Yeah I had a giggle about that also. He argues: “cloud abstractions are the wrong shape”, then what they actually ship is: a different abstraction, with even more hidden constraints.
I'm very curious how they deal with subscription levels/noisy neighbors.
Here is my regular "hard prompt" I use for testing image gen models:
"A macro close-up photograph of an old watchmaker's hands carefully replacing a tiny gear inside a vintage pocket watch. The watch mechanism is partially submerged in a shallow dish of clear water, causing visible refraction and light caustics across the brass gears. A single drop of water is falling from a pair of steel tweezers, captured mid-splash on the water's surface. Reflect the watchmaker's face, slightly distorted, in the curved glass of the watch face. Sharp focus throughout, natural window lighting from the left, shot on 100mm macro lens."
My observations have been that image generation is especially challenged when asked to do things that are unusual. The fewer instances of something happening it has to train on, the worse it tends to be. Watch repair done in water fits that well - is there a single image on the internet of someone repairing a watch that is partially submerged in water? It also tends to be bad at reflections and consistency of two objects that should be the same.
I mean, your prompt is basically this skit: https://www.youtube.com/watch?v=BKorP55Aqvg ("The Expert" 7 red lines: all strictly perpendicular, some with green ink some with transparent ink)
I couldn't imagine the image you were describing. I've listed some of the red lines with green ink I've noticed in your prompt:
Macro Close Up - Sharp throughout
Focus on tiny gear - But also on tweezers, old watchmakers hand, water drop?
Work on the mechanism of the watch (on the back of the watch) - but show the curved glass of the watch face which is on the front
This is the biggest. Even if the mechanism is accessible from the front, you'd have to remove the glass to get to it. It just doesn't make sense and that reflects in the images you get generated. There's all the elements, but they will never make sense because the prompt doesn't make sense.
The last point (reflection by front glass versus mechanism access so no front glass) is the only issue I see with it. Other than that I can easily visualize an image that satisfies the prompt. I think that the general idea is a good one because it's satisfable while having multiple competing requirements that impose geometric constraints on the scene without providing an immediate solution to said constraints as well as requiring multiple independent features (caustics, reflections, fluid dynamics, refraction, directional lighting) that are quite complicated to get right.
To illustrate that there aren't any contradictions (other than the final bit about the reflection in the glass). Consider a macro shot showing partial hands, partial tweezers, and pocket watch internals. That's much is certainly doable. Now imagine the partial left hand holding a half submerged pocket watch, fingertips of right hand holding front half of tweezers that are clasping a tiny gear, positioned above the work piece with the drop of water falling directly below. Capture the watchmaker's perspective. I could sketch that so an image model capable of 3D reasoning should have no trouble.
It's precisely the sort of scene you'd use to test a raytracer. One thing I can immediately think to add is nested dielectrics. Perhaps small transparent glass beads sitting at the bottom of the dish of water with the edge of the pocket watch resting on them, make the dish transparent glass, and place the camera level with the top of the dish facing forward?
A second thing I can think to add is a flame. Perhaps place a tealight candle on the far side of the dish, the flame visible through (and distorted by) the water and glass beads?
Without the last point with the watch glass it is also easier to imagine for me. Still, you'd have to be selective.
Do you want it to actually look like macro photography (neither of the generated images do)? Then you can't have it sharp throughout and you won't be able to show the (sharp) watchmakers face in a reflection because it would be on a different focal plane.
Dropping the macro requirement, you can show a lot more. You can show that the watchmaker is actually old, you can show the reflection, etc.
Something has to give in the prompt, on multiple of the requirements. The generated images are dropping the macro requirement and are inventing some interesting hinging watch glass contraptions to make sense of it.
Yeah, fair enough. I figure "macro" sees sufficiently loose use that a model should be able to make sense of it but to get the prompt into perfect shape that ought to be replaced with something like "a closeup showing X, Y, Z in perfect focus". Still the only real problem I see is the aforementioned contradiction regarding the front glass. Short of that single detail an artist could easily satisfy the description as written to well within reason.
Yeah I dunno bud, I have a degree in film and three Emmy awards for technical production (an expert), I could shoot that prompt (unlike the so called "expert" in the skit). Canon EF 100mm Macro USM at f32 should be able to produce that, focus doesn't need to imply aperture, and a quick google search shows me there are loads of front gear pocket watches available. Also it produced something very clearly not shot with a 100mm anyway, as the telephoto compression is wrong.
Far be it for me to add to a comment by an expert from someone who only whipped out his macro lens for ring shots at weddings and - about 2 hours ago - a picture of our latest newborn. However, I think most photographers is that situation wouldn’t shoot at f/32 due to diffraction and would focus stack instead.
Of course, a text to image model shouldn’t really need to worry about that sort of thing.
Yeah I dunno bud, I've watched a few watch repair videos on youtube and have seen macro photography which other people did.
Sure there are pocket watches where the movement is visible from the front (you'd still likely service them from the back, but alas). Even if you'd do service from the front where the glass is, you'd still have to remove it to drop in a gear.
Anyway, I think that we aren't really talking about the same thing. I'm nitpicking your prompt while you constructed it to mostly see the performance of the model in novel situations and difficult lighting and refraction environments. And that's fair.
How satisfied are you with the generated image results? What would you do different when shooting this proposed scene yourself?
Reasonable people can disagree - I think you made some good points, I've been sitting for the last 20 minutes wondering where the DoF at 32 on a 100 runs out, maybe you're right I'm not 100% sure.
The prompt I did mostly to see how it does with the gears and the tweezers, and the perspective of the gears (do they.. I don't know the opposite word of distort, straighten?, but do they seem like they're actually round, could they work?) I think those are really hard things for AI, the glass distortion, reflections the DoF etc were just to see how it approached that, and like the other comment below said, I tried to pick something that that wasn't likely to be in training data, so it reasoned about it more.
Nano was able to spit it out consistently, Images 2 really struggles, and has yet to complete one I was satisfied with, whereas with nano it nails it almost every time, the 2 images I showed originally are the first shot of the prompt with the models. (here are the 3 other gens from Images2: https://drive.google.com/drive/folders/1s8gik_x0B-xDZO6rOqoz...)
How would I shoot it? I wouldn't, fixing a watch in water is a dumb idea. ;)
Thanks, I need to get off Zight, they used to be such an nice option for fast file share but they've really suffered some of the worst enshittification I've seen yet.
As far as I recall it stared in 2014 or so, yes metrics could/were still gamed, but there was still a belief in VC that OSS projects could turn into Red Hats. First I heard of it was when a VC told me they were "looking for the next docker" and mentioned something about Rancher OS and how quickly it's stars/follows were growing. In VC you tend to have conviction builders, and conviction buyers. I suppose what happened was some conviction builders used growth of a project on gh as part of a leading indicator (valid or otherwise), and conviction buyers picked up on that as a method.
Vercel acknowledges a security incident, which nobody is claiming doesn't exist. What they don't acknowledge are this person's vague implications about impact elsewhere.
Note: what follows is absolute 100% speculation based on nothing but gut feelings.
Theo has long been Vercel supporter and was sponsored by them several times. In this case it could be a combination of him being genuinely interested in Vercel (a rare thing) and hopes for future sponsorships
Ah, Theo with his vast insights and connections into everything. That man gets around, and his content is worth it's cost.
Theo's content boils down to the same boring formula.
1. Whatever buzzword headline is trending at the time
2. Immediate sponsored ad that is supposed to make you sympathize with Theo cause he "vets" his sponsors.
3. The man makes you listen to a "that totally happened" story that he somehow always involved himself personally.
4. Man serves you up an ad for his t3.chat and how it's the greatest thing in the world and how he should be paid more for his infinite wisdom.
5. A rag on Claude or OpenAI (whichever is leading at the time)
6. 5-10 minutes of paraphrasing an article without critical thought or analysis on the video topic.
I used to enjoy his content when he was still in his Ping era, but it's clear hes drunken the YT marketer kool-aid. I've moved on, his content gets recommend now and again, but I can't entertain his non-sense anymore.
I just wanted to chime in and say I think he is knowledgeable; he's not a con. I know you didn't say that, but people might have the impression he doesn't know what he's talking about. He does know, and I've learned quite a lot from him in the past.
However, since the LLM Cambria explosion, he has become very clickbaity, and his content has become shallow. I don't watch his videos anymore.
Not that I ever had confidence in his technical knowledge, but it went to zero when he confidently asserted that there was no possible way a single server could handle the massive traffic some NextJS app he had made was serving. He then posted the bill - which was about $5K IIRC - and I was able to determine from the billed runtime and memory that a modestly-spec’d RPi could in fact handle it.
He's about as knowledgeable as the junior you hired last week, except that he speaks from a position of authority and gets retweeted by the entire JS slop sphere. He's LinkedIn slop for Gen Z.
I don't watch his content, but I felt comfortable posting his link as I believe he's generally considered a reputable guy? His tweets sometimes come up in my for you tab and he seems reasonable and knowledgable generally? Maybe I'm wrong and shouldn't have linked to him as a source.
He's kind of like an LLM in that his content has the surface texture of something substantial, and sometimes it's backed by substance, yet it's often half-true or totally off the mark too. You'll notice if you're previously acquainted with what he's talking about, otherwise he seems to be as you described.
I don't think he's a bad guy or that he's trying to be misleading. I suspect he wants his content to actually carry value, but he produces too much for that to be possible. Primarily he's a performer, not a technologist.
Yeah that's my mistake - Sorry! I just went back and and re-read this tweet: https://x.com/theo/status/2045870216555499636?s=20 - I had read it as via their Linear/GitHub - I should have known better and double checked what I read when I posted this, please accept my apology, I can't edit my original comment now.
> @theo: "I have reason to believe this is credible. If you are using Vercel, it’s a good idea to roll your secrets and env vars."
> @ErdalToprak: "And use your own vps or k3s cluster there’s no reason in 2026 to delegate your infra to a middle man except if you’re at AWS level needs"
> @theo: "This is still a stupid take"
lol, okay. Thanks for the insight, Theo, whoever you are.
You'll have to ask @ErdalToprak on Twitter on that one. I just thought it was funny that this slopfluencer, who's taken money to advertise Vercel, ostensibly believes that using a VPS/k3s is "a stupid take."
That first link is not relevant to the point of my comment. I was not complaining about paywalls. The comment also doesn't address whether paywall bypasses would be acceptable for non-text links.
Regarding the second link, I'll happily engage with something specific dang said on this topic if you want to link to it, but a link to every time he said the word "paywalls" is not a productive contribution to this conversation.
I am autistic and I also enjoy the sharp edges, I rub my wrists up and down them sometimes and generally play with them, I find it very satisfying. I also suspect the laptop might not be as easy to carry around when open if edges were rounded?
I noticed there was a respiratory epidemic that killed 25 chimps naturally quickly, one would imagine that would have quite a societal destabilizing impact?
> a respiratory epidemic that killed 25 chimps naturally quickly, one would imagine that would have quite a societal destabilizing impact?
there were several seemingly destabilizing factors, sort of a perfect storm, each contributing to further disconnect and polarization.
the group grew too large (and displaced other groups), but then ended competing for the best food among themselves, and having trouble socializing and bonding in such a large group.
subgroups forming, first fluid but eventually creating a split
loss of older alpha males exacerbating competition between males
loss of the few individuals that still maintained some relationship with the other group (the last one doing so actually died in that epidemic while the split was already well underway)
it is indeed an amazing read. my take away is that the root cause was mainly the group becoming too large, this affected socialization and cohesion, and thus the group was unable to cope with everything that came after.
Makes me wonder if civil war is more common for larger countries. Reminds me of the phenomenon where Latin American countries pretty much all broke up after independence from Spain.
My initial instinct is that they were just reestablishing social order among the group after such a dramatic event.
Edit : I just read the paper and the discussion does a good job at laying out the entire landscape that contributed to the disruption. Pretty fascinating but also totally explainable due to the circumstances explained, which in and of itself is wildly fascinating!
Sudden power vacuums are often filled by the most opportunistic individuals in human culture. People who are frequently more concerned with personal gain over the collective well-being of the group. It's why assassinating heads of state usually just makes the situation worse.
Plenty of people stepped into power vacuums not to make themselves rich but to save their nation Napolean, Tito, Cincinnatus, arguably George Washington.
Just like in human analyses of geopolitical situations, the explanations that rely on broad abstractions of human nature or resource competition and paint a teleological narrative always end up breaking down when you do a deep dive into the history and specific circumstances. When you get into the nitty gritty of every unique geopolitical situation it's actually much more difficult to pull out a generalizable lesson imo. At some point we have to accept that we can't cross the same river twice
I think some of the individuals who died were key in linking the two groups (they were "the glue" that prevented disruptive aggression), and after they were gone, the split cemented and later turned into aggression.
I wonder if chimps are sophisticated enough to believe in omens? Perhaps they saw the sudden deaths are some sort of sign that the established structure was weak or immoral.
I could imagine if you where friends with someone and a bunch of their friends suddenly and mysteriously died, personally, I wouldn't kill that friend, but I might call the cops.
Here is an example of one of their core growth plan items from the strategy above:
"Social Media Campaigns, Organic and Paid
Driving key messages around digital hygiene, decentralisation, and security on social media platforms to raise awareness."
The whole pdf is basically a collection of the remedial "go-to" SaaS growth blog posts everyone thinking about startups read: make content, build a community, turn your community into advocates, write about things people care about etc etc.
Given I've done this stuff for some 20+ years now, here is what is missing and frankly what most folks miss/don't want to admit:
This document basically has no ICP, who is the ideal customer? What is their persona? Who specifically are they, like, super specifically! You can't start with "oh anyone who wants anon-privacy first msg'ing!" That would have been like me at digitalocean saying "oh it's for anyone who needs a VM" - you can't execute a series of steps with that, you can't boil the ocean so to speak, we had to work through communities one at a time, we did: rails, node, php, devops/config management, in that order, split up over quarters and years, maybe it looked like we just...did developers, but we didn't, we slowly worked our way through all the developer communities slightly tailoring towards them while keeping things general enough.
The biggest problem here tho is the classic vitamin vs. aspirin problem. They're selling "better privacy" and "decentralization" - these are vitamins for the vast majority of people - they're things people say they care about in surveys but don't actually switch apps for. The 85% of adults who "want to do more to protect their privacy" aren't switching off WhatsApp. Are they the most secure messenger, or are they a token ecosystem with staking? Those attract fundamentally different people with different motivations...so just bolting them together creates confusion.
Folks need to stop thinking "we're going to do marketing" = "we're going to build a business" marketing, go to market, growth.. these are tiny components of overall business strategy. </rant>
I just can't help but imagine those long lived rats escaping and taking over our cities. We already struggle with the rat population and they only live 3 years, imagine if they lived 15.
Also I love rats and totally agree they're tough pets because they don't live long enough. We had many of them growing up and spent a fortune removing tumors.
I thought rats would make great pets but the idea that you get attached to a guy who would die in a couple years is quite discouraging. You updated my knowledge.
My grandmother had an African grey parrot she had inherited from someone.
Hey there, sorry to confuse you but that was a bit of a joke, real rats max lifespan is 7, but we have those other rats that will live to 80 or even 90... But yes, the parrot issue is real, people will get those without thinking for even a second about how old they can get. We have one in our family that has outlived three owners.
Funny because I have dyslexia and read excreting power as exerting power, and then had to read your "Exerting" underneath 4 times to understand the mistake. I guess it's the phonics, dyslexia is so weird tho, ha.
Hey do you have certain fonts that are better? I was working with a dyslexic student last week trying to find fonts that work better for his online classes. All the research pointed towards a handful that didn't seem to really improve processing for the student.
They tried all sorts with me in school, I seem to recall it's related to trying to add shadow to hint to the brain the direction the letter should be etc. I found it more annoying than helpful. Probably a very unpopular opinion but I think teaching someone with dyslexia to read and write neurotypically is probably unhelpful and finding audio visual learning methods is a considerably better way to have them retain knowledge. I think you can get to a basic level of competency but speed and recall, at least with me, never really came. One thing I found once that was cool was an app that present each word at a time only in the center of the screen, but it felt extremely mechanical I was so focused on the words once I was done there was basically no meaning left if that makes sense. I'm autistic with dyscalculia also, FWIW. I mostly think in sounds, pictures and movies, for whatever reason my brain doesn't have a great framework for symbols that don't have those things inherently attached to them. ¯\_(ツ)_/¯
I'm very curious how they deal with subscription levels/noisy neighbors.
reply