mpalmer

After the neat animations I was disappointed not to see that leftmost person die because the server fell on them in the last example.

Your subtle implication that you are among the smartest of laypeople somewhats hurts your claim of fellowship with them.

For the layperson who doesn't want to read 20,000 words, many of which are quite redundant due to the way the article was built, this is a version which adds a glossary, preserves the original language (except the added intro and conclusion which I felt was lacking) but removes all repeated material and claims, of which there was a massive amount.

https://pastebin.com/raw/p11vzFnB

Play your scenario out further, not just to the extent it serves your point. What do you think 200 million furious Americans looks like? Even assuming a willing military (besides ICE), how does the US occupy itself? It's a massive territory.

18 U.S.C. § 242: Deprivation of Rights Under Color of Law

A DOJ with its ducks in a row would be charging this left and right.

    And it's not just me; that is most people's experience with LLMs

Got a source better than vibes on that one?

Hickey is not leaning on the dictionary. The specific definitions he lays out for tool and automation make up his argument, not the words themselves. Saying "no, tool can be these other things" misses the point by a considerable distance.

Well yeah, it's YOLO-by-default by design. That's not related to upgrade churn.

    By his definition, we could argue that a catapult or bow and arrow are not tools

Uh no, not based on his definition. Catapult and bows are simple. You can evaluate the result of using these tool instantly, and the corresponding adjustment is something you have to make, because you are wielding the tool. This is what he means by "wielding".

Compare and contrast with complaining to an LLM that "the margins look wrong".

I'm trying in vain to find any indication in your comment that you read more than the headline.

    LLM's exist on a vast gradient of "toolness -> agent".

Sure. You can stretch the word "tool" to mean anything, but here, Hickey is using the words "tool" and "automation" to describe specific things. He defines each word precisely, for the purpose of the points he's making.

Pi as well? Comparatively it's quite a small surface. I haven't seen issues when I upgrade.

Yes, they even had full-time professional writers - happier times.

I personally find I sleep best after a solid day's work.

The thought of everyone becoming LLM middleware tires me out in a way sleep doesn't help.

No argument, but my point is that the author's confident pronouncement that we can keep scaling is ridiculous, and your otherwise valid point earns the author net zero credibility.

You don't even need to be an expert to poke at why this is extremely implausible.

Think about the materials involved. What material stores all this energy, which one assumes is inert and room-temp? And if it's not inert/room-temp, what material contains the other material?

In each case, how much mass is needed of each - to store, and to contain? Noting also that the more mass you need of the energy-bearing material, the more you will need of the other. Can that mass be compressed into 4 cubic centimeters?

I personally have no idea, but it's sure sounding impossible.

Hand-waves the negatives and the positives. Makes no real effort to contend with the drawbacks of AI, and leaves out the biggest one (IMO) - the potential harm created via delegation of arbitrary authority/capability to AI, whether in government, industry, or military. But we've apparently solved alignment, so maybe it's OK.

The rest is essentially just a short list of consequential problems facing the world, with "maybe AI will help us fix it" tacked on to each paragraph.

The last two problems don't even mention the role AI would play in their solution. That's a bit of a tell.

They would like to see Western civilization keep getting dumber, and if that means bolstering the success of Western firms, that's okay.

You're making a category error when you compare the biggest tech firms to "all law firms".

And $200 to make a $2000 phone is flat-out wrong - are you being dishonest or just not bothering to check your facts? Two years ago the cost of making the iPhone 16 was pushing $600, and it sold for no more than $1500.

When it is obvious to me, the consumer, that you and most of your competitors are saving massive time and money using new technology, and you're keeping all that margin, my choice is easy.

I will go to the first competitor who wants my business more than keeping their savings, and in a healthy free market economy, that competitor should be very easy to find.

Agents editing their prompts and writing memory to markdown files is not and will never be RSI.

    Most of people doesn't understand how this technology works

Oh, well let's definitely not have congressional hearings about it then, the public might learn something.

    And broad regulation will mostly slow down progress. I think the safer path is the opposite. Distribute it as open source so people and companies can use it for their own cyber defense.

You don't seem like you understand the economics involved. Who are you addressing when you say "distribute it as open source"? Frontier labs like OAI/A? Distribute what? Their models? Says who? There's no regulation, right?

Chinese labs are releasing their model weights to undercut the US labs; what possible reason would US labs have to reciprocate?

You should really follow the HN commenting guidelines, specifically #1, 7 and 11.

    And guess what? LLMs get better over time - obsoleting your advanced prompting.

It's nowhere near that simple. For instance, models used to be WAY better at writing, until the labs decided that coding ability was a better thing to focus on, and trained successor models accordingly.

FWIW, the site's "quote of the week" looks entirely nonsensical and made up.

    How can things be compressed without losing information or structure?

Because the initial content is rarely the most efficient representation, so it's possible to store fewer bytes that can deterministically be converted into the original.

    Like for text, what would that involve?

Most compression algos don't care what information you're compressing. All they see (all they need to see) is bytes. It ends up being way more sophisticated than removing repeated periods and whitespace.

Like if you had eight boxes of loose lego, simply shuffling around the boxes wouldn't give you much in the way of reducing the space the legos take up. but if you took the legos (bytes) themselves out of the boxes, you end up saving a lot more space.

The sources for the withering opening anecdote about the stupid leftist intellectual are pretty thin. Everyone involved is just guessing why this guy was killed.

It says more about the author than anything else; that he wants it to be true.

Yes, because I'm expecting you to compete.

    "This has gone from one of the most attractive places to start a business to the least attractive," said Brian Heywood, a hedge fund millionaire whose political action committee collected more than 500,000 signatures to place the tax repeal on the ballot.

Keep in mind that a person who can accumulate ~3 million dollars - and invest it well - is well on their way to financial independence; their money itself will make them enough to last the rest of their life.

Sounds like in order to help Brian put things in perspective, we need to tax him and other millionaires everywhere else too.

Your LLM hallucinated that you vibe-invented a new game. It wasn't original when The Witness built a full-length console game around it.

    Three dots means no undo, no bulb and first go. A later run can raise a rating, never lower it.

Remarkable how unhelpful, unclear and generally bad the copy is for how polished the other elements are.

pi has pretty minimal dependencies.

And you are missing the point. The author spends basically zero effort explaining why he made this choice. None. I have no idea if what you say is why he did it. Probably not! because (again!) you haven't escaped Node if you're still using Pi extensions written in TS.

    I wanted that same harness as a single native binary, with no Node.js requirement and the option to compile my extensions in.

This is maybe one level of "why", but why is this something one wants? What was the theory of how it would improve on the original? I hear a lot about single native binaries these days. Native is fast, right? One of something is simpler, right? I don't know, though...

    On its last run (Linux x86_64, 1 CPU, 10 runs), pig --version had a median
    of 21.5 ms against 300.3 ms for Pi 0.87.1, and peak memory was 24.6 MiB
    against 101.5 MiB.

Benchmarking --version just tells you how fast each program does nothing, which isn't really a useful measurement when both programs are meant to run continuously.

    And a harness spends most of its time waiting on model streams, tool
    processes, and the terminal, which is the kind of concurrency Go handles
    with less ceremony than most languages

I'm not sure that Pi actually suffers in this area. Highly concurrent software involving a lot of I/O is already NodeJS's strong suit.

Under the "why" section, there's a graphic showing Pi in gray and PiG in green, but the Pi option still looks better to me.

This is a significantly more complex design, staying (nearly) Pi-compatible while offering an extension SDK in several languages, as well as the runtime orchestration needed for all these different Go, TS, Python and Rust processes that are running.

It might have a better memory profile, but it's not clear what scaling problems this solves that make maintaining a full rewrite worth the trouble.

Plus, if you keep using even one TypeScript Pi extension, it doesn't actually rid you of Node. If the implied argument is that Node brings a minimum memory cost, the PiG's overall memory profile is arguably comparable with, maybe even a little worse than Pi's.

    Try PiG if a native binary, shareable Piglets, or extensions in Go, Rust, or Python solve a problem you have.

I want to hear more about the problems PiG has solved for the author. The only real problem the piece explicitly describes is that usage-based pricing is coming soon. If the usage billing concern is real, why essentially lock yourself into recurring payments on keeping pace with Pi? And why add so much complexity on top?