Finally got around to test progress on image generation again.
My go-to test is creating an entire fake Instagram influencer from scratch. That nicely tests consistency between images, spacial understanding of the scene, prompt following on minute details, ect. It keeps me up to date on what the problems are when you fake photos of people (or fake entire people in general). I mostly create women, because that's more fun to me and also because this tests model censorship more effectively - the commercial models are a lot more touchy when creating women than when they create men.
The main result of my most recent session is particularly funny: Nano Banana 2 is another significant step forward on photo-realism, but it is exceedingly difficult to get it to produce images of conventionally beautiful people from scratch. Getting just a portrait of a woman that is above a 7 requires a lot of coaxing. If the major focus of the prompt is on some other detail, it will generate the most mid women you've ever seen. Nano Banana 1 was perfectly happy to just spit out 10s. You start the prompt with "photo-realistic full body shot of an attractive female college student..." and you could focus on scene, clothes, body position, camera equipment, ect., and it only needed minor coaxing for some body types and poses (as long as you kept it SFW). But Nano Banana 2 will often simply ignores instructions that coax other models towards conventional beauty. I wonder why. Peak body positivity seems long past. Did earlier models train predominantly on pictures of influencers on social media (because they post so much), and now photos of the rest of humanity have a more proportional ratio in the training data? Or are they trying to stop me, in particular, from creating and monetizing an Instagram e-thot? (I'm not, of course, I've lost interest in image generation, again, very quickly).
Other than that: prompt following is truly impressive now. You can pick scene, clothes, and body positions (either by describing them or supplying reference photos), and it will usually one-shot them down to the correct head tilt angle. Consistency (same person in different images) requires a bit of care, or ideally tons of reference images. We're not completely out of the uncanny valley for faces created completely from scratch, but this is where I notice the most progress (Nano Banana 1 makes beautiful people, but they look like influencers with the filters maxed out in the best case, and like very good paintings in the median case). Around 1% of images still have extra limbs or other easy tells.
Oh, and making images that help explain a technical concept is still hilariously bad. A straight rip-off of an existing image with a liberal dose of detail errors is the best you can expect. Ah, factual correctness in every detail... the old nemesis of AI still lives on.
They're making super dangerous airborne diseases in ferrets... For no good reason at all. [...] So there is no value to this research. All we know is that 'this specific disease could be super dangerous' and they helpfully put its genome up on the internet.
But so far as I can tell nobody had anything to gain besides publishing some 'good' papers.
I think this misunderstands what they are trying to do. There has been, for a long time, a large community studying virus evolution and spread. And if you monitor influenza, globally and in different species, you ideally would want to know what you're actually looking for. No use having petabytes of genetic data, but no way to actually analyze it, or really, make some useful predictions.
Gain of function research tries to help with that. Identify which changes/mutations are actually worth watching. Identify what will spread fast, what will go airborne, what will kill, and what might jump to humans. They hope that next time, we'll have a bit more advance warning, or maybe a vaccine approaching the effectiveness of the polio shot. Having so damn much antigenic drift won't safe influenza if the vaccine directly targets what makes this specific strain so successful.
And all this isn't just a "saving humanity" moonshot/insurance against a black swan event. There's very practical applications - it would be nice if we didn't have to destroy 100 million chickens every 5 years because the farms got infected with bird flue, again.
At least that's the dream. Whether that's actually doable and/or worth the risk is another question. Maybe those people should really be send to Antarctica, or onto a decommissioned oil rig only serviced by a very slow ship. Maybe their current security measures are fine, I haven't updated on the COVID lab leak story an a while. But during the thick of it, I found the arguments of the counter-side more convincing.
pbmonster
1d ago
Though I know some people say OpenTTD is now the best version of a railroad builder
They must be using a peculiar definition of "best". It's a good game. It's elegant in its simplicity, and it's amazing how large you can make the system using so few rules. The UI is historic, and hilariously bad. It has been hugely influential on transport games.
But it's not Railroad Tycoon 2. The depth and complexity is just much higher there. The game gains so much by having an amazing campaign. It gets genuinely difficult towards the end.
we can just look back to what Iran looked like in the 60s and 70s under the US-backed Shah.
Wait, wait, your scheme hinges on Iranian royalists/secularists/liberals fighting your (ground) battles? Ramming through regime change from the air, once again going out to win hearts and minds (successfully this time, for real!), and then having them do the dying? Watch from afar while they de-islamize every single institution and hunt down the mullahs? I'm very skeptical. They don't have history, especially recent history, on their side. And neither the training, the cohesion, nor the morale.
Frankly, the IRGC is going to eat them alive.
the kind of violence you're describing- massive use of MANPADS and sophisticated explosives- doesn't just magically happen. It happens when some other country with sophisticated weapons factories is sending them weapons.
Well, if the IRGC doesn't have absolutely enormous stashes of at least small arms (and probably even drone/rocket parts plus warheads) in the mountains, they didn't plan even a week ahead. I'd be very surprised. Also, I'd expect Russian surplus equipment and Chinese dual use goods to make it across the border. But for any of that to even matter, they'd first have to lose the cities, which would be orders of magnitude more bloody than Mosul or Falludscha.
pbmonster
5d ago
How do you imagine the country to look like under US rule?
How do you "easily" stop the IRGC (and its successor insurgency groups) from perpetually setting the oil fields on fire, blowing up the pipelines, attacking every single supply route you run through the mountains, firing MANPADS at every single helicopter and airlifter that dips below 15k feet and generally IED-bombing, droning, mortar-ing and rocketing every US installation in the country? How do you stop them from infesting every town you turn your back on for five minutes?
Is it just a million miles of barbed wire fences with autonomous auto-cannon turrets? Because you certainly can't just kill them all. Trying that always results in recruitment automatically out-pacing your kill rate, and Iran has a - for all intends and purposes - infinite recruitment base.
Overwhelming force; casualty rate will be higher, but achieving ultimate victory will be popular.
Are you sure the US could achieve ultimate victory over Iran and rule it for 100 years? I'm not convinced. Even after removing all rules of engagement, glassing half their cities and sweeping every single mountain valley 10 times, it still would just be insurrection after insurrection.
I'm actually having a hard time thinking of a single combination of society and geography more capable of resisting foreign rule.
The... province would certainly have close to zero economic output. The infrastructure necessary for any sort of economy would be to easily destroyed by insurgents. So what's even left of the idea? Trading army brigades for a thin justification of genocide?
pbmonster
7d ago
people who use the phrase use it in the way you would say "late-stage cancer"
Or if they want to make a more historical point: they use it in the way you'd say "Late Republic", the period of the Roman Rebublic that was characterized by civil wars, mass slavery / slave rebellions and internal instability.
Same result, really.
pbmonster
7d ago
IMO you’re better off dove hunting if you want to switch to mostly birds.
Interesting, never met anybody doing that. Those birds are tiny, right? Can you reliably shoot enough for a full meal for multiple adults? Can you fill a freezer to have some for the off-season? Do you need a dog to find the birds, or do you go looking for them yourself?
pbmonster
7d ago
Same. I think my snobbishness saves me. If I swipe through Youtube shorts, my brain doesn't dump dopamine, it goes "this is shit" over and over again and gets more annoyed with every swipe.
Video is the wrong format for most content, and short form video is the wrong format for very close to absolutely all content. The very few exceptions to that rule where discovered 10 years ago on Vine, and done in thousands of variations since.
pbmonster
7d ago
It's coming, and soon. Zero-knowledge proofs for age are in the design of the age-verification framework of the EU Digital Identity Wallet, and in the specs of the Swiss eID law they passed a while back. Both involve an app on your phone holding your ID and your crypto keys and generating ZKP responses to things like age requests.
Both designs are decent in my opinion. Once you've come to terms with the slippery slope that we'll soon have digital ID checks everywhere, all the time, there's not much to criticize. It's probably the best way to do it, if we agree that we need to do any of that. But also, it's pretty far from a "very easy way". This scheme absolutely needs a central authority (probably a national government) doing the final ID/age check and then the issuing of crypto keys. I'll be curious how the US handles this. I expect Google/Apple to take over that task, since the majority probably won't trust the government to do it right...
pbmonster
12d ago
ion engines still have much higher specific impulse, but are only viable in space
True, I ignored those since they're not practical to push payloads to Mars or beyond. Insufficient thrust to power ratio, if you want to move serious cargo you need so much power that thermals become unmanageable. Other engines use the propellant as coolant, which is kind of genius - but also requires large propellant mass flow. Still, very useful for low mass probes!
And you're still sidestepping the point that upper-stage nuclear rockets (the original topic) and large nuclear payloads are completely separate issues.
Sorry, didn't mean to sidestep. I think we're just talking about different scenarios. Launching a nuclear thermal rocket from earth is an option for a non-stop express to anywhere in the inner system. There's plenty of use cases for those.
But up-thread and side-thread, people were discussing "serious space travel", "permanent Mars colonies" and "asteroids" with nuclear propulsion. In that case, I assumed we would do more things than just sending resources from Earth to X. A system-wide economy requires more fuel, and while the hydrogen can come from anywhere (luckily, since you need to refuel all the time), the Uranium can only come from Earth. And once you start pushing asteroids, you need to get up a whole lot of fuel.
OK, so maybe same scenario, but different timescales.
pbmonster
13d ago
One is the launch, from Earth, of a nuclear-powered rocket (e.g. NERVA).
"Current" designs (well, currently available 1960's designs) of nuclear powered rockets aren't useful for launching from the surface. While they have by far the best efficiency/specific impulse of all engines available today, they have catastrophically terrible thrust to weight ratios. Absolutely useless engines for first stage and even most second stage applications. You'd only want to use them in space - then their low thrust doesn't matter, and they use their high fuel efficiency to cut down time of a Mars transfer by a factor of 3.
it's a lot fairer to compare that to an A-bomb like Little Boy (64kg) rather than just the primer of an H-bomb
The vast majority of atmospheric tests where tactical warheads with a boosted fission core. Those - just like H-bomb primers - always contain subcritical amounts of plutonium (4kg) for efficiency and safety (they can only fission if explosively collapsed correctly into a critical mass) reasons. Pretty much the only devices with larger amounts of fissile material are H-bombs with second stages and tampers. But even those are much, much lighter than Little Boy, and they weren't tested all that much.
And I'm not even sure how valuable nuclear rockets would be for long space trips (there are lots of options once you're up there).
Extremely valuable! Even the most primitive and conservative designs outperform chemical rockets by several hundred percent (again, in specific impulse). More batshit designs (nuclear pulse propulsion and nuclear salt water rockets) are probably technically doable today, and offer orders of magnitude more specific impulse. Those would actually unlock the outer planets and the asteroid belt, and maybe Alpha Centauri.
Sustained fusion is already difficult enough in containment, actual fusion propulsion is probably orders of magnitude more complex than that. I have no hopes to still be alive when it arrives.
One rocket's worth of nuclear material in the atmosphere is barely a blip.
I'm not convinced. One NERVA style nuclear thermal rocket engine contains hundreds of kilos of uranium. Put one as an upper stage engine on a SpaceX booster and you can lift another 100 tons of cargo to orbit - which quite frequently will be 100 tons of U-235 (or 233, since we'd probably quickly get into thorium breeding if we'd consider such a project). We want to fuel an economy the size of a solar system, after all, and earth is the only place in this economy where it would be economical to mine Uranium.
Compare this to the ~4 kg of an H-bomb primer, and vaporizing a nuke fuel truck sounds a whole lot more catastrophic that an atmospheric test.
The interesting part is the "vaporizing" here. I'm pretty sure that most failure modes of such a launch would not vaporize a significant fraction of the payload or even the engine cores. The "fallout" would quite literally be tens of thousands of 1-kg pits (and a few fuel pellets) raining down from the explosion. Compared with the alternative, that contaminates a much smaller area. Manual clean-up would be possible, economical and necessary from a proliferation (and ecological, of course) perspective.
pbmonster
14d ago
This is the truest blackpill IMO: collective agency has been eradicated from Westerners, and it will take at least a century to rebuild the necessary infrastructure to produce it again
Overly pessimistic, I hope. I think it could/would be rebuild much quicker through a collective crisis. I'm watching China with a lot of interest partly because of that. I think a descend into a Cold War era international order, followed by a defeat of the West at the hands of China - could be a relative minor thing like Sputnik - would jump start the entire thing right quick. Probably requires media alignment and effective political leadership, but I assume that would emerge more or less naturally.
pbmonster
14d ago
I remember when I was a teenager I could feel such passionate crushes and such intense butterflies, but by the time I made it through college I couldn't really feel much of that at all.
I think that's normal. That's what growing up is, and it happens to everybody. That first crush, that first kiss, that first love really hits different, but the butterflies mostly are just adrenaline. And the novelty makes it special and forms stronger memories.
It's sad in a way, but I find it reassuring that it happens to most people, even the people who end up being married to their first crush for 60 years.
pbmonster
14d ago
With those brands, you can also always get a 15 year old model with around 100k miles. It will perform equally well as the 3 year old model, but be substantially cheaper to buy, easier/cheaper to repair (do research, get a model that has a reputation for being reliable and common), and it will not depreciate at all if you drive it for 2 years and 20k miles.
I'd just get a Toyota with hybrid drive from that era.
That said, I agree that nuclear (or WMD) inspection is at least theoretically possible.
I always wondered about that in the case of highly industrialized nations (or nations aided by one of those). How hard would it be to secretly build a large centrifuge stack and then obtain either a lot of ore or a bit of high assay low-enriched fuel (basically 20% U-235)? With tight integration, you could spin that into a bomb in a week.
Are the fourteen eyes really that good? Will their spies notice and report inconsistencies in, I don't know, centrifuge bearing part inventory and then locate where exactly those ended up in time? If you keep pouring state grants into small modular reactor startups (there's over 100 of those today) and two dozen of those companies end up needing 20% enriched fuel right around the same time, and all those fuel shipments get confused at the post office and they end up getting U-238 pellets by accident... will the fourteen eyes see?
The classic nuclear threshold states are pretty clear, I guess. If Japan or South Korea want the bomb, the time window to stop them will be tight. Still, I'd be curious if all their centrifuges are accounted for (and if there's bunker busters on the shelf that already have those coordinates pre-programmed). But could Australia or Canada cook the books at their mines and start a little stockpile on the side? Could Germany repurpose all that fuel just sitting in those mothballed reactors on the down low?
pbmonster
22d ago
That doesn't make much sense. Being aware of, and agreeing to, a power imbalance doesn't make it go away.
And I agree, every couple should have those discussions. But going into the discussion with a younger man would obviously change the stakes, and the style. There would be, on average, more room for compromise. And sure, if both actually want the exact same thing in every aspect of life, that doesn't matter. But most relationships don't have a 100% match, so compromising is important.
Same with the threat of divorce. As long as one party is significantly more fucked by separating, there is a power differential. Agreeing to it doesn't make it vanish.
pbmonster
22d ago
But none of what you're describing would have been unknown, to either party, even at the start of the relationship.
That doesn't make a difference, does it? A professor dating his student is an obvious power imbalance, no matter how aware both of them are at the start.
just sounds like they're both giving something up (money or choice of where to live) and both getting a lot out of the relationship.
Sure, still different than a couple graduating together and deciding where to move together - who then also get a lot (but often different things) out of the relationship.
Can you define this for me? What does "power differential" mean in the Western context?
There's only very few 10+ year age gap relationships in my extended bubble, but those I can think of have clear power differentials: the guy already owned a house and was established in his career when she left grad school. This means, once she decided to enter that relationship, he got to choose the city they would live in. She's also, by not pushing for it in a prenup, not on the title of the house.
At some point, his pension scheme is going to allow him to retire, maybe even retire early. Whether she will continue to work or retire extremely early herself - together with him - will probably not feel like her choice.
She could have pushed against all that, but by being older, a lot of the default choices were already locked in by him. It would have taken a lot of effort to change some of those defaults, and realistically, the relationship would not have survived that effort.
Oh the other hand: free rent, lots of disposable income, friends in similar situations, a network to boost her own career... certainly nice perks, but I bet she wonders how much of that would survive a divorce.
pbmonster
22d ago
That's the thing, I haven't notice the frequency of incorrect output to go down significantly! It just gets more and more difficult to detect the errors.
If you enjoy history, I can recommend the Asian Civilisations Museum. One of the better ones I've visited, specifically it is orders of magnitude better than every single anthropological/historical museum I've visited in Japan. The National Museum is more about Singapore itself, but I also found it well done. If anything, the museums are air-conditioned, which will make them very much appealing once you're exploring the streets. The city also has a nice selection of different styles of temples, I decided to visit all the major ones just as a mission to see many different parts of the city on foot.
If you like Asian food, I've had some of the best Chinese and Indian food of my life in Singapore, both street food and fancy restaurants. Selecting restaurants beforehand is probably worth it.
When I asked Singaporean people what they like to do, they suggested going to one of the mega-malls, with the cable cars to Sentosa Island and (of course) to Marina Bay. The two former I flat out hated, they are not worth doing, unless you really have time to kill or need photos for Instagram - they look much better in photos than in real life. No matter what, absolutely do not go swimming at the beach at Sentosa, that must be just about the worst touristic beach in Asia. Marina Bay (the towers, the gardens, the indoor jungle, the supertrees at night, and the area around it) are more or less obligatory for tourism in Singapore, but just like the main Botanical Gardens and the Zoo I found all that just extremely... mid. Unless you like park design and architecture, it won't even fill a full day.
Two weeks is a lot of time for a "young" city like Singapore. I would strongly recommend long excursions into Malaysia instead of staying in the city, and I regret not leaving for trips sooner.
pbmonster
23d ago
Seriously, compare gpt-4 and gpt-3 output, this is not something that can really be disputed by any thinking person.
I dispute it. Both suffer exactly the same problem: the output they produce is frequently wrong in subtle and insidious ways. This makes both equally useless for work that requires correctness, especially correctness you can't write unit tests for.
No, that's not the problem. "Prominent researchers" don't ever get close to the contagious ferrets unless it's necessary for a press photo to go with the release of their most recent Nature paper... The people producing the data on the oil rig lab, or in the antarctic darkness or who are stuck on the slow quarantine supply boat are grad students and lab techs. They'll do it for the paper, the title, the story and the love of the game.
The problem is that oil rigs and research labs in Antarctica are more expensive than university basements in cities.
More options
Context Copy link