cmrdporcupine

Some people get this way the moment they get a whiff you've touched an LLM. Like, beyond rude.

Frankly, it's usually just gatekeeping (or just being threatened) hiding behind some moral system they just assume you should share with them.

Junie in particular, or just their various other "AI Assistant" / "Chat" things? They have multiple confusing overlapping products/modes going on. "JetBrains AI" -- which may or may not be same as "Junie"?, "AI Chat", then there's the "Agents" tab which integrates outside coding agents.

Their coding agent "Junie" is actually really good. The problem is they force you through only a few curated models. And try to funnel you through them to subscribe to do it, too. It's not like Pi or opencode that lets you just point at an OpenAI API endpoint and then choose whatever model. If I recall you can even actually use local models, or openrouter etc, but only use the specific model variants from their curated list. It's insane. They tried to stick themselves in the middle, and failed. Nobody will use it.

They've also done so many pivots on AI, all of them weakly and annoyingly done. They couldn't even pull of FiTM copilot well, they were super late on it and it had all sort of bugs.

They could have just taken the loss and written an amazing GUI coding agent that integrated well with their IDE -- amazing integrations with their refactorings, etc. -- but that played well in other people's ecosystems but they refused and tried to own the whole pipe. Or something. They just confused me.

Their latest attempt is to allow you to connect to and run other people's agent harnesses inside the IDE and it works well enough but it wasn't enough to grab me since I (like you) use herdr on another machine to do the coding.

I've been using their products since ~IDEA 3.x. Almost my entire 25+ year career really. My fingers are wedded to their keystrokes.

Relational algebraic model (which SQL is a debased form of) is literally based on propositional logic. Or at least that's the claim. Propositional & predicate logic is evidently part of human language, grammar, and present in human thinking systems. Though it would be -- as you are intimating -- insane hubris to make one singular claim about how the brain represents knowledge.

Commenter is self-promoting their "graph DB" related product.

Giant citation needed. This is completely unsubstantiated. And contrary also to whole schools of epistemology. Also nothing to do with TFA.

"Yeah, well, you know, that's just, like, your opinion, man"

I think in the real world LeanStore doesn't do the the MADV_DONTNEED bit at all though because of performance concerns? Overhead is high (syscall cost, and frequent thrashing about in the kernel VM subsystem). If I recall the vmcache paper gets into this (Virtual-Memory Assisted Buffer Management, SIGMOD 2023). Lots of talk about TLB shootdowns and throughput concerns.

In my own experiments I confirmed the same thing. Just using explicit free lists / allocator techniques was far more performant.

And actually-existing LeanStore on github doesn't use it, anyways. And strangely never has (looking through git history). Even though the paper (2018) describes it. Maybe Leis or Alhamsi can explain.

Curious if Umbra and the CedarDB stuff that came from it do use this approach, too. I seem to recall the Umbra talk speaks about using it.

As a lazy "eventually reclaim when I've got free cycles" technique maybe ok? But revealing that the actual LeanStore C++ code never does it.

This is already happening, and it's scary.

But this is about humans malevolently using AI as a tool, not about AI autonomously doing "its own thing" (ahem David S Pumpkins) against "humanity."

It is instead already existing anti-social elements within human civilization harnessing machinery to accomplish their own selfish and destructive aims, regardless of the broader cost. Which is nothing new -- it's the story of all the horrors of the 20th century, as well.

Ask yourself who benefits from ideological attempts to hide the human driving the process from the agent doing its work for them.

If we imbue these machines with apparent autonomy, this is nothing more than an attempt to hide the man behind the curtain.

If we're conjuring SF worlds ... I still prefer the outcome of (RIP) Iain M Banks' Culture universe: hyper smart artificial intelligent consciousnesses admire and love us for what we are, and respect the experiences we have, and see part of their job -- among the myriads they take on -- to care-take, assist, and work with us.

That's a perhaps overly optimistic view, but feels like one worth aiming for.

In reality we're so far from consciousness and independent goal-directedness in AI -- let alone any kind of energy and real computational efficiency... I agree with LeCun in broad strokes. I think he's overly dismissive of LLMs perhaps but I do think he's right that talk of "wiping out humanity" is theatre for people playing corporate power games.

Well, I worked there for 10 years and I can tell you that they're incapable of making a singular focused decision and executing on almost anything.

There's likely all kinds of internal competition and disagreements and dysfunctions.

So there's that.

But in the last instance, revenues win. OpenAI and Anthropic have to fight for new revenue sources while Google already has a firehose. The motivations are fundamentally different.

I have said this before here, but. I don't think they care.

Google gains very little by having us use their models for coding. It doesn't help them in their core mission ("to organize the world's information and make it universally accessible and useful ^W^W^W^W^W^W[and turn it into ad revenue]"), nor would it be anything close to substantial revenue compared to the ads revenue firehose they already have.

They will push AI in a direction that serves their existing goals and service the coding agent model only as is necessary.

They will go through the usual transition of "can't release a model" to "won't load in a harness normal people can use for 3-4 weeks" to "it's smart as hell but completely inept at tool use and coding" to "now it's behind everyone else" ... like every Gemini release.

"For antitrust reasons, it’s helpful for the US government to mediate or at least enable these discussions — they don’t need to participate, but do need to issue a narrow waiver for certain kinds of safety conversations. " - Dario a couple weeks ago.

Yes, he was talking about safety, but IMHO they're likely already IMHO pushing the boundaries of cartel type behaviour. And they will use safety as the cover to make it happen.

I suspect we'll see serious price fixing and the DOJ do nothing about it because of the inroads these people have with the Trump regime.

Whether that survives contact with Chinese open weight models is hard to say.

Sorry, I'm GenX. Growing up they showed us "Old Yeller" in the school gym every year like that was some kind of treat.

Taking GPT-6 "Sol" outside behind the shed and giving it a merciful end is about the best outcome possible.

Huge misstep releasing it.

6 Sol was worse than 5.6 Sol from my own experiences. Far worse.

Will see if this remedies things.

Aka "we made an oopsie last week and released what should have been called GPT 6 Terra with the name GPT 6 Sol"

Imagine if you will two product proposals coming across Dario's desk.

One is:

1) Complex IDE integration where our AI model is called to assist users in making decisions, assisting them to do the things they already intended to do. It appears more as a function of the IDE rather than something separate. It shows off the intelligence of the model but doesn't show off any autonomy.

2) A separate tool that can be branded Anthropic. It takes over completely. It advances the narrative that we've already been spreading that AI is going to make some human jobs obsolete. It looks to employers (the actual people who spend money) like it replaces an expensive developer even if at first it requires one. It has minimal to no integration point.

Which do you think they'll approve. I think it has less to do with the user motivation and more with the producer's.

I am simultaneously deep down in the agentic treadmill and also supremely frustrated by this.

And I don't think it was ever necessary to go to the point where people just gave up authorship. These were choices made by adopting the "I'll do everything for you" agentic "harness" model that shipped with Claude Code but it was never inevitable.

e.g. we completely dropped fill-in-the-middle completion OG CoPilot auto completion model. That combined AI authorship with a human always in the mix and I actually really enjoyed it. It's just that the models involved were pretty stupid. We totally could have had IDE / shell / tooling integration that kept people in the driver's seat while automating parts of the drudgery away. Instead what we got was a simple chat loop with "oh, whatever, you go do it" being the ultimate result. Cuz, you'll totally review everything after and understand it, right?

The things should end by quizzing you on what was just made and if you don't pass, just throw it away. That'd be funny to watch.

68020 and onwards could also get substantially more work done in a cycle than a 68000. So Amiga 1200 or Atari Falcon not only had substantially higher bus speeds they were just ... better... at the CPU level in terms of how much work could be done.

Even if they "solved" that, the problem is it's the LLM that "knows" it, not the team.

Which is really the same problem with coding.

The agentic model of it just taking over and doing everything is poisonous to effective long term team work.

We're well past the point where it's about the quality of the work they produce. It's the way they integrate (or rather, don't) into human practices.

> I don't know if the dominance of the PC was a foregone conclusion by 1985

It absolutely was. Simply on account of the name on the grey box.

If it hadn't been IBM, maybe somebody like DEC may have gotten their shit together to take the market instead of farting around with overly high margin products and eating themselves from the inside.

But the way "business" culture was in the 80s the "respectable" productivity machine could not have come from the games/"home" market. Apple maybe but they opted instead to go close off the Apple II line (which in fact had a strong productivity sector thing going on, especially in schools) and push the Macintosh which was a good product but started off weak in terms of capability and costly on price.

In the early to mid-80s it was either going to be a CP/M machine or a CP/M-like machine (like the PC). These are Visicalc/Lotus-123/WordPerfect machines to their customers, and they cared not at all about graphics capabilities until later. Just needed 80 columns, floppies, and reasonable amounts of memory.

Hell there was terror in the microcomputer market when IBM made moves to go "downmarket" into it with the PCjr (which was a flop). The hype leading up to it, and the noises about how it would be a Commodore and Atari killer was insane.

I mean, that's just pedantic. Of course the display frame rate was fixed. The point is what the effective game loop responsiveness was. which was often a fraction of the actual display's capability. If you go back and play some of the classic games of that era, there's often low responsiveness to joystick movements etc.

Graphic fidelity on the 16-bit machines was great. Latency was sometimes bad, depending on how much they were trying to do in the game loop.

Despite its obviously far lower MIPS rating the 6502 could just get more "video game" type work done in a clock cycle than a 68000. 68000 interrupt responsiveness was also pretty terrible.

Looking back 68000 had little business in video game type machines. I like the 68000 from a programming POV but it's faster for doing a pile of fast integer calculations than it is at moving memory around and responding to interrupts.

But Atari and Commodore were also imagining their machines to be targeting productivity more than games, so.

ChatGPT also seems down for me.

Frame rates on games on the ST were terrible, which just came down to there not being enough cycles in a single frame to really move stuff around. Sprites or tiles would be great but are a bit specific. Blitter hardware subsumes all of that under a more general utility. The STe's blitter when it finally came along was great. But by then it was too late, nobody was going to make games for it because the market was too small.

Once clockspeeds got up to what the Falcon could do, it was unnecessary. But again, too late.

This is a great post, thanks. I've been following prototype systems since I first logged into lambdamoo in 91, and Self is an old intellectual muse of mine but this is the first time I'd seen Korz.

You might be interested in my lang/environment "mica" which attacks this same dimension of problems from a relational/datalog angle and then makes object/prototype identity hopefully emergent out of that: https://github.com/timbran-project/mica -- relational dispatch feels like a broadening of multiple dispatch.

I feel like I may have ended up at a similar place as Korz without realizing it. I will need to dig into this paper.

Speaking also of multiple dispatch and prototypes, there's also this paper from an old acquaintance: https://www.cs.cmu.edu/~aldrich/courses/819/salzman-pmd.pdf which I used to find interesting (and built my first version of "mica" back then around the idea).

yeah, substantially agree. But I also think if only Atari and Commodore teams could have just been one company... it would have been glorious. Somehow. The market couldn't afford to be split like that, the pie wasn't big enough.

Also I was an Atari ST user back then, and loved it. But the Tramiel's didn't "get" software in the early days. They failed to make the right investments to keep TOS/GEM actively developed (e.g. as a Mac alternative) until it was too late (they hired Eric Smith to work on MiNT and then MultiTOS in the early 90s but they were already done as a company by then).

And then three fundamental hardware mistakes with the ST were:

1. not shipping it with the Blitter from the start so it couldn't compete on games, obviously with the Amiga but even with 8-bits of the time

2. forcing the ROM size down to 192kB so they had to size optimize the crap out of GEM before shipping (waste of precious development time), especially including tossing away all of GDOS (which had support for proportional fonts and printer drivers etc). A 256kB or 512kB ROM would have solved a lot of problems with what I'd think wouldn't have been an insane BOM cost bump.

3. Minor but long term annoying -- tying the cartridge port to only have 128kB of address lines. If they had given it 256kB or 512kB address ability their later expansion issues (OS upgrades etc) would have been much improved.

As for the Amiga, great machine but a) it cost too much at the start so failed to catch immediate fire (the 500 rectified this to some degree) b) interlace.

The alternative to "open weights" at this point is "American controlled."

And Dario and Sam have already made it clear that it's America First.

The rest of the world isn't going to accept a regulatory regime which imposes American hegemony. Maybe when Silicon Valley was playing all utopian like they used to. Not now.

Open weights is the most reasonable counter-power we have.

guess i'm old cuz i still feel like it's weird and antisocial

I mean, those of us with Marxian biases define it that way -- not right wingers -- and see "social democrats" in the 21st century as being the people who successively turned the term to mean instead anything-where-government-does-stuff. That's effectively what it came to mean in the US.

Notably, the definition given by Marx and other 19th century socialists (worker controlled production) is doesn't even have to involve government or the state at all. It just came to mean that once proximity to power came into play. The "soviet" word merely meant local worker councils, not a state form, for example. But once the Bolsheviks took state power, the power of the actual soviets was successively demolished in favour of a centralized one party state.

So, notably: Self-called "Marxist-Leninists" in the USSR, and Social Democrats (or "democratic socialists" in the US), and Right-wingers all far prefer the term to mean "government controls things." The latter because they can then point fingers and call things authoritarian (but somehow exempt corporate power from critique).

But it is not at all clear that 19th century socialists .. from Proudhon through Marx etc -- who coined the term -- meant this at all.

Also this hard line between "communism" and "socialism" is fictitious and very schematic. All communists call themselves socialists, and almost none of them agree on what that means.

Social democrats trying desperately to sound respectable clearly want a nice hard line there. But they also have no intention of bringing about socialism.