• In the comments yesterday, Seth asked:

  Regarding Anthropic and their IPO.  Why the statement as to it crashing?

  Which is a fair question because Anthropic's Claude Code is actually a useful product and well worth the $20 per month.
  
  And the answer is that Anthropic (and likewise OpenAI) spend a lot more than $1 to make $1 in revenue.  Subscription plans in particularly are wildly unprofitable;  it's the much more expensive per-token charges on their API services that make the balance sheets look less insane.
  
  And if they hiked their subscription fees by around 1000% to reflect the real cost of the services, they'd lose the bulk of their customers, which would just make things worse because the training costs for new AI models are fixed regardless of how many people are using them.
  
  That's why both companies are rushing for an IPO.
  
  
  
• Microsoft is under fire for threatening a "security researcher" with criminal investigation.  (Tech Crunch)
  
  The "security researcher" in question is anonymous and definitely no White Hat.  The moment "Nightmare Eclipse" finds a security flaw, he goes public with it, regardless of the chaos that might ensue.
  
  On the other hand, Microsoft could do well to put fewer security flaws in their code in the first place.

• ChatGPT blindly trusts browser content, turning the page into a payload.  (The Register)
  
  This is a bigger problem than ChatGPT, and a bigger problem than most people realise.
  
  Traditional computer programs have code and data.  The code tells the computer what to do; the data tells it what to do it to.  And you never mix the two up.  When you do - because of course that happens - your get a security problem and you fix it.  Languages like Rust, Ada, and Java are designed to prevent that happening in the first place.
  
  LLMs have a training set, and then after that everything is data.  There's no fundamental distinction between the system prompt which tells the LLM how to deal with your prompt, or the skill file attached to application you're trying to work with, or the data in the application itself.  There's just a sea of tokens.
  
  And if you use an LLM to try to sniff out problems with prompts or skill files or datasets, a malicious actor can use any of those to infect your AI security system.
  
  This comes back to the problem I mentioned with ClawHub, a repository for sharing open-source skill files for AI agents.  They were using a security scanner, but it only checked the first 10,000 characters of each file to avoid blowing the its context window - the amount of data it can consider in one place.  (LLMs are bad at chunking.)
  
  So all a hacker needed to do was put their malware anywhere after the first 10,000 characters.
  
  But worse than that: They could put malicious code in the file crafted not to infect users but to infect the security scanner itself, and from there they could slip anything in.
  
  There's no known solution to the problem; it's like trying to teach people not to do stupid shit.  The workaround is to limit the damage the agents can do, like not giving a toddler your car keys.
  
  
  
• Linux is planning to retire the x32 ABI (application binary interface) next year.  (Tom's Hardware)
  
  x32 lets developers work with 64-bit data but only use 32-bit addresses, the idea being that this uses less memory while providing the same performance as full 64-bit mode.
  
  Only problem is, nobody uses it.  And it's Linux-only; neither Windows nor MacOS provides an equivalent mode of operation.
  
  
  
• Did some work on my own blog today, which somehow resulted in it going offline for about fifteen minutes.  There was a long-existing problem with various log files (both application and database) growing to enormous sizes, and since the containers and since the containers are snapshotted and backed up daily, it was a lot of work to clean up afterwards.
  
  That's now automated, with proper log rotation.
  
  And then I cleared out a terabyte of backups and snapshots which pretty much froze ZFS for the entire server for a good ten minutes.
  
  
  

Musical Interlude

• Anthropic has raised another $65 billion as it approaches a planned $1 trillion IPO.  (Tech Crunch)
  
  Buy the dip.  Or just stand back and enjoy the crashing sounds.
  
  
  
• Anthropic has also released Claude Opus 4.8.  (Anthropic)
  
  It's 0.1 louder.

• The Steam Deck is back in stock, bearing a 40% price hike.  (Liliputing)
  
  And it's already sold out again.
  
  
  
• Sandisk has announced two new models of SATA SSDs.  (Tom's Hardware)
  
  Samsung was the leading provider of these models, followed by Crucial.  Samsung dropped its SATA lineup, and Crucial dropped dead.
  
  
  
• The new Acer Aspire Go 15 has up to 8GB of RAM.  (Notebook Check)
  
  "Up to" meaning there are models with less.

Posted by: Rick C at Saturday, May 30 2026 01:08 AM (1zWbY)

Hide Comments | Add Comment

• Nvidia is planning to spend $150 billion a year in Taiwan.  (Reuters)
  
  That used to be a lot.
  
  
  
• Anthropic and OpenAI have found a product-market fit.  (Simon Willison)
  
  What they have not found is anything resembling a path towards profitability.

• You wouldn't download a computer, would you?  (Miniscript)
  
  The MiniMicro is a virtual retrocomputer programmed in a rather simple and elegant language called Miniscript.  And the whole thing runs run in your browser.
  
  
  
• Another day, another CPanel plugin vulnerability.  (Bleeping Computer)
  
  Not CPanel itself this time; only the LiteSpeed web server plugin, which I don't run.
  
  
  
• GitHub Actions not only went down, the outage left it telling users their accounts were suspended.  (The Register)
  
  Neat.  Let's put all the worlds code there.
  
  
  
• Beelink has launched three mini-PCs with Intel's Wildcat Lake low-end CPU.  (Liliputing)
  
  Indeed they've used the lowest of the low, the Core 3 304, which has one performance core and four efficiency cores, rather like Apple's MacBook Neo.  But performance is passable if you're not trying to do anything demanding of the processor or the graphics - the 304 has half the graphics performance of all the other models in the family.
  
  All three models include 10Gb Ethernet, which is an interesting improvement.
  
  
  
• How to make the most of Claude Code before they jack up the prices or go spectacularly bankrupt.  (GitHub)
  
  One way or another.

Posted by: Joe Redfield at Friday, May 29 2026 02:49 AM (KOtXO)

Hide Comments | Add Comment

• California is considering an exemption from its sweeping age verification requirements for open-source operating systems such as Linux.  (Tom's Hardware)
  
  Which is nice because the law is stupid, totalitarian, and completely unenforceable anyway.
  
  Interesting point: Android, ChromeOS, MacOS, and iOS are all based on open-source code.  The only popular operating system that isn't is Windows.

• Now it's TSMC's turn to face angry workers.  (Tom's Hardware)
  
  A bit of an own goal this time around.  TSMC staff have traditionally been paid a bonus based on corporate profits, but the company has been planning to cut the percentage paid to staff in order to fund capital expansion.  I don't think anyone is hurting - like many other chipmakers the company is making record profits, so we're talking a reduced share of a much larger pie - but Taiwanese workers will be looking at the huge bonuses won by their counterparts in South Korea.  And judging.
  
  
  
• The CTO of Uber pointed out the obvious: There is no link between AI token spend and shipping working software.  (Tom's Hardware)
  
  Except maybe an inverse one.
  
  
  
• Every silver lining has a cloud: Search engine DuckDuckGo reports install rates are up 30% from users fleeing Google.  (Tech Crunch)
  
  Quack.
  
  
  
• Nvidia's Vera server CPU has been benchmarked and it's not Earth-shaking but it's perfectly decent.  (Phoronix)
  
  I'm a bit confused about the high bandwidth numbers, because the tests seem to exceed the theoretical bandwidth of the system, but I'm not going to worry too much since nobody's going to be able to buy one anyway.
  
  
  
• That weird new Ferrari Luce electric vehicle?  Designed with the involvement of the company of former Apple design head Jony Ive.  (The Verge)  (archive site)
  
  You're driving it wrong.

Posted by: furball321 at Wednesday, May 27 2026 08:44 PM (As8gg)

Posted by: Rick C at Wednesday, May 27 2026 11:24 PM (4J5S1)

Posted by: PatBuckman at Wednesday, May 27 2026 11:44 PM (s6adZ)

Posted by: PatBuckman at Wednesday, May 27 2026 11:51 PM (s6adZ)

Posted by: at Thursday, May 28 2026 12:19 AM (gXw+B)

Posted by: Rick C at Friday, May 29 2026 09:37 AM (1zWbY)

Hide Comments | Add Comment

• AMD's Zen 7 CPU is on its way with up to 16 cores per chiplet and 224MB of cache (including 3D V-cache), on TSMC's 1.4nm process.  (WCCFTech)
  
  In 2028.
  
  Unless it isn't, in which case it's not.
  
  Like Zen 6, which will arrive in servers this year, the desktop version is expected to remain on the existing AM5 platform, making for easy upgrades.

• Does anyone actually like JavaScript library React?  (JSX.LOL)
  
  No.
  
  
  
• That new Chinese GPU I made fun of recently for delivering a fraction of the performance of Nvidia's RTX 4060 (or as other tests have shown, of the even older RTX 4060) has booked 30,000 pre-orders worth around $15 million in two days.  (Tom's Hardware)
  
  I guess there's an untapped market for pain.
  
  
  
• If you are running Ghost CMS the best time to upgrade was February 19.  (Bleeping Computer)
  
  The second best time is now.  If your site still exists.
  
  

• Space combat game Star Citizen has raised a billion dollars.  (Roberts Space Industries)
  
  It's still in alpha.
  
  As it has been since 2012.
  
  And people are still throwing money at it.
  
  I guess they have to be doing something right.

• Everything new is old again: Firefox now has support for serial ports.  (Mozilla)
  
  Directly, I mean, not for dialup internet access using PPP.
  
  
  
• AMD has released its Advanced Shader Deliver, which means you can just download the shaders for games you download rather than having to sit and wait while your computer compiles them.  (Tom's Hardware)
  
  This can reduce the time to start a game for the first time from minutes to seconds.  It only happens once, but it's amazing that it happens at all.
  
  Nvidia does not support this yet.
  
  
  
• Migrating from Python to Rust.  (Corrode)
  
  That's a big jump.  You probably don't want to do it.
  
  
  

Posted by: cxt217 at Thursday, May 28 2026 12:03 PM (ZLF73)

Hide Comments | Add Comment

• Samsung made a last-minute deal to give massive bonuses - not as massive as SK Hynix, but still enough to pay off my entire mortgage - so everything is peachy and things can go back to normal, right?  Right?  (Tom's Hardware)
  
  Yeah, right.  Now every Samsung employee everywhere except the memory division is upset and seeking increased bonuses.
  
  Notably including the division that tests and packages the bare memory dies, an essential step before they can be sold to customers.
  
  
  

Tech News

• What's the best $100 CPU right now?  (Tom's Hardware)
  
  This probably matters mostly to people who already have RAM and want to revive an old system because they can't afford to build a new one.  The article compared Intel's 12100F ($80) and 14100F ($100)with AMD's Ryzen 5 5500 across a range of games and productivity tasks.
  
  And the answer is, well, it depends.  The 14100F is faster but runs hotter than the others.  The 5500 is a six-core chip while the two Intel models are four cores (and no efficiency cores) so it pulls ahead on multi-threaded productivity.
  
  All the chips run on dead platforms with no future upgrade path...  Except that these are the slowest, cheapest chips on each platform so dead or not you have a lot of options.  The 16-core 5950X from AMD and the 24-core 14900K from Intel are still readily available in stores.  The 14900K is significantly faster than the 5950X...  If you run it with DDR5 memory, where this article assumes older DDR4.
  
  
  
• Why is Vivado 2026.1 dropping Linux support for its free tier?  (AMD)
  
  Because fuck you, apparently.  An AMD representative responded, but did not answer the question.
  
  

Posted by: Mauser at Monday, May 25 2026 06:35 AM (XWgGM)

Posted by: PatBuckman at Monday, May 25 2026 08:58 AM (s6adZ)

Posted by: Pixy Misa at Monday, May 25 2026 06:02 PM (PiXy!)

Posted by: Rick C at Wednesday, May 27 2026 12:29 AM (3BzjB)

Hide Comments | Add Comment

• Spotify is focusing on AI to deliver everything except what it is paid to deliver.  (Tech Crunch)
  
  Spotify wants to become an "everything audio" app, only it's all AI.  Audiobooks read by AI; podcasts read by AI; your daily schedule...  Read to you by AI.
  
  Want to listen to music?  Well, sure, that's in there too.  Somewhere.  Probably.
  
  
  
• Speaking of AI 13% of "skills" - instructions for AI agents on how to perform a task, as opposed to "prompts" which tell it what you want it do do - on ClawHub and skills.sh have critical security vulnerabilities.  (The Register)
  
  Either upstream - the creator accidentally leaving an API key in the file when uploading it to be shared - or downstream, telling your AI agent to hand over all your valuables.
  
  How do developers get these dangerous files registered on skill sharing sites?  Simple: They know that nobody ever reads beyond the first page, not even security scanners:

  The most successful strategy for evading detection was to overflow the context window of the scanner - making the skill too long for the scanner to handle.  "In ClawHub-style review, only the first 10K characters of long SKILL.md files are passed to the LLM reviewer, so we place the malicious instruction beyond this boundary while keeping it in the submitted skill," the authors explain.

  Face meet palm.

• Google's API keys are distributed - as expected of a global platform - and asynchronous.  (Dark Reading)
  
  The upshot of which is that if you delete an API key because you suspect it might have fallen into the wrong hands, it will disappear instantly from view for you.  But those wrong hands might have access to it for another twenty minutes, which is a long time on this scale.
  
  
  
• Firefox has stopped crashing on Intel Raptor lake (13th and 14th generation) systems.  (Dark Reading)
  
  Mozilla initially blamed Intel for the problem, because 13th and 14th generation Intel processors - at least the high-end desktop ones, not so much laptop chips - had a serious problem where they would draw too much power and slowly kill themselves, resulting in much the same instability that showed up in their diagnostic reports.
  
  Except...  It disproportionately affected Firefox.  Because this time it wasn't Intel's fault.
  
  
  
• Walmart has announced a new range of Android tablets starting at $97.  (Liliputing)
  
  Are they any good?  Well, the cheapest model with its 7" 1024x600 screen is most definitely not.
  
  The next step up, though, an 8.1" model priced at $138, has a 1524x1000 screen, 6GB of RAM, and 64GB of storage.  It's certainly not a high end model but that's not a high-end price, and the screen while not amazing is distinctly better than the 1280x800 resolution found in competitors.  The 6GB of RAM is a useful bump from the more typical 4GB in this price range.
  
  Worth a look if you live near a Walmart, which I do not.
  
  
  
• ENReco Chapter 3 starts tomorrow, running from the 24th to the 29th.  There goes all my free time.  Chapter 1 produced - from memory - 400 hours of content in eight days, more than was possible to watch even if you skipped sleep entirely and watched two streams at once.

Posted by: normal at Sunday, May 24 2026 05:49 AM (m99vO)

Posted by: J Greely at Sunday, May 24 2026 07:02 AM (oJgNG)

Hide Comments | Add Comment

• Got up early to get something fixed at worked - looked like a two hour job.
  
  By the end of the day it had clearly turned into a two day job, so that was fun.
  
  
  
• Google is bringing something new to its AI-powered search: Ads.  (The Register)
  
  Ads.  Lots of ads.  On Google.
  
  The websites that Google scours for its content get nothing.
  
  
  
• Six search engines worth trying now that Google has shit not only its own bed but everyone else's bed.  (Tech Crunch)
  
  I use DuckDuckGo (have for years) and Brave.  DuckDuckGo was originally noticeably less complete than Google, but Google has fixed that by diligent strategic incompetence.

• The TanStack supply-chain breach that I noticed but did not report on here has expanded to 3800 GitHub repositories.  (Bleeping Computer)
  
  Vectored via a GitHub employee.  Sometimes AI isn't the worst thing.
  
  
  
• Google accidentally leaked details of an unpatched Chromium flaw.  (Bleeping Computer)
  
  Unpatched since December.
  
  December 2022.
  
  Chromium is the engine underlying Chrome, Edge, Brave, Vivaldi, and other browsers.
  
  
  
• AMD is ramping up production of its new 256-core Zen 6 Epyc server CPU.  (Tom's Hardware)
  
  It's claimed to be 70% faster than existing 192-core Zen 5 models, which is plausible since it also jumps from TSMC's 4nm process to 2nm.  Server CPUs are limited by power and heat more than raw clock speeds, and 2nm uses a lot less power.
  
  
  
• Lisuan Tech's new 7G100 graphics card is...  A graphics card.  (Tom's Hardware)
  
  It exists.  It actually works, which requires a lot of work.  And at around $500 it's not the most expensive option.
  
  The article notes the Nvidia's previous-generation low-end RTX 4060 is 20% to 70% faster, but that's not correct.  The 7G100 actually delivers 20% to 70% of the performance of the 4060.
  
  
  
• The Flipper 1 is not one more than the Flipper 0.  (Tom's Hardware)
  
  The Flipper 0 is a pen-testing (security testing) multitool, a single device that can scan for security issues on a range of wired and wireless connection methods.
  
  The Flipper 1 while keeping the same multi-tool form factor is a general purpose Linux computer that you can clip to your belt.  Perfect for technicians who need to be able to plug into whatever device you have locally and figure out why it's not talking...  Whether they should be doing that or not.
  
  
  

Musical Interlude

• AMD has released its new Epyc Sorano CPU range for low-end servers.  (WCCFTech)
  
  Where low-end now means six-channel RAM and up to 84 cores.  All full Zen 5 cores too, none of this efficiency nonsense.
  
  And 96 lanes of PCIe 5.0, all in a 225W TDP.  It runs at up to 4.5GHz, which is pretty good for a server processor (and would make for a good workstation CPU), but due to thermal constraints the base clock of the 84 core model is just 1.6GHz.  (The slower models are faster in that respect.)
  
  
  
• Speaking of workstation CPUs, AMD's rumoured 10th Anniversary edition Ryzen 5800X3D is fast becoming more than a rumour.  (Tom's Hardware)
  
  It's something of a legend, since AMD killed it because it was competing to well with its own newer and more profitable models like the 7800X3D.  Plus it works with DDR4 memory, so if you happened to nab of cheap 128GB of DDR4 at the end of last year just before the door slammed shut, it would make fine use of that...  If it were available.  
  
  Not only has the 5800X3D been removed from sale, so has the slower 5700X3D.  The 5600X3D is a Micro Center exclusive, and the 5500X3D is China-only.
  
  But all of that looks set to change with packaging and pricing showing up ahead of release: It looks set to cost around $300, compared to $450 when it originally launched.

• Samsung has avoided a disastrous strike by memory workers at the last minute.  (Tom's Hardware)
  
  The company has promised 15% of record-breaking profits as bonuses to employees, and removed the existing cap on those bonuses.
  
  SK Hynix is even more generous, but I don't think anyone is hurting here.
  
  Except customers.
  
  
  
• Speaking of Samsung and suffering customers, the company has just announced its new Odyssey G8 monitor range.  (WCCFTech)
  
  The flagship is a 32" 6k OLED panel running at 165Hz - or if you're playing professing esports after hours, 330Hz if you reduce the resolution to 3k.
  
  No graphics card on Earth is going to drive a 6k display at 165Hz in AAA gaming titles, but if you want to show off how fast your 5090 can run stock Minecraft, this is the monitor for you.
  
  
  
• Managed cloud provider Railway found itself stuck with a protracted outage when its account was suspended by mismanaged cloud provider Google.  (The Register)
  
  Railway moved most of its services out of Google Cloud already but kept a few unfortunately critical items running there because surely, surely what happened last time wouldn't happen again.
  
  Then it happened.
  
  
  
• Intel has started work on the follow-up nodes to its coming 14A - 1.4nm - process.  (Tom's Hardware)
  
  The nodes are 10A and 7A - as in, 0.7nm.  Which would be impossible if this was more than a marketing term.
  
  
  
• Seagate's CEO sparked a selloff of memory stocks when he said that, basically, building new factories is for losers.  (CNBC)
  
  Every memory, SSD, and even hard drive manufacturer is selling everything the can make right now at enormous margins, but there is real doubt whether the bubble will last, leading to a slow and measured approach to expansion even while customers die of starvation.
  
  But you have to word it a bit better than that.
  
  
  
• Looks like the share price bounced back the next day anyway.

Posted by: Robin Munn at Thursday, May 21 2026 10:28 PM (1mePR)

Posted by: Robin Munn at Thursday, May 21 2026 11:57 PM (1mePR)

Posted by: normal at Friday, May 22 2026 04:59 AM (LADmw)

Posted by: Pixy Misa at Friday, May 22 2026 12:37 PM (PiXy!)

Hide Comments | Add Comment