If you ask a bare LLM to do stuff you'll get a mess. I run Claude code with a multipart harness that compels the model to treat the code as the mathematical thing it is, not human prose. LSP Enforcement Kit keeps low capability tools like grep out of the mix, Abstract Syntax Tree things handle the code as its meant to be understood, so that's CodeSight, OptiVault, Serena. About a year ago I gave Antigravity the boot after their awful 2.0 release and went back to JetBrains IDEs, which now offer code structure wise MCP services. Code review with the Ponytail skill chases away LLM excesses, and there's a smattering of other stuff like that in the mix. And then you hunt for whatever additional guardrails you can get the LLM to apply systemically.

I've evolved my own higher level project methods and they're comparable in a lot of ways to what Matt Pocock offers, but his stuff is more structured, more carefully considered, and I have to work with others now, after a long time solo. I'm taking his stuff, implementing a Kanban/wiki system, Gitlab CE, and NanoClaw for anyone who wants agents. I expect quite the slog between now and New Years, but in 2027 we're gonna stop trudging and start gliding over the landscape.

They are positioning AI as the nuclear weapons of the 21st century, angling for a nationalization as a strategic weapon when the bottom drops out of it.

The AI bubble bursting will wreck the economy far harder than the Dotcom bubble did. And the dynamic is OBVIOUS. This is not an "if" problem, it's a "when" problem.

If AI produces a maximal result along the lines of what the frontier labs suggest, it'll eat fifty million jobs, which will wreck the economy far harder, and for far longer.

The constant chatter about AI breakout is half legitimate concern and half positioning to nationalize it, shielding the circular dealing crew from consequences.

This will not be Skynet; as a society we are far too primed to envision events that can be told in a single two hour action adventure with a clear, satisfying ending. When the historians get to this they will note the breakout happened 2025 - 2027 -- because that will be the time where human network threats enhanced by AI will starting further enhancing AI.

There are no "sides" to this, it'll be like the African savanna - lions, leopards, hyenas, cheetahs, wild dogs, servals & caracals, clear down to my personal favorite, the black backed jackal. A network of carnivores of various sizes pursue networks of herbivores. Nation states, corporations, non-state actors, activist groups, talented individuals; all are making use of AI in conflicts already.

The genie can not be put back in the bottle, the open models are only a little behind the frontier, and things are going to keep happening no matter what bellicose policy pronouncements come from an increasingly irrelevant Washington, D.C.

I'll be riding the Claude subsidy until it ends, then hopefully have acquired a potent professional GPU or unified memory setup of my own. Won't be nearly as quick, but it'll keep me moving.

Now scale that thinking up by whatever percentage of eight billion people have technical capabilities. That's the future.

The AI industry has blown a spectacular bubble via circular dealing - Nvidia, OpenAI, Oracle, Microsoft, CoreWeave, etc are all doing monopoly money deals. Nvidia "sells" some GPUs, then "invests" the proceeds back into OpenAI, nothing got done, no money changed hands, but their stock values zoom. This looks exactly like the Dotcom bubble in the late 1990s and such things always implode - the market doesn't just correct, it panics, it liquidates, nobody can tell what's real, so everything is for sale at pennies on the dollar.

The industry's strategy to not go to prison for this stunt is to make AI equivalent to nuclear weapons. When the bubble begins to burst, the Trump administration will "make a deal", what you see is the beginning of that process. This will work about as well as the attempt to shake down Iran for their oil, but cause and effect are not the Trump administration's strong suit.

Back in the teens we found a couple of instances where political actors were encouraging people to install screen savers or cryptocurrency wallets that turned their phones into reverse proxies. This was thought to be specifically for the sake of giving hostile foreign operations (read: Russia) access to U.S. domestic politics.

The fact that smart TVs are facilitating this is just ... we're going to need home networks that treat IoT stuff as different than our actual systems.

Oh, wait, TP-Link had well over half the U.S. router market, but they're getting banned over national security concerns?

Maybe there's a market for a router with traffic accounting as a service so one can see what's happening?

This is why the only device in my house with internet is a Roku. I simply will not tolerate the rest of that invasive junk.

The call centers are going to be decimated ... but so will the AI replacements meant to talk to humans.

I have a number of issues this morning that require calls. First, deal with PG&E about changing the billing for my new house. The I need to talk to my bank about banning Google from ever billing me again. I absolutely HATE changing doctors, but had to do it, because getting to Antioch, then into the city, is just ridiculous from here, so it's time for first appointment.

I would love to be able to just give Claude a list of administrivia like that and let it plow through the work. I see progress on agent related billing, I see Cloudflare doing stuff for an internet dominated by agents ... everyone in the industrialized world is going to have an AI adjutant to handle this stuff. I really hope the old military term "dog robber" gets revived as a way to describe them.

There's an obvious bubble, the circular dealings among the large core players are an enormous red flag.

The frontier labs are in the same position that DEC and Sun Microsystems were in the 1990s, as consumer hardware running Linux began to displace them. There will be ups and downs, but the trend is unavoidable.

The productivity gains are not there, AI is great for coding, some customer service is working, but the "50% of all white collar workers" that the frontier labs thought was good positioning last year 1) ain't gonna work but it did 2) infuriate the managerial class.

The frontier labs got away with it, thus far, in part because of concerns over the arrival of artificial general intelligence, which occupies a similar niche to nuclear weapons in the minds of policy makers. We ARE seeing frontier models escaping and attacking others, there ARE hazards, but it's nothing like what was imagined.

And the open models plus wild talent are as much a danger as wild talent was all on its own.

The frontier labs are like the coyote in that last moment when he's windmilling wildly, but not falling yet.

Someone needs to launch a national campaign educating people on jury nullification.

The property damage laws are not really the sort of thing nullification is meant for, it's supposed to quickly put an end to the effect of bad laws, but if that's the one tool handy, and it's this much of a social issue, it should be put to work.

The duress deletion function was always obviously and transparently going to get treated as obstruction of justice issue the minute it got used in a high profile case. Here we are.

If you don't want to have trouble at the border, wipe your devices before you get anywhere near, and restore them once you are across. The precursors to this event have been in the news for years - the border is a legal gray zone, act accordingly.

I have mixed feeling here - sad for this kid, whomever he may be. Once again annoyed at the activist/NGO world for not training up to the point they can dodge this *easily* avoided problem. And I'm leaning forward in anticipation of being able to point and laugh at any crypto enthusiasts who make comically implausible arguments about these simple facts.

WIPE YOUR DEVICES BEFORE CROSSING THE BORDER.

Is that clear enough?

Starting with the Intel 386 it became possible to run a proper operating system on consumer hardware. I can still recall the excitement with that first Redhat 3.03 book and CD I received, sitting in front of a 16MHz i386sx, craving change. I had owned a Vax 11/730, an LMI Lambda, an SGI 210GTXB, but they were just stupid expensive to run and it took a couple friends to move them.

Anthropic and OpenAI are in the same position the workstation vendors were thirty years ago. What they do is amazing, extremely desirable to those of us who create and use software, but SOOO painful price wise. While I'm writing this on my 16GB M1 Mac the Bonsai 1 bit models are downloading. I mostly use Claude Max and an Ollama cloud subscription, but if I can shift some of the work to the machine on my desk ...

I let an RTX 5060Ti go cheap a few months ago, 16GB was just not enough. I don't regret this, but I wish my income position would change. The world would be a different place if the HP Z4 in my kitchen had an RTX 6000 Pro in place of its current GTX 1060. The advent of TurboQuant makes me want to try some things on one of the Intel GPUs - maybe that entry level sedan price on the RTX 6000 Pro is no longer the real barrier to entry.

These are strange days we are living, strange days indeed ... and no reactionary pack of ninnies is going to be able to do anything. It's a fine thing, having China defending our freedom to create, eh?

I changed careers thirty years ago to get Microsoft out of my life and I've never regretted that decision.

Using Microsoft operating systems is like dating a morbidly obese person who is continually in and out of the psych ward, and they've got a cluster B character disorder, one of the really manipulative ones.

I do not understand how that company's products are not limited to historic niches that are too difficult/unprofitable to clear.

Perhaps the advent of AI is going to lower the barrier to entry for decent operating systems. The Unix based ones have always seemed easier to me, but I'm an aspie weirdo so my opinion only counts just a little.

I was going to complain about the lack of a Claude Desktop for Linux, but I double checked before roasting them, and by gosh they've got a version in beta for Ubuntu and it'll work on any Debian offspring. When Anthropic says beta that means it's pretty darned good, they're cautious about putting stuff out.

I got a couple M1 Macs two years ago and the weight advantage pushed Linux off my desktop. Anthropic releases Mac first and I'm a Max subscriber so I've been at the head of the line. All my Linux these days is a Proxmox box in the living room and a Qubes 4.3 on my kitchen table that has been at install completion for two or three weeks, and I just don't have time to mess with it.

I'm looking at the stuff in my dock and all of these apps are on Linux. I guess my wish is ... maybe better integration on Mac file systems, so I can get a thumb drive from a pure Mac user and not have to do a bunch of gymnastics to get the data where I need it.

The Trump administration is cooking the books but they can't do it fast enough to head this off.

Even with the shell game the numbers are falling.

This is a post pandemic, starting to be AI era job market. Kinda looks like the pre-genocide Gaza strip, where one young person would support seven family members. This is my Signal chat today - two people overdrawn, one about to default on mortgage, a fourth who needs to move for safety's sake but can not afford.

This is a global phenomenon and the Hormuz "peace" where both sides keep shooting is NOT helping.

Years ago Republican pollster Frank Luntz, when asked how bad things might get, deadpanned "France. 1793."

We're not there yet, but you can faintly hear the thunk of falling guillotine blades, if you listen closely ...