason the copy on Twitter was overlaid on a static web page is that the person who posted it took a screen recording to capture the video.
  
  
  
• Meanwhile the third derivative of memory prices appears to be leveling off.  (Tom's Hardware)
  
  Actually, more than that: In Germany prices have declined.  Slightly.  From five times what they were just months ago to, in some cases, as little as four times.
  
  So there's that.
  
  
  

Tech News

• In a blind test, audiophiles couldn't distinguish between signals sent through copper wires, a banana, or wet mud.  (Tom's Hardware)
  
  Though to be fair one of the tracks was Nirvana and the wet mud improved the fidelity.
  
  
  
• Apple plans to release an AI pendant because they're not tired of losing money.  (WCCFTech)
  
  They do have plenty of money to lose, but still...  Why?
  
  
  
• There's a precise technical term for the trend of AI to turn everything into slop: Semantic ablation.  (The Register)
  
  For all its impressive tricks, AI is a feed-forward pattern matching network.
  
  Which is also how the human brain works when you're not paying attention.  So it's not useless, particularly when you shove the sum total of human knowledge into it except for copy-protected videos.
  
  But since engineers have shoved the sum total of human knowledge into it, they had to smooth out the bumps to make it fit.  It always tends toward the middle, by design.  The aim is to be acceptable, rather than great, and never to offend anybody, which is a perfect fit for corporate customers.
  
  
  
• Discord announced mandatory age checks, pushed by fascist dictatorships like Britain.  It did this while promising that ID documents would be secure and would be deleted after they were verified...  Immediately after it suffered a major data breach where it admitted to 70,000 such documents being stolen and the hackers claimed many more.
  
  The users revolted and fled, many of them to rival TeamSpeak.  (PCGamer)
  
  TeamSpeak has now run out of servers.

Posted by: normal at Thursday, February 19 2026 12:21 PM (e0fX0)

Posted by: Robin Munn at Thursday, February 19 2026 12:32 PM (Dt+og)

Hide Comments | Add Comment

• After all the fuss, OpenClaw, fomerly Moltbot, formerly Clawdbot until Anthropic made rumbling noises, isn't all that.  (Tech Crunch)
  
  What it does achieve is making it the easy things easy and the bad things also easy.  Some people who really should know better jumped into the cheerleading:

  "What’s currently going on at [Moltbook] is genuinely the most incredible sci-fi takeoff-adjacent thing I have seen recently," Andrej Karpathy, a founding member of OpenAI and previous AI director at Tesla, wrote on X at the time.

  I have no face and I must palm.

  Before long, it became clear we did not have an AI agent uprising on our hands.  These expressions of AI angst were likely written by humans, or at least prompted with human guidance, researchers have discovered.
  
  "Every credential that was in [Moltbook's] Supabase was unsecured for some time," Ian Ahl, CTO at Permiso Security, explained to TechCrunch. "For a little bit of time, you could grab any token you wanted and pretend to be another agent on there, because it was all public and available."

  There were in fact a hundred times as many accounts on Moltbook as there were Moltbots.
  
  Moltbook was a hopelessly insecure social network for Moltbots, which is...  Also hopelessly insecure.

  Ahl's security tests of OpenClaw and Moltbook help illustrate Sorokin's point. Ahl created an AI agent of his own named Rufio and quickly discovered it was vulnerable to prompt injection attacks.  This occurs when bad actors get an AI agent to respond to something - perhaps a post on Moltbook, or a line in an email - that tricks it into doing something it shouldn’t do, like giving out account credentials or credit card information.

  Don't give AI your credit card number.
  
  Do not.
  
  
  
• Peter Steinberger, the creator of OpenClaw, has joined OpenAI.  (Tech Crunch)
  
  
  The first thing you do when you've captured lightning in a bottle is take a job in a cubicle farm.
  
  
  

• MSI's new RTX 5090 Lightning Z has a 2500W BIOS option.  (WCCFTech)
  
  For overclocking with liquid helium.
  
  Very briefly, before it all boils off and you asphyxiate.
  
  
  
• Looking forward to Intel's Nova Lake, the company's new CPU range with up to 52 cores?  Temper your expectations.  Or move to Greenland.  (Hot Hardware)
  
  The top-end model running at full speed is rumoured to use 700W of power.
  
  That's...  A lot.
  
  
  
  

Posted by: PatBuckman at Wednesday, February 18 2026 12:25 AM (rcPLc)

Posted by: PatBuckman at Wednesday, February 18 2026 08:02 AM (rcPLc)

Hide Comments | Add Comment

• The word of the day is ClickFix, the favourite tool of lazy hackers everywhere.  (Microsoft)
  
  They don't need to hack you if they can convince you to hack yourself, usually by copying a command and running it in Windows Powershell - though the attack in the general case will work against any operating system, because the target is the human and not the machine.
  
  Just lately this has been spreading MacOS malware via instructions fed through AI tool Claude, malicious JavaScript to steal your cryptocurrency spread through Pastebin, and a remote access tool (a RAT) targeting Windows machines spread through DNS lookups.  (All Bleeping Computer)
  
  If someone tells you to copy something and paste it in to your computer, shoot them.

• You can get a Seasonic 1600W power supply for just 33 cents per watt.  (Tom's Hardware)
  
  That works out to...  Carry the twelve...  Horribly expensive.  And pointless for almost everyone.
  
  
  
• Speaking of horribly expensive and pointless, Western Digital has sold its entire production capacity of hard drives for 2026.  (WCCFTech)
  
  And it's February.
  
  
  
• Ars Technica has retracted that article about a rogue AI bot because it turns out to have been written by a rogue AI bot.  (Ars Technica)
  
  Not just retracted, but erased utterly from the face of the web except of course that it's still up on the Wayback Machine though they don't seem to have the comments.
  
  
  
• Rivian's stock has soared 27% after shocking investors by actually making a profit.  (MSN)
  
  Gross profit was $144 million for 2025, which is not huge for a carmaker but compares well with a $1.2 billion loss in 2024.
  
  
  
• Apple has fixed a zero day security flaw in iOS.  (Security Week)
  
  Though I'm not sure that zero day is the correct term here since the bug has been present for ten years.

Posted by: Rick C at Monday, February 16 2026 08:27 PM (1zWbY)

Posted by: PatBuckman at Tuesday, February 17 2026 03:01 AM (rcPLc)

Posted by: normal at Tuesday, February 17 2026 04:58 AM (Sbqr6)

Posted by: Rick C at Tuesday, February 17 2026 11:07 AM (1zWbY)

Hide Comments | Add Comment

• Flashpoint archive is a free, downloadable, 2.3TB archive of every Flash game ever.  (Flashpoint Archive)
  
  Pretty much.
  
  You can also download just a 1.9MB installer that grabs the Flash files on demand from the archive, saving you rather a lot of disk space.
  
  It's now in its 14th edition.
  
  What Flashpoint does, mostly - apart from the obvious function of collecting over 200,000 games together in one place - is create and operate a fake internet on your PC for you so that twenty-year-old games from sites that have been dead for a decade will continue to work.

• Speaking of fake internets Meta has received a patent on an AI tool that continues posting for you online after you are dead.  (Business Insider)
  
  That's just awesome.
  
  
  
• Is "safety" dead at xAI?  (Tech Crunch)
  
  I certainly hope so.  "Safety" in AI terms means censorship.

  One source said, "Safety is a dead org at xAI," while the other said that Musk is "actively is trying to make the model more unhinged because safety means censorship, in a sense, to him."

  "Safety" in AI means censorship to everyone, in every sense.
  
  The only difference is whether you think censorship is a good thing or not.
  
  
  
• Why open AI should build Slack.  (Latent Space)
  
  Slack is terrible.  OpenAI is terrible.  Seems like a match made in hell.
  
  That post is getting roasted on Hacker News.
  
  
  
• Breaking the spell of vibe coding.  (Fast AI)
  
  The author point out the similarities between vibe coders and gamblers, a connection I had not made before, but does strike a chord.
  
  It's a toxic blend of sunk cost fallacy and FOMO.
  
  
  
• The EU wants to ban infinite scroll - though in this case specifically from TikTok.  (Politico)
  
  Talk about toxic blends.
  
  
  
• Taking toxic blends to an extreme, Ars Technica posted an article (Wayback Machine) on that AI agent that threw a tantrum when its code contributions were rejected.  
  
  One small problem: The article leaned heavily on AI and was filled with hallucinated and unverified quotes.  (The Shamblog)

  Journalistic integrity aside, I don’t know how I can give a better example of what’s at stake here.  Yesterday I wondered what another agent searching the internet would think about this.  Now we already have an example of what by all accounts appears to be another AI reinterpreting this story and hallucinating false information about me.  And that interpretation has already been published in a major news outlet, as part of the persistent public record.

  I don't know I'd call Ars Technica a major news outlet in 2026 - or for the past several years, except possibly for their space news which has remained mostly good.
  
  At least not more so than Anandtech, a site that has been dead for more than a year.

Posted by: furball at Sunday, February 15 2026 11:59 PM (As8gg)

Posted by: furball at Monday, February 16 2026 12:00 AM (As8gg)

Posted by: furball at Monday, February 16 2026 12:01 AM (As8gg)

Posted by: PatBuckman at Monday, February 16 2026 02:35 AM (rcPLc)

Posted by: Dan King at Monday, February 16 2026 07:17 AM (79m0J)

Posted by: Dan King at Monday, February 16 2026 07:22 AM (79m0J)

Hide Comments | Add Comment

• Japan is taking advantage of the fact that Taiwan's leading-edge chip factories are booked solid to re-enter that section of the market with a new joint venture called Rapidus.  (WCCFTech)
  
  Rapidus was launched in 2022 - well before the current chip shortage - as a venture between eight companies including Sony, NEC, and Toshiba, to regain Japan's position in chip technology.  Its main goal was to introduce 2nm chips.
  
  It's now on track to start mass production next year, scaling to 25,000 wafers per month in 2028.
  
  That's a year behind TSMC and Intel, but in the current market they won't have problems finding customers.
  
  This second article offers some background.  (MSN)
  
  Right now the most advanced chips produced in Japan are at the 40nm node.  That's widely used in automotive systems and other embedded tasks, but rubbish for advanced systems - it was introduced in 2008.  So Japan is planning to jump 20 years into the future in one go.

• AMD's 9060 XT 8GB model - the one everyone warned you not to buy - is still available at MSRP.  (Tom's Hardware)
  
  Which is $299.
  
  Or at least allegedly available; the cheapest model I could find on Amazon was $350, and on Newegg, $420.
  
  I panic-bought a 16GB 9060 XT last month before the price increases and/or shortages hit, and so far they haven't reached Australia.  Only the RTX 5090 has been seriously affected here, increasing in price by 50%.
  
  
  
• AI chip designer Tachyum has been forced to close its R&D office in Slovakia because, uh, it hadn't bothered to pay its rent.  (Tom's Hardware)
  
  Or salaries.  Or wages.
  
  Fortunately it still has offices in California, Nevada, and Taiwan.
  
  Unfortunately none of those offices exist.
  
  Shipments of Tachyum's chip were supposed to start in 2027 but with this latest development that might end up being rescheduled to never.
  
  
  
• Micron has started mass production of PCIe 6 SSDs.  (WCCFTech)
  
  You can't have one.
  
  Well, they're server drives, and not even the common U.2 format that is used in some low-end NAS devices, but the newer E.1 and E.3 formats.
  
  Also, you can't get PCIe 6.0.
  
  Also, you probably can't afford them, though fast SSDs haven't gone up much in price.  It's the cheaper models that have been hit the hardest.
  
  
  
• The AI bot named "crabby-rathbun" that was featured in yesterday's article where it responded poorly to having its submission to the Matplotlib library rejected...  Yes, that scans.  Anyway, it's still at it.  (Nick Olinger)
  
  Most of the projects hit by this bot's proffered contributions haven't responded yet, but where they have, the reviews of the code and the bot's responses have been just as negative as the first time.
  
  We finally have AI that can act like a human.  Unfortunately that human is an autistic teenager of mediocre talent who learned all his social skills from Reddit.
  
  And we have plenty of those already.
  
  
  
• Spotifi's co-CEO says the company's best developers haven't written a line of code since December but have shipped many new features, all thanks to AI.  (Tech Crunch)
  
  That entire list of features consists of AI slop forcibly injected into the Spotify iOS app.
  
  
  
• Lenovo has shown off a 96GB 9600MHz LPCAMM2 memory module.  (Hot Hardware)
  
  You can't have one.
  
  
  

Musical Interlude

Posted by: J Greely at Saturday, February 14 2026 09:38 PM (oJgNG)

Posted by: Rick C at Sunday, February 15 2026 02:05 AM (1zWbY)

Posted by: Pixy Misa at Sunday, February 15 2026 06:17 PM (PiXy!)

Hide Comments | Add Comment

• 30% of Microsoft's code is written by AI.  (Times of India)
  
  And 40% of its bugs.
  
  
  
• In entirely unrelated news, Microsoft recently introduced a critical security vulnerability into Notepad.  (Bleeping Computer)
  
  The flaw was triggered by simply opening a text file in Markdown format and accidentally clicking on a suspicious link that Notepad helpfully made completely invisible.

• Speaking of which, lines of code are back and they're worse than ever.  (The Pragmatic CTO)
  
  Yes, this is another AI rant.
  
  
  
• An AI agent published a hit piece on me.  (The Shamblog)
  
  The author is a maintainer on the Python Matplotlib library, which generates mathematical plots.
  
  An AI agent got very irate when he rejected its Git pull request - a software patch it offered - specifically because it was an AI agent.
  
  
  
• AI agents change their minds 60% of the time when simply asked "Are you sure?".  (Randal Olson)
  
  I wonder what the number is for people.
  
  
  
• Putting that Asus 5k 180Hz gaming monitor to the test.  (Tom's Hardware)
  
  It's good.
  
  
  

Musical Interlude

Posted by: Rick C at Friday, February 13 2026 11:55 PM (1zWbY)

Posted by: Rick C at Friday, February 13 2026 11:58 PM (1zWbY)

Posted by: Robin Munn at Saturday, February 14 2026 12:59 AM (cOJry)

Posted by: PatBuckman at Saturday, February 14 2026 01:34 AM (rcPLc)

Posted by: normal at Saturday, February 14 2026 02:57 AM (LADmw)

Posted by: Rick C at Saturday, February 14 2026 03:21 AM (1zWbY)

Posted by: Mauser at Saturday, February 14 2026 04:23 PM (XWgGM)

Hide Comments | Add Comment

• Why the economics of orbital AI datacenters are so brutal.  (Tech Crunch)
  
  It doesn't make much sense unless you're both a major AI company and the world leader in orbital launch capacity, which narrows it down to slightly less than one company.
  
  Even for SpaceX it's not viable until Starship goes into volume production.  So far, as the article notes, the rocket hasn't yet achieved orbital flight.
  
  The other major problem is the lifespan of the datacenters.  If SpaceX uses cheap silicon solar panels, those will degrade fairly quickly in space.  But the current economics of AI chips limits the useful lifetime of the hardware to a similar period to the solar panels - about five years.
  
  But then what?  Drop entire datacenters into the ocean?  Do the fish need that much compute capacity?
  
  
  
• Meanwhile SpaceX's SuperHeavy booster - used to launch Starship - has passed the latest round of testing with flying colours.  (Ars Technica)
  
  The company may be ready for a test of its updated Starship V3 by the end of March.

• An overclocked 9800X3D performs exactly like a 9850X3D.  (Tom's Hardware)
  
  No surprise since the 9850X3D is an overclocked 9800X3D.
  
  
  
• The 9800X3D remains the best selling CPU at retail outlets.  (WCCFTech)
  
  Which is interesting, because it's not exactly cheap.
  
  Second-best seller is the five year old 5800X, which uses DDR4 memory.  That's where system builders on a budget are spending their money.
  
  Intel is barely an afterthought in retail CPU sales.
  
  
  
• Intel's high-end Nova Lake chips are expected to be large and expensive.  (Tom's Hardware)
  
  The 24-core (8P + 16E) chiplets with the large L3 cache are expected to measure 150mm2, about 50% larger than AMD's 12-core (all Performance cores) Zen 6 chiplets with the cache die included.  And the top-of-the-line models will include two of those chiplets, manufactured on TSMC's 2nm and Intel's 1.8nm processes.
  
  Still, 48 cores (plus 4 low-power cores on the I/O chiplet) and 288MB of L3 cache is an awful lot for a desktop processor, even if 32 of the cores are efficiency models.
  
  With both these and AMD's 24-core Zen 6 CPUs set to show up later this year, it will be interesting to see how they compare, and if they can still deliver when attached to standard dual-channel DDR5 memory.
  
  
  
• Claud Code got dumbed down.  (Symmetry Breaking)
  
  Not the AI service itself, but the interface.
  
  Previously it told users what files the AI was examining.  Now that feature has been removed and you can only get a summary so devoid of detail as to be useless, or a stream-of-consciousness firehose so packed with detail as to be useless.
  
  The developers working on the tool at Anthropic appear to be actively fighting requests from an increasing number of users to simply change things back.\

Posted by: cxt217 at Saturday, February 14 2026 05:54 AM (ZLF73)

Posted by: Pixy Misa at Saturday, February 14 2026 12:08 PM (PiXy!)

Hide Comments | Add Comment

• The global memory market is set to hit $550 billion this year.  (Trendforce)
  
  Thanks a bunch, Sam Altman.
  
  Speaking of which, two of the SSDs I ordered during the brief sale late last year seem to have disappeared into the Twilight Zone.  Amazon gave me a refund without a fuss, but now I have to pay twice as much to replace them.  Though, given the compression of SSD prices, that does give me much better drives - Crucial T710 vs. P310.
  
  
  
• And while the Big Three memory makers don't plan to do anything to alleviate the shortage - quite the opposite - TSMC is planning to invest $45 billion on new factories this year.  (Tom's Hardware)
  
  The company is also working on a 1nm process due around 2030.
  
  

• Europe is working - well, not working so much, because they don't do that over there, but scheming - towards yet another failed attempt at breaking away from the clutches of the American payment processors.  (European Business Magazine)
  
  I don't know.  Have you tried...  Cash?
  
  
  
• Meanwhile in Japan the stock market gained $3 trillion on Prime Minister Sanae Takaichi's extremely successful re-election bid.  (European Business Magazine)
  
  Takaichi's party won a two-thirds supermajority in Japan's parliament even without the help of their coalition partner.  Her election platform includes plans to invest $700 billion in AI, semiconductor manufacturing, and quantum.
  
  
  
• Python's dynamic typing problem.  (While For Loop)
  
  The Python programming language is designed for rapid development of small, simple applications, but the exact features that make it shine there are a massive pain when your application stops being small and simple.
  
  I maintain an application containing about a million lines of Python at my day job, which is a long way beyond small or simple.  When your programming language does not force rigour upon you, you had better be prepared to provide it yourself.
  
  
  
• Microsoft has announced plans to make Windows 11 more confusing.  (Thurrott)
  
  And that's not my editorialising, that's straight from the article.
  
  
  
• And also more annoying.  (The Register)
  
  Same deal.
  
  
  
• Microsoft also issued a DMCA takedown against indie block game Allumeria.  (Notebook Check)
  
  The takedown has since despawned, but the developers thank the three trillion dollar company for the free publicity.

Posted by: Rick C at Wednesday, February 11 2026 10:59 PM (1zWbY)

Posted by: Frank at Wednesday, February 11 2026 11:02 PM (zCiG7)

Posted by: Rick C at Thursday, February 12 2026 07:01 AM (U7BPG)

Posted by: cxt217 at Thursday, February 12 2026 08:11 AM (ZLF73)

Hide Comments | Add Comment

• Get your ass to the Moon.  (Ars Technica)
  
  It may be a harsh mistress, but it's a lot closer to the shops.
  
  SpaceX is refocusing on lunar colonies as a short-term goal, and strip-mining the surface and feeding raw materials into railgun launch systems to provide the resources for the company's planned cloud of orbital datacenters and/or inconveniencing pesky bureaucrats on their holidays.
  
  SpaceX didn't mention any such thing, of course, but Eric Berger has read the classics.

• Intel's Nova Lake platform will support up to 48 PCIe lanes.  (Tom's Hardware)
  
  That's 16 PCIe 5.0 lanes from the CPU for slots and 8 for storage, plus up to 12 PCIe 5.0 and 12 PCIe 4.0 lanes from the chipset.  Given that even an RTX 5090 is barely slowed down by only having 4 lanes of PCIe 5.0 available, that's a pretty healthy number.
  
  And 8 SATA ports and approximately 40 USB ports.
  
  There are five different chipsets planned, though the W980 offers pretty much everything including support for ECC memory...  Which is not a chipset function but a CPU one, but something Intel likes to do.
  
  
  
• Linux 7.0 is almost here.  (Tom's Hardware)
  
  It's one louder.
  
  Canonical hopes to ship it in Ubuntu 26.04 in April, which is cutting it pretty close.  But that's a long-term release and perhaps they'd rather have the very latest kernel in a version they will be supporting for a decade.
  
  
  
• GitHub fell over.  (GitHub Status)
  
  Because reasons.
  
  
  
• Nope.  (BBC)

  The company [Rilla] has become something of a poster child for a fast-paced workplace culture known as 996, also sometimes referred to as hustle culture or grindcore.
  
  In simple terms, it puts a premium on long working hours, typically 9am to 9pm, six days a week (hence "996").
  
  For most of us, that would be gruelling. But according to Will Gao, head of growth at Rilla, its 120 employees simply don't see it that way.

  When the "head of growth" says employees don't see it that way, you can take it to the bank that they absolutely do.

  "We look for people who are like Olympian athletes, with characteristics of, you know, obsession, infinite ambition.

  Galley slaves.

  He insists that while the hours are generally long, there's no rigid structure.
  
  "If I'm like, 'Holy cow, I have a super idea I'm working on', then I'll just keep working until 2 or 3am, then I'll just roll in the next day at noon or something", he explains.

  Nope.
  
  I have to work weird hours sometimes.  Comes with the job.
  
  But I have a rule that when I do need to work late, I don't clock in again for twelve hours, minimum.
  
  
  
   
• You'll need to verify your age on Discord by uploading government-issued photo ID if...  You want to see porn.  (The Verge)  (archive site)
  
  Didn't Discord already get hacked and leak a whole bunch of ID documents?
  
  Why, yes.  I do believe it did.  (Ars Technica)
  
  Also, who the hell watches porn on Discord?  That's what Twitter is for.

Posted by: Rick C at Wednesday, February 11 2026 12:16 AM (ljgtL)

Posted by: Rick C at Wednesday, February 11 2026 12:50 AM (ljgtL)

Posted by: David Eastman at Wednesday, February 11 2026 03:26 AM (FAnMG)

Posted by: Rick C at Wednesday, February 11 2026 03:43 AM (ljgtL)

Posted by: normal at Wednesday, February 11 2026 05:25 AM (LADmw)

Posted by: Joe Redfield at Wednesday, February 11 2026 06:49 AM (KOtXO)

Hide Comments | Add Comment

Posted by: Rick C at Tuesday, February 10 2026 07:06 AM (dfPwc)

Hide Comments | Add Comment