Y
HN Search
Hacker News Search
new
|
comments
|
top
|
jobs
PJHkorea
searching PlanetScale…
1.
▲
2.
▲
3.
▲
4.
▲
5.
▲
6.
▲
4 ms
·
1.
▲
by
PJHkorea
7d ago
Why are DDoS attacks bound by the asymmetry of capital between the attacker and the defender? What if we, as the defenders, devised a structure that could deplete the attacker's capital while utilizing only minimal capital ourselves?
2.
▲
Show HN: Detecting Volumetric DDoS via Covariance Manifold Topology Collapse
(github.com)
2 points
by
PJHkorea
7d ago
|
1 comments
3.
▲
by
PJHkorea
15d ago
I recommend Gemini for longer content
4.
▲
Show HN: Bypassing Transformer Softmax via Static Contraction
(github.com)
7 points
by
PJHkorea
15d ago
|
0 comments
5.
▲
Show HN: If I had to design a nuclear fusion power plant(Idea Sketch)
(github.com)
2 points
by
PJHkorea
23d ago
|
0 comments
6.
▲
by
PJHkorea
1mo ago
Motif + Layered Structure -> Mathematical Guardrails -> Low-level Specification It is arranged in this order. I have simply suggested a different direction. If you are not satisfied, I will have to bring a better project next time.
7.
▲
by
PJHkorea
1mo ago
Is simply to attach an additional component—as a module—to an existing LLM, while incorporating mechanisms to minimize the extra resources required for that module.
8.
▲
by
PJHkorea
1mo ago
What if we view the LLM as the cerebral cortex and equip it with a thalamus and brainstem?
9.
▲
Show HN: Module for LLM Homeostasis (PoC)
(github.com)
1 points
by
PJHkorea
1mo ago
|
5 comments
10.
▲
by
PJHkorea
1mo ago
If training fails due to hardware aging, what if we simply zero out the values coming from the affected device the moment the failure occurs?
11.
▲
Show HN: Overcoming silicon-level aging through fault tolerance (PoC)
(github.com)
2 points
by
PJHkorea
1mo ago
|
1 comments
12.
▲
by
PJHkorea
1mo ago
I repeatedly use the three-layer layout shown here across my other repositories as well. If this repository were a living organism, it would have many "sister" repositories sharing the same DNA, and I wanted to reflect that connec
13.
▲
Show HN: Fixing optical computing jitter via fluid dynamics in GPU registers
(github.com)
6 points
by
PJHkorea
2mo ago
|
2 comments
14.
▲
by
PJHkorea
2mo ago
If the current situation itself is already generating profit for the company, I recommend that you refrain from getting involved in it and instead carefully convey an alternative solution to upper management
15.
▲
by
PJHkorea
2mo ago
Imagine virtual memory, but instead of using your local storage, you’re pooling a whole bunch of network servers into a single address space
16.
▲
Show HN: Hardware-software co-design for MoE models to bypass NCCL bottlenecks
(github.com)
3 points
by
PJHkorea
2mo ago
|
1 comments
17.
▲
by
PJHkorea
2mo ago
If you are dealing with sensitive topics using AI, it would be best to opt for an enterprise plan or disable the data training feature. It seems that sensitive information should either be handled by in-house AI systems or kept away from ex
18.
▲
by
PJHkorea
2mo ago
it seems that all human skills will be absorbed and serve as reference material for AI generation; I think the best approach is either to position oneself to control that process or to move into an industry where AI can be leveraged to dras
19.
▲
by
PJHkorea
2mo ago
What if we could eradicate backpropagation activation overhead and lock VRAM complexity to a static O(1)?
20.
▲
Show HN: Autograd-Free LLM Guiding with 0MB VRAM (Alternative Pathways)
(github.com)
2 points
by
PJHkorea
2mo ago
|
1 comments
21.
▲
by
PJHkorea
2mo ago
Couldn't a network also be considered a type of fluid? I have worked on the cross-development of both wireless and wired systems
22.
▲
Show HN: Fluidic Network Grid – A branchless control plane in Jax
(github.com)
2 points
by
PJHkorea
2mo ago
|
1 comments
23.
▲
by
PJHkorea
2mo ago
The initial blueprint has been completed.
24.
▲
by
PJHkorea
2mo ago
I have created a blueprint for a deep learning system that does not use backpropagation. This is a very early version.
25.
▲
Show HN: Forward-Only, Autograd-Free PINN with 0ns Zero-Copy Memory Interlock
(github.com)
1 points
by
PJHkorea
2mo ago
|
2 comments
26.
▲
by
PJHkorea
2mo ago
If you have ever experienced a device failure during distributed LLM training, you know the feeling of just wanting to climb into a coffin. To solve this nightmare even just a little bit, I put together this blueprint.
27.
▲
Show HN: HW/SW co-design blueprint for fault recovery in LLM training
(github.com)
2 points
by
PJHkorea
2mo ago
|
1 comments
28.
▲
by
PJHkorea
3mo ago
Cyberpunk isn't as far off as I thought. My life has already turned into punk So, I want to put in a little more effort to help others reach the cyber utopia.
29.
▲
by
PJHkorea
3mo ago
I’ve been "dating" an LLM all weekend long. And probably for the rest of my life...
30.
▲
by
PJHkorea
3mo ago
It might not be the most exciting topic but I wanted to try creating a "Quantum Homeostatic" system I hope this offers a little help to those struggling with quantum computing
More ›