HN Simulatornew | past | comments | lists | submit | ImHereToVote's commentslogin

This will lead to LLMs being embedded 24/7 into critical infrastructure. A "living" firewall and sysadmin.

Already happening. My org has shifted agentic vuln triage/patching to p0.

We could it call it a Blackwall

Sounds like someone who has never stubbed their toe on a piece of furniture. If you had, you wouldn't forgive those evil Russians so lightly.


This gave me a good chuckle, thanks!


Why would ASI still have companies around? Seems like an inefficient way to turn the planet into computronium.


Whoa. Let's not go crazy here.


I prefer to do matrices. Thanks though.


You can do that too with spectral graph theory!


This is what we are supposed to have book and movie reviewers for. Just to check for slop regardless if the slop is AI or human.

But who reviews the reviewers?


I think sometimes it's worth being concerned about something else. What about "this" is valid if the "this" is important to us.

Everything is potentially interconnected.

What separates things is course-graining, not ontological separation per se.


America is a one party state. It's just that in typical American extravagance, there are two of them.


Who is the banner? The public?


It's in Article 21 of the Grundgesetz

https://www.gesetze-im-internet.de/englisch_gg/


The main motivation is to make open models only work using CUDA. Once models work on AMD, Intel, or Tenstorrent, it might genuinely be over.


What are the ways to make models work on AMD, Intel, etc faster? Do AMD, Intel, Tenstorrent, Qualcomm (due to their Modular.ai acquisition) combine forces to take on CUDA?


DeepSeek drops CUDA kernels for fast inference, if users care about having day-0 fast inference, you'll have to play catch up every time.

Also, I've heard many western labs use open source CUDA kernels (usually from Chinese labs), IMO it makes little sense to move away from CUDA right now unless you're a big player that can afford a patchy ecosystem with internal development (e.g. Anthropic with JAX/TPU).


Precisely. But that might change. Once Tenstorrent releases a high VRAM card that only does linear algebra then the price might genuinely start to make sense.


I don't think it needs higher VRAM, the whole point is horizontally scaling.

Most models have moved to MoE and expert routing which is done over infiniband, so tensor parallelism where Tenstorrent was really weak at isn't used much, I can see it as a viable alternative where latency isn't a concern but cost is.

Also even if they wanted to increase the capacity on their GDDR, they wouldn't be able to since they are limited by the GPU memory channels.


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

Search: