Hacker Newsnew | past | comments | ask | show | jobs | submit | xpl's commentslogin

Is this a joke? If it's not trainable / differentiable when why do it in the first place? It's just as inefficient and inflexible as it gets compared to tool calling — you have to statically bake programs in the weights, model cannot introspect it and modify, it has very limited IO capabilities, bad performance, bad everything. Its like a weird brainfuck-esque VM — cool that you can do it, but for what except some lulz?

But maybe it's just too genius and I don't understand it.


I'd tend to agree, the only good points I've seen were made by @hedgehog [1] here in this thread:

    I'm not sure about the rest but a significant problem with high frequency tool calling (especially in training) is that it breaks batching.
and then later by @ACCount37 [2]:

    I'm less interested in turning programs into transformers and more interested in turning programs into subnetworks within large language models.
In theory, if you can create a very efficient sub-net to replicate certain tool calls (even if the weights are frozen during any training steps, and manually compiled), this might help with making inference much more efficient at scale. No idea why in general you would want to do this through the clunky transformer architecture though. Just implement a non-trainable, GPU-accelerated layer to do the compute and avoid the tool-call.

[1] https://news.ycombinator.com/item?id=47367986

[2] https://news.ycombinator.com/item?id=47363909


We at Avride are hiring in Austin for a full-time, onsite position.

In this role, you'll help build state-of-the-art 3D data annotation tools that advance autonomous driving.

We're looking for a seasoned UI engineer with a strong React background and relevant experience building highly stateful, low-latency interactive applications. Experience with 3D graphics is a plus.

Apply here: https://job-boards.greenhouse.io/avride/jobs/4012877009


Hi, I'm Kibru senior React engineer with strong experience building complex, highly stateful UIs. Is there any flexibility on remote for this role?

Unfortunately, no. We allow hybrid work, but only if you're willing to relocate immediately (U.S. only).

> Tesla is famously anything but a car company because their cars are mediocre in every way except the battery range

I can't say that I'm a big fan of this guy... But I can tell you this: I learned to drive only after moving to the U.S. recently, and when I had to choose my first car, I found Tesla to be the best among many I tried. It's just awesome, and I don't even use their FSD, the car itself is superb (at least the latest "3"). Minimalistic, no BS, drives well, quiet, comfortable. The same feeling I had with the first iPhone, compared to other phones.


sqlite with extensions, scales to millions of docs easily


can you tell me more


A similar project of mine (generated back when GPT-4 was out, along with comment threads and articles):

https://crackernews.github.io/


> What are some examples?

"Weaponizing Wikipedia Against Israel" https://aish.com/weaponizing-wikipedia-against-israel/


Using that site as evidence is about as credible as linking to a Ben Shapiro vlog


Why not use actual Ethereum as a base layer? If you want speed, build (or use) an L2 on top of it.

I can hardly see any value in "yet another private blockchain" — just use a database, duh.


Greed. There's no technical basis. I am pretty sure they considered it and even though L2s do not pay much rent to L1, Stripe wanted absolute control (which goes against decentralization) and all the fees in their pocket. Just hoping this doesn't become another trend.


ethereum is expensive, and there's scaling limits, see all the posts coinbase puts out for their base chain.


Those scaling limits are temporary though. PeerDAS in the next hardfork should increase scalability even more on the rollups.

Most of these L1s will likely end up becoming L2s in the near future, especially if they can rake in revenue via sequencers


Probably 50 years ago, some skeptics were saying, "Why database, just use a pen and a notepad, duh".


exactly my thought. i have next to zero interest in yet another l1


They say:

In the API, all GPT‑5 models can accept a maximum of 272,000 input tokens and emit a maximum of 128,000 reasoning & output tokens, for a total context length of 400,000 tokens.

So it's only 270k for input and 400k in total considering reasoning & output tokens.


Only at first glance. It can easily render things that would be very hard to implement in an FPS engine.

What AI can dream up in milliseconds could take hundreds of human hours to encode using traditional tech (meshes, shaders, ray tracing, animation, logic scripts, etc.), and it still wouldn't look as natural and smooth as AI renderings — I refer to the latest developments in video synthesis like Google's Veo 3. Imagine it as a game engine running in real time.


Why do you think this is so hard, even for technical people here, to make the inductive leap on this one? Is it that close to magic? The AI is rendering pillars and also determining collision detection on it. As in, no one went in there and selected a bunch of pillars and marked it as a barrier. That means in the long run, I'll be able to take some video or pictures of the real world and have it be game level.


Because that's been a thing for years already - and works way better then this research does.

Unreal engine 5 has been demoing these features for a while now, I heard about it early 2020 iirc, but the techniques like gaussian splattering predate it.

I have no experience in either of these, but I believe MegaScans and RealityCapture are two examples doing this. And the last nanite demo touched on it, too.


I'm sorry, what's a thing? Unreal engine 5 does those things with machine learning? Imagine someone shows me Claude generating a full React app, and I say "well you see, React apps have always been a thing". The thing we're talking about is AI, nothing else. There is no other thing is the whole point of the AI hype.


What they meant is that 3D scanning real places, and translating them into 3D worlds with collision already exists, and provides much, much better results than the AI videos here. Additionally, it does not need what is likely hours of random footage wandering in the space, just a few minutes of scans.


This thing looks like a complete joke next to the Unreal Engine.


Check out this Russian Telegram channel on AP https://t.me/usadminvisaprocessing/1

There are people who file collective mandamus lawsuits, so you might consider joining one of those groups.

However, from what I've seen, it could be a waste of money, as there isn't convincing evidence that it accelerates AP. These lawsuits usually take many months, and AP often resolves "by itself" before the lawsuit reaches a resolution.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: