Website streamed live directly from a model
flipbook.page
flipbook.page
Genuinely one of the most impressive demos I've tried in a long time. I was able to use it almost like a living version of a classic illustrated Haynes workshop manual.
Went to the website, typed in "Jeep Wrangler JK engine bay with components labeled" (Since I'm intimately familiar with JK engine bays). Seems like a pretty analogous test to what you did, if anything an even easier test.
Let's see what we get .. a very nice looking diagram of a wrangler engine bay with components labeled, looks good.
But wait ..
- The brake fluid reservoir is on the wrong side of the engine bay
- Where the brake fluid reservoir is, it's labeled as the coolant overflow tank, and while the actual coolant overflow tank does exist in the diagram, it has no label.
- The battery is on the wrong side of the engine bay.
- The top of the front grill is labeled as the "oil filter cap".
- The oil fill cap is in the wrong place.
- Half of the battery is labeled as the fuse box, when the fuse box is correctly shown, but unlabeled, on the other side of the engine bay.
- It shows two different windshield washer reservoirs next to each other.
I could keep going on ...
Now I tried clicking on the incorrectly labeled coolant overflow reservoir and it switches to a new page which now shows a completely different looking coolant overflow, but now it's at least located in the correct place in the engine bay.
But of course it doesn't look remotely like the actual coolant overflow container. It also shows the radiator cap as on the top of the coolant reservoir, when in reality it is very much on the top of the radiator itself.
Like .. I can find fault with every aspect of it. But of course, if you didn't actually know much about the topic it'd all look fairly believable. The story of LLMs basically.
I attempted to explore the works of Kinoko Nasu/TYPE-MOON through its characters and the relationships across works and it was mostly nonsense. Sure it had some broad relations correct, but it presented a tiny set of meaningful characters and only attempted to touch Fate/Stay-Night and Tsukihime.
Even more damning was that it produced garbled text for a few of the textual representations and often even if the lettering was clean, the grammar was off.
Will Silicon Valley executives ever accept this reality? If we acquiesce and admit that LLMs are a good tool for prototyping and boilerplate-reduction, but not finished products-- is that when the bubble finally bursts?
I used to feel job safety in the knowledge that AI labs weren't likely to solve the hallucination problem. Then it dawned on me that they don't need to — they just need to reduce our collective expectations.
We've never had to act with surgical precision except in matters of math/science/engineering.
Like how you fill up your coffee cup up to a level probably +/- 50ml each morning.
When you fill up your coffee cup in the morning, I sure hope you aim accurately and don't pour half of it all over your desk. And don't even get me started on the process of making coffee that isn't completely unpalatable.
I first asked it "how big are geckos". It gave me a cool comparison diagram between three gecko extremes (leachianus, Jaragua dwarf gecko, and leopard gecko, if curious). Info all looked correct. Drilling into the Jaragua brought me to a less-impressive page with utter gibberish text and duplicated info boxes. So it goes. I drilled further, but they were more esoteric topics I'm less versed on (lamellar setae), I can't evaluate the accuracy without further research.
I also gave it something broader: "tokay gecko". More duplicate info boxes, and for some reason it "drew" two geckos on top of each other. Kind of cute, but tokays are extremely territorial, so happy cohabitation isn't their default (though it's not unheard of).
Still, despite the issues, I thought it was very neat.
It did a diagram that has absolutely nothing to do with the actual stage, not even close. And tells me a complete whole slew of completely wrong information. It shows pod of dolphins that teach you to dash attack (you know it by default). It shows a power sonar crystal (the sonar is a default ability, there is a "power" sonar I guess, but it is not obtained from crystals, and while the game features crystals, there are none in the game until level 3 and they look nothing like the diagram's). It shows air pockets... which are just bubbles (In the game, there are actually air refilling bubbles, but air pockets would refer to a small bit of open air in an underwater tunnel, like, the actual, you know, real life geological feature.)There are some medusas far off in the background in the image (They're yellow. The ones in the game are clear. They are also not present until later levels). An exit cave leads to the sea of silence (An actual stage. Wrong game.). A random cave says "Health source" (???? You do heal by eating fish but???). There is no warp ring.
So basically, the ONLY correct elements in the diagram are the presence of dolphins and the fact the diagram is labeled "Home Bay". Every single other element on this is wrong and would be wrong for all iterations of the Home Bay.
For a visual search tool, this sucks at visuals.
Like every other AI demo I've tried ever, impressive on the surface, but the system fundamentally doesn't understand what it is doing
But still coming from a frugal background I still cannot wrap my head around this
and it's really slow. I didn't end up waiting. Not a slight to the creators, let them create. It's just really freaking slow I didn't wait.
The sample videos on the tweet are very very cool.
Unfortunately it didn’t really work for me, I’ll try it out in a few days when the traffic’s died down.
Gemini generateContent request failed: { "error": { "code": 429, "message": "You exceeded your current quota, please check your plan and billing details. For more information on this error, head to: https://ai.google.dev/gemini-api/docs/rate-limits. To monitor your current usage, head to: https://ai.dev/rate-limit. ", "status": "RESOURCE_EXHAUSTED", "details": [ { "@type": "type.googleapis.com/google.rpc.Help", "links": [ { "description": "Learn more about Gemini API quotas", "url": "https://ai.google.dev/gemini-api/docs/rate-limits" } ] } ] } }Good luck.
snEED
fEED
sEED
chUCKI would have been so impressed if it got it right.
Asked it about how to count, half of the diagram was incomprehensible/wrong. Asked it about cryptocurrency basics, again, the way the information made no sense and several points were vague to the point of being incorrect. Asked it about the seahorse emoji, it made up a shit wholesale. Entirely fabricated page.
This is what we're reshaping the economy and computing world around?
Remember when video generation looked like this 3 years ago? https://xcancel.com/minchoi/status/1911275924458270906
Maybe try prompting GPT-2 or 1 again? https://huggingface.co/spaces/mkmenta/try-gpt-1-and-gpt-2
The first automobiles could only go 10 mph. The first flight only lasted 12 seconds.
A good investor or inventor sees the potential of an idea, not just the limits of the current instantiation. This indeed gets details wrong, but anyone who thinks we’re not headed in this direction in the next half-decade or so will likely be woefully incorrect.
Give it room to improve. Don’t claim victory before the battle has even begun!
Reading this, I couldn’t help but think of Ray Kurzweil’s idea that we’ll eventually spend most of our time in virtual environments near 2035 with help of nanobots. This feels like a specific, functional twist on that prediction.
Instead of just a digital world to hang out in, the 'neural computer' is like a specialized VR generated by AI to solve a specific problem. It’s a version of reality that adapts itself entirely to whatever task is at hand. It's fun!
https://flipbook.page/n/12267bbfdeb043c3aa477337950b2b71
- M2 is labeled as GPU
- GPU is labeled as M.2 and RAM?
- RAM is labeled as GPU
- Random plant inside the case?
- This is also not a typical layout for a SFF PC
Great demo, interesting transitions and UI, but the model / generated information is definitely not correct.
You can do this with any agentic coding model now... just ask OpenAI codex : generate me an image given this query: "the query" and you will get the same kind of outputs.
If often identifies some of the points of interest correctly, but their spacial location relative to each other is completely insane. Like, not even close to reality.
It's neither streaming NOR live, by my definitions.
It's generating single images, one by one, based on a prompt. So it's automating image generation via calls to Gemini. That's it.
It's not generating "websites", it's generating an image based on a prompt, then generating a new image based on the previous image, and the position of where you clicked it, and a prompt.
The author has massively exagerated the complexity of this, you can one-shot this with a single prompt in Claude or Codex.
It was "network is the computer", now it will be "model is the computer", and the model will be like one large ("multi tenant" - it will know on its own how/when to separate tenants' data and when to analyze it all together) model living on tens/hundreds of millions of nodes in AWS ... the AWS itself will be just that model.
For this to really be practical you'd need a way to run networks many times faster and more efficiently than today's GPUs. This is too slow to work even with cloud GPUs powering it.
Maybe someday.
General design was correct, and it included the name of a town just nearby.
Not a surprising result, but made me reflect on what a weird world we now live in.
This very well could be a sneak-peek into how educational resources might look like in the future.
A person who knows something puts up a page, and uses hyperlinks to link to other pages. Those other pages have information from other people who are sharing their knowledge.
Of course, that isn't how the internet works now. Everything is a platform/app that wants to maximally surveil and control their users as much as possible.
* Not talking about the work of the person who made it; it's very cool and great UX. Moreso that the models themselves are an expensive, smeared, inaccurate/dishonest approximate version of actual human experience/expertise.
screenshot: https://images2.imgbox.com/ff/84/j2FCxyrD_o.png - top right callout
I wish I could share all of the things I clicked on afterwards, exactly in the order in which I tried to tell the story. When I try to use the share feature on anything below the root node, I get "This page could not be saved for sharing." However, the video generation does work in that share modal, in the way I would expect. [1]
Super cool project.
[0] https://flipbook.page/n/aa99d756f5aa4fd6bc617106c8d5077c
[1] https://drive.google.com/file/d/1_YUr8NoIhB5DEYQkR_G3QjcgNz5...
I went from Cat Photos into History of Victorian Cat Photos With Props like Miniature Tea Sets And Velvet Chairs And Humorous Captions On Calling Cards In Visually Ironic Aristocratic Cooperplate Font The Victorian Meme Script With High Stakes Expectations Anchored In A World With Human Dignity As It Relates To Modern Memes in just a few clicks.
Oddly specific, but that was exactly what I needed to see today.
But that's totally missing the point: this is a demo of what seems like a totally new interaction model that could not previously have existed. It's a small glimpse at a possible future where those nit-picky details get better.
Remember when image models produced grotesque hands? Now look at GPT Image 2 and Nano Banana Pro; those original nits are mostly a thing of the past.
Take a look at this not for its immediate flaws but for the new doors it opens and the expansion of our collective imagination.
Neat project though!
For example, here I asked for a mechanism of converting the circular motion to wing flapping, and it has no idea what to do at all.
https://flipbook.page/n/21f96ba33aa94852bc1f567bc5cd23bf
The core problem with using AI to learn is that if you don't know about the specific area (which you don't, that's why you are learning it), it can (and absolutely will) fake knowledge without you ever noticing it, and be utterly wrong, disseminating false info, and teaching you (or your kids) wrong world models.
In the age of such enormous computing power, this sort of thing is pure waste.
MS Encarta CDs were faster and more in-depth.
This is built from the collective works of all humans throughout history who have strived to make infographics, illustrations, and communicate knowledge - with 0 actual credit or reference to them (or financial compensation, if they’re still alive).
Instead, who is making money from this? Google, as providers of the model - and maybe the founders of this product, if they ever choose to monetize it somehow.
I’m not even going to get into how the results it produces have just enough “insight” to appear valid but the moment you inspect it up close, it’s completely wrong in most details, and replete with ornamentation that doesn’t actually add to meaning - a Potemkin village of knowledge - because the common answer to this criticism around here is “just wait 6 months bro the models will definitely solve all those problems”.
(some of the things I’ve tried for reference)
https://flipbook.page/n/56fed5ac8e164467b1d6151a6d5068ae
https://flipbook.page/n/deeb4d846d1a44738aa70d8973fc5765
https://flipbook.page/n/335fb5d4c4d8428d82e8a43fc4f7a4e8
We are not better off investing billions of dollars in computers doing this over paying humans to write and illustrate and make cultural artefacts. We are not better off putting this in the hands of kids rather than meaningfully designed resources & curriculum designed by humans.
What are we even doing.
That’s the biggest problem with the current wave of AI tooling - it’s so easy to make a cool demo all while completely missing the point of what actually is good for human flourishing.