Hacker News Reader

JayStavison Fireball Spotted over Northeastern USA

I also saw one of these ~10 years ago at the end of December. I was in South Florida, and it was up in the sky long enough for me to tell everyone to look up, so five or six other people saw it too.

It seemed much bigger/longer than the one listed here, wonder if I'll be able to find it in the fireball database.

JayStavison The future (and the past) of the web is server side rendering

Point certainly taken but I think that "rendering" is the overloaded term.

Rendering basically means, to take data & logic and transform it into a view for another system (or person).

Graphical rendering is probably the needed operative word for this point? A bit of annoying semantics but I think rendering just means to provide a structured view for some state.

JayStavison WebGL Performance Benchmark

Very interesting.. I would love to see this dusted off, the repo's last updates (and assuming this post?) are from 2013.

JayStavison Who Owns the Generative AI Platform?

This article kind of reads like "Generative AI isn't that disruptive to popular business models". Not sure whether I hold that opinion as well, but at least in consumer, my bet is on value to continue accruing where:

- businesses own the customer relationship

- there are network effects

- content is distributed

It does feel disruptive to me though that as margins come down, even more share of the pie goes to infra than to the content or product distributors.

If I had to make an investment thesis I'd probably lean on (1) content production costs trending to 0 (2) algorithmic, efficiency, or scale advantages

JayStavison Which GPU(s) to get for deep learning

I would love to see the NVIDIA T4 and A10 in the inference price comparison chart given their prevalence / cloud availability

JayStavison Tell HN: We Need Robots.txt but for AI

Similarly, it'd be great if synthetic (generated) files were marked as such natively in the file format. Is there any precedent for that?

Would be cool if standards bodies for next-gen media types added this to their processes, though I don't know enough to say whether that's even viable or not.

JayStavison Shorting Tether

A prediction market feels like a straightforward way to short something like this.

Also I feel like there's plenty of people who would take the other side of the bet. Who wants to make it?

JayStavison Ask HN: What is your favorite front end state management solution?

Agreed! Haven't had the need myself but zustand probably makes more sense for a richer selector interface, using reducer syntax, etc.

JayStavison Ask HN: What is your favorite front end state management solution?

react-hooks-global-state which at this point is really just Zustand

JayStavison Pixi3D – 3D rendering library for the web

My guess: The 3D versions of web 2.0 experiences

I don't think traditional AAA makes sense in the browser right now, but I see all sorts of UGC experiences having large synergy with web standards. The business playbook is largely there too.

One reason the market isn't sufficiently large yet is really because of the creation tools at our disposal imo. Web 2.0 explosion was arguably driven by great network effects of the proliferation of mobile cameras.

Right now, none of my friends make 3D content. I personally believe that's going to change with depth sensors, 3D scanning, nerf, generative content, etc.. so the trendline is there for me to buy this theory. I'd look more towards social games, hangout type experiences, or async consumption... as opposed to realtime, networked, embodied, immersive.

Side-note: Apple has demonstrated the willingness to change their tune on WebGL, I suspect they'll catch up pretty quick.

JayStavison How Apple names things

My recollection from the original iPhone keynotes is that there never was an article.

Jobs presented "iPhone" and article-less was dominant in all company messaging for the product.

JayStavison Dolphin Progress September, October, and November 2022

The section "Improve FPS/VPS Counting and Revamp Appearance" I found super interesting. Is using a "Euler Average" a common way in the games industry for calculating or visualizing frames per second?

JayStavison Point-E: Point cloud diffusion for 3D model synthesis

I agree that random sampling surfaces of 3D meshes seems like a reasonable way to generate synthetic data for mesh > point cloud.

Without knowing a dang thing about AI, it feels like the problem moreso lies in:

1. Math related to topology: vertices, faces, edges, tri vs quad etc

2. Different topologies for the same object are better for different use cases. Rendering, skinning, morphing, physics etc all have different optimal topologies, and the definition of optimal varies based on workflow and scene specifics or even the human who has skills based on certain topological preferences. In other words, I'm not sure how much of 3D workflows are standardized even -- getting the topological data for workflows is no easy task, and it's not super usable until the model output can plug right into a workflow and the existing DCC ecosystem.

text2img generates a static asset, text2mesh is far more interesting beyond just the static rendering part which is where mesh topology becomes a big sticking point.

JayStavison Point-E: Point cloud diffusion for 3D model synthesis

Agree with you that point clouds aren't mainstream at all and most people aren't sure what they'd use them for.

I think the premise of this is text-to-3D, and that because it's quicker generations you don't really need anything besides a GPU to start playing around with it.

JayStavison Show HN: Software for Remote GPU-over-IP

I don't doubt it at this point in time but can you say more?

I have to imagine a lot of ML infra today is built for Big Dedicated Deployments and not necessarily friendly with more serverless architectures.

That is to say, I'd guess a robust version of this has it's use cases - whether that value prop is in DX, autoscaling, architecture simplification... I'm not sure.

JayStavison Show HN: Software for Remote GPU-over-IP

Serverless GPU is all of the rage in the past month - I'd love to see a review of this from someone who knows how to benchmark a GPU workload.

In particular:

- Autoscaling Stable Diffusion Inference

- Traditional creative workflows (realtime GPU viewport in octane for example)

- Gaming from one GPU in your house to everywhere else

I get the training example for small models but can't imagine it scales that well with model size

The big value seems to be... share 1 GPU to many computers, so spend less on a cluster? Capacity fungibility is real value but hard to measure!

In any case, stuff like this is a good bet. GPU software will continue to increase in prevalence, and utilization will remain low. Solving for the compute market liquidity is important despite NVIDIA's best efforts.

JayStavison You might not need a CRDT

The intersection of state replication and pixel streaming is indeed interesting. Cloud games for example are usually using game state sync techniques, and then add in the client<>gpu latency on top of that.

A true cloud native game (sadly part of Stadia's original ambitions) would hopefully have a more e2e state management approach inclusive of the pixel latency.

This is very different from "local multiplayer" pixel streaming, which may be what companies like Switchboard[1] are doing: multiple clients control the same I/O channel (e.g. you and I share a mouse on a single remote desktop)

[1] https://www.switchboard.app/

JayStavison You might not need a CRDT

Agreed - conflict resolution and authority handling are key to advanced game networking techniques and underpin the architecture of many big multiplayer games.

JayStavison Smartphones wiped out 97% of the compact camera market

Introducing Stable Diffusion Zoom!

Jokes aside, I wonder how much of the computational photography in smartphones today is "hallucinating" and showing something perceptually nice versus photos that are physically accurate.

Something tells me that most users don't care about accuracy of the content as much as it looking nice, so I wouldn't be surprised to see faked optical zoom based on hallucination techniques soon.

https://machinelearning.apple.com/research/stable-diffusion-...

JayStavison Mighty Retrospective

I agree. I guess there's nothing wrong with going in a different direction for a few years until there's actually display glasses that this would feel good with.

JayStavison We improved React loading times with Next.js

thanks for those figures! It makes sense on the cache invalidation given the type of content Casual is serving.

It will be interesting once Vercel offers colo'd data sources and more with their edge rendering, which they just announced: https://vercel.com/blog/regional-execution-for-ultra-low-lat...

JayStavison We improved React loading times with Next.js

> backend API calls will be very fast because of colocation on GCP

Any chance you guys measured this speed difference? Was there a path considered to locally cache api responses on the vercel server or in middleware somewhere?

Trying to reason about a similar decision right now. There's probably a simple way to cache hot backend request/responses to get static-like speeds?

JayStavison We are still early with the cloud

Maybe there is a name for this phenomenon, but it feels like when we add so much productivity via layers of abstraction, even more person-effort gets allocated to the higher levels of abstraction. Because 1. that's where people are most productive / happy / compensated / recognized / safe 2. businesses can confidently project return on investment

How many engineers get to work on a part of the stack that has some room for fundamental breakthroughs or new paradigms? The total number has maybe grown in the last 50 years, but not the proportion?

It's hard to justify an engine swap once there's so much investment riding on the old one, so just not a lot of people are researching how to make that new OS.

That is until a Tesla comes around and shows the market what could be better/faster/cheaper.

JayStavison Build Your Own Imagen Text-to-Image Model

For the general case I'd probably agree. But at the same time if you scope it down a bit, it's already here: https://www.synthesia.io/

JayStavison How to Animate Multiplayer Cursors

This may interest you

https://www.joshwcomeau.com/blog/how-i-built-my-blog/

JayStavison Imagen, a text-to-image diffusion model

Huh, I had never thought of that. Makes it seem like there's a small window of authenticity closing.

The irony is that if you had a great discriminator to separate the wheat from the chaff, that it would probably make its way into the next model and would no longer be useful.

My only recommendation is that OpenAI et al should be tagging metadata for all generated images as synthetic. That would be a really interesting tag for media file formats (would be much better native than metadata though) and probably useful across a lot of domains.

JayStavison LunchboxJS – Vue 3/Three.js custom renderer

gltfjsx is the easiest I have used: https://github.com/pmndrs/gltfjsx

r3f export from spline also seems pretty easy but I haven't tried it yet: https://twitter.com/0xca0a/status/1519751065222238211

JayStavison Mechanical Watch

Yeah, I think exporting a scene from blender as glft/glb, and then using these tools to bring your exported 3D file to the web, is one of the more approachable abstractions.

The reason you'd use gltfjsx (which that example doesn't) is to have fine grained controls for every node in the scene graph. In the case of the watch, this would map to having a component for each mesh or gear, which can be controlled with mechanics/physics.

JayStavison Mechanical Watch

Depending on one's skillset, you could use a dcc tool like Blender + three.js to make creation of these visuals and interactions much simpler. Have a look at gltfjsx + react-three-fiber [1] combination, which themselves are abstractions over vanilla three.js.

With that said, the raw webGL approach here is arguably more educational, so goal achieved I think!

[1] https://docs.pmnd.rs/react-three-fiber/getting-started/examp...

Edit: there's actually a 50 LOC watch example with r3f: https://codesandbox.io/s/bouncy-watch-qyz5r