The “not A but Z” thing and variations thereof was pretty common before LLMs. The noticeable thing about their usage of it is they’re trying to use language meant to take the reader from something they might genuinely be confused about to a surprising conclusion, and using it in a way that’s entirely banal. It relies on distance between A and Z, and genuine possibly of either. Humans tend to have way better intuition about what is surprising to other humans, and don’t make insane mistakes like LLMs do.

Rather than have “Z” be self-evidently interesting, the LLM need to tell us that it’s not “A”. Except no one thought anything was “A” in the first place, and the “Z” is barely a “B” let alone a “Z”.

This also goes for couplings of three short descriptors (“Simple. Intuitive. Seamless.”) and summations (“the important part to realise is:”), bullet point lists, etc. All techniques to say: here’s the important part, here’s the bit you should listen to.

This all being said: the tweet smells like ai to me. Wtf does he mean by sword

See my other comment, but yes, this is a surprising result. It goes against many intuitions, but it’s an extremely famous trend in the field.

This is literally one of the most famous essays in AI (it has it’s own Wikipedia article) and I mentioned it by name but sure here you go: https://www.cs.utexas.edu/~eunsol/courses/data/bitter_lesson.pdf

As for more recent stuff, people are doing experiments on it all the time, here’s Nvidia very recently trying to figure out the best mix of training data (how much task-specific training, how much general-knowledge training for optimal results) https://arxiv.org/html/2606.24747

Google’s 2022 “A generalist agent” - 1091 citations https://arxiv.org/abs/2205.06175

The entire field is built on statistics. It has many flaws, but this is like, the entire thing people are working on actively. Their goals may not be compatible with the flourishing of humanity, but finding the best way to automate various tasks is their one goal.

Even from gpt-1 people have been trying to make fine-tuned models for specific tasks and they keep failing compared to general models.

Elements of this idea do live on though, in MoE architectures, where they take a base model with knowledge of everything, then fine tune various versions of it for different things, and route your request to one of the models fine tuned for your task. This is mainly a workaround for the fact a large model with all parameters doesn’t fit in memory so easily even in the massive Nvidia datacenter gpus, if it did, we can be pretty sure it would beat the smaller “experts” in most of the tasks

Also like, china isn’t doing different to this? Deepseek (China) and glm5.2 (China) and mistral (France) and various other models are doing the same thing, because that’s the thing that works (for the narrow definition of ai success that tech companies and politicians believe in)

Apparently only the driver is in there, at least for the jump. The ramp was built by a corps of engineers (see the video). It’s a test for ways to cross rivers - all the crew would get out and I guess cross on a rowboat or dinghy, and the driver would jump the tank across and then they’d join up on the other side

I haven’t had it for years, pretty sure it was nuked by some part of some script I run whenever I start up a new win 11 installation - power tools and the like

https://youtu.be/EC64uo6tmAI

They did tests jumping rivers and lakes as a way of crossing them.

Memoir of the driver: https://t34inform.ru/publication/p-pers-4.html

“A jump… Perhaps this is the most difficult of all tank aerobatics exercises. Only the most skilled and fearless driver can perform it. Just imagine: hurling a metal colossus weighing over ten tons down a steep bank into the water, and at a speed of fifty kilometers per hour, at that. But the point is not to hurl it, but to jump—that is, to swiftly fly tens of meters through the air, and then smoothly glide across the water’s surface, keeping both yourself and the tank intact.” To do this, a rather complex program must be precisely and quickly executed in a matter of seconds. Begin accelerating at maximum speed, shifting gears quickly and at maximum engine speed in each gear. Cover the last 20-30 meters at top speed, raising the tank’s nose slightly just before the jump. As the tank lifted off the bank, the engine’s load was relieved, the RPMs increased, and the tracks began spinning at breakneck speed. To avoid losing them in the air, the RPMs had to be reduced to minimum. An impact is inevitable—a tank is not a bird, after all!—but this will make it glide. A tanker executing this maneuver has a special approach. Drink almost nothing for 24 hours beforehand. Otherwise, the kidneys won’t withstand the hydraulic shock, and blood vessels will burst. At the moment of the jump, drop the levers, bend the body sharply, and be prepared to protect your head with your arms in the event of a strong impact. Then they had to regain control of the controls and maneuver the vehicle onto the shore. Because the banks of Blue Lake were gently sloping, the sappers had to construct an 8-meter embankment with a ramp, simulating a high bank from which the BT-7 tank would jump to overcome the water obstacle. To soften the impact, a 30-meter “mat” of pine needles, approximately 1.5 meters thick, was laid in the area where the tank would likely enter the water. According to calculations, the tank’s acceleration path had to be at least 250 meters, which would ensure that the tank reached a speed of 72 km/h upon takeoff, with the calculated jump length being 37 meters. After the BT-7 tank, driven by E.A. Kulchitsky successfully crossed the Blue Lake. It was determined that the actual jump length was 5 meters longer than predicted, reaching 42 meters—a record unbroken to this day by any tank crew."

Have you ever heard of the bitter lesson? Your suggestion has repeatedly been shown not to work. People want it to work so bad but the data doesn’t bear it out: generalist systems beat specialists time and time again.

This doesn’t make ai good for society, but specialist systems isn’t the right path if what you want is things to be able to do hard tasks

Windows servers

I think the only reasonable answer is to just leave without it. This one’s on the gift giver; I can’t possibly accept it. For one, it definitely can’t go in my apartment, and it would cost a shit ton to keep it healthy. It would be an extraordinarily bad idea to take it under my care. I would be getting all sorts of new and exotic fines for all the damage it will do to neighborhood cars etc.

Lol what? Their system makes more sense

To make a circular mask in gimp, use the circle select tool, then fill your selection.

This, I’ve never understood the learned helplessness of some people. The 12,000 employee international company I work for doesn’t pay for any Adobe products. You can just use literally any other product or open source alternative, I promise you there are multiple good ones with extremely well supported professional workflows used by millions of people every day, for every conceivable Adobe product. You can learn a new workflow I believe in you, it will take like three months to get used to.

In the case that you have an earthquake or something big enough to cause the cave to collapse yeah you’re just dead. This is also true of rock climbing, or walking over a bridge

If you’re not suicidally stupid you’ll already be holding onto the rope leading to safety with plastic arrows attached every few feet pointing to the exit so you know which way along the rope is out. Never let go of the rope! (You’re usually tied to it also, but ofc you gotta unclip to switch rope or go around an anchor where the rope is attached to the rock)

Idk, pretty much all the deaths I hear about are open water divers who decided to go into a cave, ignoring the part of their training where someone shouts at them for an hour to never even think about going in a cave even a teeny little bit, no not even just the entrance, no not even just to look.

This includes people going in to recover bodies. You’ll notice the people dying doing that are often military divers who’ve never been trained for cave diving (true of the recent Maldives incident and also the diver in the Thai cave rescue). I don’t say this to be like oh training is some magical panacea but rather that you specifically really should not go in a cave even if you really want to without getting cave trained first. It’s not even remotely the same thing as open water diving.

Oh and yeah it’s definitely cave diving that’s the insane one. Never done it myself and never will but it’s an interest of mine because of all the procedure, backups of backups, redundancy etc needed to make it safe. It’s more like spacewalking in that regard.

I think a big part of the problem is the time investment and the fact most people really vibe with only a few types of games.

On time investment: in order to say anything interesting about the game, anything you couldn’t get from watching someone play it for two minutes, you’d need someone to take the time to finish it, often 50+ hours, and probably more to digest it and to play stuff around the sides (achievements, collecting. Given an 8 hour work day, and reasonable breaks, that’s easily a week or more of just playing before you start writing. Then you’ve got to write with better insight into mechanics, design etc than the average guy, which a lot of critics fail at. Not just what mechanically does the game do, but clearheadedly why is that different, what makes that good and bad, how do they reinforce or work against the games themes.

On subjectivity: I could not review a first person shooter. I could not review an RPG like Skyrim. It’s not because I couldn’t play those games, or even derive some enjoyment. But rather that I know my mind, and I know that I have a strategic optimizing mind that left alone would rather play spreadsheet simulator stuff like rule the waves or dwarf fortress or football manager. Complexity is my god. But that’s not the route to good design for most people. Similarly, guy who only plays CSGO probably shouldn’t be reviewing disco Elysium.

All of this put together means an ideal reviewer would be working on a narrow selection of games, taking weeks or more for each one. And they’d be competing with people making excellent video essays for free on YouTube or whatever. Text is already a niche (hence declining film/book reviews also)

Not liking it isn’t the same thing as not doing it, or even not knowing how to do it well. It’s a necessary evil, one of many things in life like having to do laundry, or waking up earlier than you’d like to. Sure there are some people who don’t do those things either (and fall out of society), but there are few who deeply enjoy them. So too with small talk, it’s for colleagues and acquaintances. The way someone becomes my friend is immediately when I realise that they are open to deeper talk

Imagine being a professional music listener. That would be the life. Sure I’d have to write a ton, like every day, and go to a bunch of events with record execs probably. Still sounds alright

I’ll remind you that sailing ships big enough for piracy very much cost more than a house. If nothing else, they’re way bigger than a house, and literally house dozens of crew for long stretches of time. A huge amount of modern capitalism was invented exactly because ships were so expensive and risky (insurance, large scale banking, corporations, lots of contract law, risk assessments, time value of money etc)

But yeah. Valuable space freight passing through areas of space with complicated conditions with many safe harbours for the pirates, for one reason or another. Political (government doesn’t care that they pirate against those other guys) geographical (countless small islands/asteroids makes searching for them futile), or otherwise

People have walked on the moon.