Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Website streamed live directly from a model

flipbook.page

Post reply on HN
Print view View original

111–120 of 125 posts

Re: Website streamed live directly from a model

#111

Post by wartywhoa23 »

maplethorpe wrote: post #75
jazzypants wrote: post #44

Earlier quoted context omitted.

Do we ever simply accept that LLMs weren't made for this kind of detail-oriented work? I can't imagine something like this ever being anything other than a toy which can't be trusted. 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 think the unfortunate fact is that most jobs in the world do not require accuracy, so an inaccurate result has a negligible impact over an accurate one. 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.

I predict that this illusion of "(in)accurate enough" will last long enough to trigger a cascading avalanche of failures across all fields of human endeavours, an I'd be pretty cautious to bet on quick recovery or even survival of this civilization after that.

Re: Website streamed live directly from a model

#112

Post by mlsu »

Very cool. I think what's so remarkable is that this is basically the (ultra crappy *) version of the original hypermedia.

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.

Re: Website streamed live directly from a model

#114

Post by creamyhorror »

dugidugout wrote: post #33
thegrim33 wrote: post #23

Earlier quoted context omitted.

I decided to test it out myself. 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…

It does poorly on creative concepts as well. 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…

Are TYPE-MOON relationship diagrams the new pelican benchmark?

Re: Website streamed live directly from a model

#115

Post by totallygeeky »

I truly don't understand how people get excited for any of these products. They're comically unreliable, so usage implies you don't know how wrong it is or don't care.

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?

Re: Website streamed live directly from a model

#116

Post by VerTiGo_Etrex »

ITT: same mistakes as when GPT-1 and the first GANs for images were released.

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!

Re: Website streamed live directly from a model

#117

Post by gavinray »

bluehat974 wrote: post #101

Seems related to https://metauto.ai/neuralcomputer/ 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…

I had mentioned the same thing on Twitter and the author replied, FYI:

https://x.com/GavinRayDev/status/2047024420200522129

Re: Website streamed live directly from a model

#119

Post by rogerrogerr »

zan2434 wrote: post #51
martianlantern wrote: post #3

Cool project, but just a side thought I was having about how do people have resources and the money to make things like this and make it avl for public, I mean it's fair to say they have their own GPUs or if they are using api keys for gpt or Gemini with enterprise subsidized inference But still coming from a frugal background I still cannot wrap my head around this

I am unfortunately just paying for this out of pocket! Didn't really expect it to blow up like this.

Please write a blog post about the experience of making / hosting / paying for this. I’d love to hear about it.
Post reply on HN