pbmonster
2hr ago
Yes, all of those. Instead of burning hundreds/thousands of tokens on game state/memory each turn, you can run a simple state machine for comparatively free compute. This will also be more deterministically correct (you could even do correctness proofs if you really want to, or just have the LLM write a bunch of tests), because if you keep game state as tokens, you currently have no guarantees/tests.
pbmonster
6hr ago
it told me that it could fix these problems. And it did so in an interesting way: After each move in the game, it spelled out the game state in text. After that, it stopped making errors
And that's the worst way to solve this problem. What it should have done is write a tool call (to a state machine in python it itself wrote, mapping your rules to logic; or an open source chess engine; or a geodata software suite - whatever is required), and then just using the language model for interacting with you and the tool. This solves many problems created by the lack of a word model - much better than plain text ever could.
I suspect GPT would have solved the problem this way with a little nudge, or if called from a coding harness instead from the chat window, because many of those capabilities were added over the last few months. Previously difficult world model questions like "I'm driving from Bordeaux to Prague. The drive will take two days. Find a spot to spend the night, roughly in the middle, where I can go rock climbing. Don't add more than 60 minutes of additional driving." are now trivial, since the LLM just creates a hand full of Google Maps calls. Without tool use, arguing its way to correct solution was not possible - the travelling salesman cannot be argued with.
Unfortunately, this only shifts the problem a little. You now need good tools for manipulating the world. Many already exist, but others need to be created now. Some of those tools need a lot of non-LLM AI themselves (lots of robotics is like that, but getting a robot to fold laundry seems to be in reach), others probably need an entirely new approach (getting an LLM to call a CAD kernel to create 3D parts and assemblies seems like a monumental task, both in old-fashioned logic and vision capabilities).
How can society better support the men who sincerely look up to Clav as role model? Is there a way to become as viral as Clav by doing pro-social things (so offering a viable competing worldview)?
A counter-example to Clav would be a Youtube channel like Sickos. It's six childhood friends who try to make viral videos by doing Redbull-fueled stunts into water, snow and dirt, and a big part of their videos is just six bros ribbing each other, goading each other to bigger stunts, partying together and just hanging out.
They've been doing that for half a decade now, and their fans (mostly tween boys, from the crowed they pull at fan meetup events) got to watch them progress pretty considerably in their extreme sports skills, get rejected by girls they hit on, meet long-term girlfriends for the first time, and become financially successful on Youtube. While most are conventionally attractive men, several of them are self-admitted short kings who struggle with approaching women. While most of them are extremely athletic, their wide variety of sports almost guarantees that at least one of them sucks at the sport they're filming at any given point, which gets exploited for laughs. They show that all that is OK. And a recurring part of their videos is showing that they're afraid before big stunts, and that overcoming fear is worth it.
They speak Clav's language, but mostly make fun of it. They're very aware of status (they often approach girls in clubs by telling them they have "surfed Nazaré" - highest wave in the world in winter, they surfed it in summer, neither of which most of the girls will be aware of) and "chester-maxxing/peakcocking" (they often go outfit shopping together before partying and end up buying crazy fits), but they always show that not taking those things to seriously is important.
Certainly not without its faults (getting this good at surfing/skiing/motocross it's an expensive way of life), but a much better world view than red pill/PUA/looksmaxxing culture.
So you would need to mandate TPM chips in every device with more than four kilobytes of address space or something.
This is the direction the wind blows. I think it's almost inevitable. Phones, tablets and laptops are basically there already (TPM/secure boot/ect.), the one thing missing is that the boot loader on a few phones and most laptops isn't locked yet. But it will be, soon, just like the phones. The industry wants it that way, and politics wants it, too.
The next steps are easy. The only bootable OS on those devices comes with age checks and a locked app ecosystem. The only browsers available will cooperate. Then websites will be required to implement hand shakes dependent on keys in the TPM, and only serve data to valid devices.
And sure, you'll be able to get around it for a while, especially on niche hardware. But if industry and politics cooperate, getting onto Instagram will soon be as difficult as getting a 4K Netflix stream on a "custom operating system" (i.e. only by the grace of Usenet/torrents).
pbmonster
13hr ago
Russian serfs didn't have to be afraid their wives or children would be sold off to a far away plantation at any time, for one.
pbmonster
6d ago
Archive.is basically abuses the fact that most publications want their content to show up on search engines. It pretends to crawls the html once and then just displays the cached version to the user.
pbmonster
7d ago
My favorite autobiography is Barbarian Days by Will Finnegan. It's special, because most people with interesting lives aren't very good writers, and most good writers didn't live all that interesting lives. Finnegan is different. He's an exceptionally good writer (the book got him a Pulitzer), and he had an extremely interesting life - which he mostly spent surfing. It's pretty low drama, but full of interesting people, places and times.
He reads the audio book himself, an does so well.
It certainly helps if you like surfing, but I don't think it's required.
pbmonster
7d ago
Follow-up: Anyone have a reliable way for Substack? I'd gladly get a subscription along the lines of "unlock 10 pay-walled articles per month", but I'm not paying several hundreds of dollars per year for just a hand full of authors... they really need to rework their subscription model.
Also, by my impression going up the corporate ladder of some BigCorp has a much better floor AND a better ceiling unless you're exceptionally confident in yourself. But it probably depends highly on the field.
Absolutely depends on field. The vast majority of Germans (and basically all other Euros) cannot independently reach real wealth by being wagies. There are exceptions - if you have the education (enough credentials in the correct field) to really go up the corpo ladder or to become an expert at the local branch of a US company/startup, you don't need to risk anything - but the overwhelming majority of people simply do not have a route to break into a 6 figure salary while employed. The distribution of salaries is just to narrow, and there's almost no long tail in Europe (things look orders of magnitude more dire if you aim above 200k: less than 1% of adults in Germany earn more than 200k EUR, and that number includes all sources of income - so there's a lot of business owners, partners and capital income recipients in that block).
But if you own a business? There's no cultural monetary limits. No education/credential requirements. "Just" the soft skills, the drive and some luck (which you need to climb the cropo ladder as well) can get into the top few percentile points of income/wealth. And sure, you need to build momentum. You need to pick a field that allows for commerce (because selling things scales much better than selling labor) and that allows you an easy path to exploit the labor of the people you hire. But both come practically built-in with many blue collar careers: many of the people in those field do not want (or cannot) directly compete with you, they want to be hired; and most of your customers don't only need to buy labor, they also need physical products (which you can sell them with good - and often obscene - margins).
pbmonster
11d ago
I'm curious what those particular folks think of this today.
Strongly against, as always. Getting fingerprinted should require a warrant. Collecting this data without suspicion of a crime is a human rights violation, the data collected this way will eventually be misused, and almost all of it is ineffective security theater anyway - if you're a danger to society, this won't stop you.
I'm also definitely jaded with time. The stuff is unavoidable, every part of the West is on its way to become a surveillance society. In the intermediate future, car licence plates will be tracked by toll cameras, people will be tracked by facial recognition and gait analysis through surveillance cameras and the new eID systems they're building will be required daily to do anything at all online.
pbmonster
12d ago
Even in that fight Halmich got caught in a grapple.
Boxers do that all the time, right? No real penalty for doing so, so you don't have to avoid it. Getting on his inside and using her shorter levers and better technique might even have been advantageous. But yes, catastrophic for her in an MMA fight.
pbmonster
12d ago
Yes. Same goes for the fighting style itself. Black Widow goes in close, often for a quick take down.
The tactic that worked for Halmich was staying out of range with far better leg work, provoking strikes at his range limit to tire him out, dodging/blocking them easily and then going for low risk ripostes as he obviously ran out of gas after a few rounds. Doesn't make for good cinema.
pbmonster
12d ago
We don't, same way we don't know that he didn't have insider knowledge of what the next set of "random" tasks he excelled at would be. But he's know to be comically competitive and overambitious, it would be out of character for him to throw a fight or cheat. But yes, we can't know.
There is/was a German TV host that can serve as our "slightly above average man": Stefan Raab. In the early 2000s, he had several successful TV shows, one of them was pretty simple: a challenger needs to beat him in a tournament of random tasks. Quizzes, shooting pool, racing cars, rock climbing, race of cutting bread loafs, classic tug of war, ball games. The details are not terribly important, but Raab was comically difficult to beat. He destroyed a long list of (perfunctorily) competent looking opponents.
He's also relatively tall and heavy. And he had the bright idea to challenge Regina Halmich, female world champing in flyweight boxing for 12 consecutive years, to a boxing match. He reportedly trained for 4 weeks, and she destroyed him in 6 rounds, breaking his nose in the process. And it was not close. He had absolutely no chance, only a freak knockout could have saved him - but he can't really touch her at all, making that difficult.
So that's one data point. I'd put Raab at around +1 SD of the adult male, both in general fitness and in boxing. That is not enough for the female +4 SD, at least not in boxing. I'm sure wrestling and MMA would have looked different... but maybe not, not sure if 4 weeks of training is enough not to end up falling on top of the female world champion - just to get choked/arm barred immediately anyway. Also, Halmich (above) is a flyweight. A heavier/stronger woman, one who additionally is trained (and allowed) to kick, might shift the difficulty for the average man again.
Still, looks nothing like Black Widow, obviously.
pbmonster
12d ago
The best female athletes in the world across a variety of sports roughly equal the athletic performance of ten to twelve year old boys.
No need to go overboard here. Let's go with 15 year old boys on elite teams. Not doable without at least some testosterone and some training.
pbmonster
18d ago
I mean it's possible that I've been reading propaganda. But what's your explanation why the carriers were inside the Persian Gulf during Operation Iraqi Freedom, but this time they're so far offshore they need to run aerial refueling operations for most strike aircraft to make it back?
It's gotta be anti-ship missiles, right?
pbmonster
18d ago
I'm not sure what exactly the Iranians have been building exactly, but they probably don't run refineries with processes as sophisticated as the Golf refineries. The thing that makes suddenly processing large amounts of heavy crude difficult is that you need to build additional cokers, and in the west those are multi billion dollar projects. How much of that is due to economical, ecological or bureaucratic reasons I can't tell. Probably all of the above, and an Iranian refinery will certainly have more slack in all three dimensions.
pbmonster
18d ago
maybe the coastline near Bandar Abbas. If this succeeds the US will (after doing minesweeping and patrolling the coast for hidden marine drones and such) declare the strait open, and the next move will belong to the P&I cartel.
This is a pipe dream. Iran has bunkerized anti-ship missiles all along the coast, not only around Bandar Abbas. Same is true for "covert bases" for mine layers (small warehouses in fishing harbors) - you can place seafloor mines from fishing boats/speed boats, after all, at night or in heavy traffic.
Also, with the state of US naval assets and the... current disposition of allied navies, minesweeping of the straight would take forever. If the IRGC gets significant numbers of mines out (and they do seem to have the capability), clearing them would take many months.
And then, once all this is done, the first tanker making a run for it might learn that Iran now actually has integrated rudimentary radar guidance into their shaheds or their 1000+km range ballistic missiles. Locating a tanker on open water using radar is not exactly difficult - its rudimentary WWII tech, radio hobbyist without further technological background can do it.
At which point the cartel will concede that, yes, the straight is actually still closed. Even if that drone/missile was shot down, because everybody knows that sooner or later one will slip through.
pbmonster
18d ago
America lacks the refining capacity to handle the domestically produced light crude
It only lacks the refinery capacity to handle it at maximum economic optimization, right? All those gulf refineries could process light crude, maybe with minor retooling. But they're not optimized for it, and a lot of expensive equipment specifically built for heavy crude would sit unused. Going the other way (refining heavy crude in a light crude refinery) is significantly more complex and actually has many years of lead time for the equipment that's required additionally.
So, the interesting question is really: at what cost? You could ban all oil exports, and refine the light crude for domestic consumption in the gulf refineries. But the price at the pump would be difficult to guess.
pbmonster
19d ago
The AI Labs could all be happy fat tech companies if they just became inference providers.
Yes, but we aren't quite there yet. Not even close, in my opinion, at least when we're talking about serious job displacement. Unless there's a phase change, were looking at years of more insane capex and opex.
And those trillions will need to be payed back with interest. We're not talking about a Netflix or Office 365 licence that every office drone just has. For millions of workers, access to those tools will rival transportation and housing in ongoing cost.
No problem, if your employer already has half the staff on SolidEdge/Ansys/ect licenses and generally does not care what toolboxes everybody gets. For the rest? Small business, low productivity labor, labour limited by hardware throughout (classic example: radiology)? They won't really contribute to paying off that debt, so they won't get a lot of tokens, and none from the good models.
We've gotten used to tech, especially software, being cheap. For the current economics to make any sense, this will come to a hard stop. On the cost side, AI is much more like an excavator than like a shovel, and it really needs to replace just as many workers to make sense building them.
And I can well imagine this never happening. Maybe they'll never get reliable enough for that much unsupervised work, especially work you can't write unit tests for.
pbmonster
19d ago
It's materially improved my productivity at work
Then you have people still in 2026 who genuinely think AI will "go away"
Note that both of those can be true at the same time. Total cumulative AI capex will probably cross $2 trillion this year, and cumulative opex is on the same order again. And that's just in the US.
If this "technological revolution" doesn't end up replacing a significant percentage of national labor costs, it actually might fade away - and the only thing remaining will be whatever open weight models can be run on cheap hardware at that point in time. And if the Chinese keep releasing last year's SOTA for free, none of the envisioned business models might hold water.
Either way, there's really only one way costs can go from here. If business AI doesn't go away, those of us still with jobs will get to work with agents costing our employers on the same order as the people they replaced.
pbmonster
26d ago
It’s more likely that they couldn’t compete with open source models, Chinese offerings, and other companies specialised in video generation.
Which open model is closest in quality?
pbmonster
27d ago
But then you drive in with your comrades and do some shooting and see how it goes from there.
Do you imagine you could have gotten the human capital present on January 6th to carry guns, let alone use them, in the Capitol? It's one thing to convince ten thousand people to go protests at the Capitol. Doesn't take much to get 10% of them riled up and caught in peer pressure to storm peacefully walk in the door.
But carrying rifles, towards the sound of (even friendly) gunfire? No way you can do that with random normies. You need to indoctrinate them, train them, and establish a chain of command. This is, of course, the point where you pull every single fed in a 100 mile radius, and now your chances of even getting onto the Capitol grounds go way down, if they don't flat out raid you while you wait for your timing.
If opposition is limpwristed, divided and feckless enough maybe you can get away with it.
Do you really think it would be? I can't imagine a crisis so severe that it would stop attack helicopter to be there within the hour. At which point you either reveal that the vanguard already has infiltrated the armed forces (realistically, all branches and the DC National Guard, because they all have bases close by), or... I guess you can play hard ball and bring out your hostages and guys with MANPADS while assets from the single USM base you've infiltrated start bombing all the other bases in the area?
There's just no way, right? You'd need an overwhelming majority of the (local) armed forces on your side before you even start.
pbmonster
1mo ago
Do they? Shahed stockpiles alone are reported to be above 6k drones, and production capacity is reported to be at least 3k drones per year, according to conservative OSINT analysts, and both might be as much as 3x higher. Israeli think tanks report 80k and 12k, respectively (lol). Have either of those been degraded significantly?
And that doesn't even start with missile stock piles.
Alright, I stand corrected. Still, sounds like it was a lot less frequent for agricultural labor (which probably always was the vast majority of slaves) than in other systems of cattle slavery.
More options
Context Copy link