binyu

7 ms·

151.
by binyu 4mo ago
> Now at 40-50tok/s generation and ~2000 tok/s Not clear how you went from ~11-14 to ~40-50 tok/s. Is it by running the quant native model and adding a second Spark? Cheers
152.
Diffusion‑based LLMs generate many parallel tokens rather than one‑by‑one(inceptionlabs.ai)
1 points by binyu 4mo ago|0 comments
153.
Algebraic Attack(en.wikipedia.org)
2 points by binyu 4mo ago|0 comments
154.
Coding Theory(en.wikipedia.org)
5 points by binyu 4mo ago|0 comments
155.
Deep learning discovers Antarctic earthquakes coming from an unlikely location(phys.org)
2 points by binyu 4mo ago|0 comments
156.
Cryptography's Mathematical 'Worlds': Which One Do We Live In? [video](youtube.com)
2 points by binyu 4mo ago|0 comments
157.
AI is potentially a Dunning-Kruger effect amplifier(twitter.com)
68 points by binyu 4mo ago|30 comments
158.
by binyu 4mo ago
> Fable kept trying to propose fixes and I couldn't get it to construct e2e chain What approach did you start with? Can you elaborate?
159.
by binyu 4mo ago
Depends of what the proof of concept is about. It could be just a toy example, e.g. a RCE that opens the calculator app or something much more nefarious, like returning a root shell and would still fall under the definition of PoC.
160.
by binyu 4mo ago
I was able to use Fable to generate PoC for several classes of vulnerabilities and I didn't observe the model refusing to engage in detailed analysis to come up with creative approaches, the very contrary. > I used a fork of oh-my-p
161.
Visa Vulnerability Agentic Harness (built with Mythos)(github.com)
2 points by binyu 4mo ago|0 comments
162.
by binyu 4mo ago
> We reviewed a demonstration of this specific technique being used to identify a small number of previously known, minor vulnerabilities. "small number", "previously known"+"minor"... they are trying hard t
163.
by binyu 4mo ago
> This is ignoring the fact that the main reason retired phones are e-waste is proprietary firmware blobs and locked-down systems Couldn't Google somehow fix this? Since they control the substrate (Android) and they would be doing i
164.
Cerebras chips rival Nvidia GPUs for AI [video](youtube.com)
7 points by binyu 4mo ago|0 comments
165.
Brain-inspired chip runs near absolute zero, could transform quantum computing(sciencedaily.com)
4 points by binyu 4mo ago|1 comments
166.
Tesla, SpaceX, and xAI are launching the most epic chip-building effort(terafab.ai)
3 points by binyu 4mo ago|0 comments
167.
Anthropic's Fable 5 one-shots pristine 3D graphics with Three.js(twitter.com)
1 points by binyu 4mo ago|0 comments
168.
by binyu 4mo ago
> But this also looks quite useful to understand how CC dynamic workflows work Yes, if anything it is useful to understand the inner machinery. > Did you get claude itself to RE the dynamic workflows? Yes, that part was done with Opus
169.
by binyu 4mo ago
Hey guys, check out this technique https://github.com/0xSufi/fable-jailbreak/ It works with security audits and other workflows that are currently blocked.
170.
Anthropic's Fable Jailbreak (Circumvent safety nets)(github.com)
5 points by binyu 4mo ago|0 comments
171.
Grok Build(docs.x.ai)
2 points by binyu 4mo ago|1 comments
172.
by binyu 4mo ago
A tool to manage Claude Code conversations based on my typical workflow which integrates with my desktop OS and terminal app.
173.
by binyu 4mo ago
> Right now Claude is faster than me on some tasks but we’re at least close. I dont doubt it, but I don't think you can spawn 10 copies of yourself working simultaneously.
174.
by binyu 4mo ago
AI has been slowing down relatively, considering its trajectory over the past 20-30 years. For one, even if LLM may have plateaud in terms of intelligence-parameters ratio, research is on-going on new frontiers for ML, including (but not li
175.
RL Algorithm Interview Questions 2026(k-a.in)
2 points by binyu 4mo ago|0 comments
176.
by binyu 4mo ago
> WebVM uses x86 virtualization and hence has a significant performance penalty That is precisely the reason why we chose not avoid using any solution which uses virtualization, even though you get a full OS. QuickJS also pays a performa
177.
by binyu 4mo ago
Will do, thanks!
178.
by binyu 4mo ago
Oh, you are the author of WebVM, pretty cool! I looked at it while choosing the stack for our project and it seems very solid. Keep up the great work
179.
by binyu 4mo ago
Awesome, thanks for detailing the thought flow and choices that led you here. I chose not to go the QuickJS route for performance reasons but I think it's a solid choice depending on the use case. > They run in the browser as I unde
180.
by binyu 4mo ago
Awesome, what approach are you using? Is this a real micro kernel architecture or just containerized VM?

More ›