HackerTrans
TopNewTrendsCommentsPastAskShowJobs

binyu

745 karmajoined há 6 anos
Cryptography - Distributed Computing - Machine Learning

Submissions

Guy is banned by OpenAI for cyber abuse, his AI appeals, another AI approves it

twitter.com
29 points·by binyu·há 15 horas·6 comments

China's 'Silicon Valley' in Hangzhou: Home of DeepSeek, Alibaba and Ant Group [video]

youtube.com
3 points·by binyu·há 23 horas·0 comments

No Boss, No Money: The Raw Reality of China's Gen-Z Freelancers [video]

youtube.com
2 points·by binyu·ontem·0 comments

Chinese hackers reverse engineer NVIDIA GPU to work on a custom PCB with NVLink

old.reddit.com
5 points·by binyu·anteontem·0 comments

Len Sassaman Forensic Dead Ends at Black Hat – USA – 2002 [video]

youtube.com
1 points·by binyu·há 5 dias·0 comments

GitHub offering to ship CD-ROMs with your source code upon request

twitter.com
6 points·by binyu·há 8 dias·1 comments

Futuristic Weapons, Threats and Security [video]

youtube.com
2 points·by binyu·há 10 dias·0 comments

Guy beats one of the best published quantum circuits for breaking ECDSA with AI

twitter.com
2 points·by binyu·há 12 dias·2 comments

Guy in his basement creates a drug to treat Alzheimer's disease using AI

twitter.com
24 points·by binyu·há 13 dias·30 comments

Anonymous GitHub account mass-dropping undisclosed 0-days

github.com
951 points·by binyu·há 14 dias·386 comments

Extropic is Rethinking Computing

extropic.ai
3 points·by binyu·há 15 dias·0 comments

Anthropic's philosopher answers your questions [video]

youtube.com
4 points·by binyu·há 15 dias·1 comments

Len Sassaman – Satoshi Nakamoto Evidence

pmarengo.medium.com
3 points·by binyu·há 16 dias·0 comments

NASA's Cold Atom Lab is creating one of the weirdest forms of matter in space

sciencedaily.com
3 points·by binyu·há 17 dias·0 comments

The notebooks Marie Curie filled with her research are still dangerous to touch

spacedaily.com
14 points·by binyu·há 20 dias·6 comments

Diffusion‑based LLMs generate many parallel tokens rather than one‑by‑one

inceptionlabs.ai
1 points·by binyu·há 21 dias·0 comments

Algebraic Attack

en.wikipedia.org
2 points·by binyu·há 22 dias·0 comments

Coding Theory

en.wikipedia.org
5 points·by binyu·há 22 dias·0 comments

Deep learning discovers Antarctic earthquakes coming from an unlikely location

phys.org
2 points·by binyu·há 24 dias·0 comments

Cryptography's Mathematical 'Worlds': Which One Do We Live In? [video]

youtube.com
2 points·by binyu·há 24 dias·0 comments

comments

binyu
·há 15 horas·discuss
Well, if anything it is good that trivial stuff like resolving being flagged automatically by a bot is also done by a bot.
binyu
·há 10 dias·discuss
I think that Vitalik is collapsing a lot of dense math and cryptography onto a more understable language aimed at the blockchain developers/community.

In a sense, Vitalik is "recruiting" with this post, his goal being lower the barrier of entry to this discipline.
binyu
·há 12 dias·discuss
so Sophon is less efficient than Cerebras?

Edit: is that Joule vs micro-Joule? I need better glasses

> Cerebras is 12.8J, Sophon is 25.8mJ

Are your figures hypothetical or do you have a working prototype?
binyu
·há 12 dias·discuss
What makes you think his reply was AI generated?

Edit: I can see a bunch of hints, most definitely. Still a good comment though.
binyu
·há 12 dias·discuss
> they scale out in 2D, we scale up in 3D.

This actually helps a lot, thanks.

> Instead of spreading SRAM across a wafer, we stack DRAM on top of the logic

Is this done with current manufacturing technologies? Does it require a special process?

> no streaming, no off-chip memory at all. ~1 kW, not 23 kW

Is this for an individual compute unit? Compared to Cerebras, what's the ratio of power used vs compute output?
binyu
·há 12 dias·discuss
Hello, kudos for the tremendous work. Could you explain the difference between your design and Cerebras?

Bests
binyu
·há 12 dias·discuss
This is particularly relevant because the guy used Andrej Karphaty's "autoresearch" [1] approach, which is extremely powerful when applied to problems that generalize well to it.

[1] https://github.com/karpathy/autoresearch
binyu
·há 13 dias·discuss
Can you provide an example or a reproducible way to trigger it?
binyu
·há 13 dias·discuss
That behavior was only observed with the short lived Fable preview, Opus and other models do not seem to pose such limitation.
binyu
·há 20 dias·discuss
> This would be a great topic but this specific article isn't worth your time.

Why? Do you think it reports inaccurate facts? Personally, I found it interesting and easy to read.
binyu
·há 20 dias·discuss
Personally, I've tried to squeeze more tok/s for a single DGX Spark deployment and DeepSeek V4 Flash but only got marginal improvements. There's work to do on fusing kernels and other optimizations that are already on antirez's roadmap so it is not worth duplicating efforts.

I've had positive experiences running GLM 4.7 via vLLM, tool calling works well and the inference is fast. Do you run DeepSeek V4 Flash on vLLM?
binyu
·há 20 dias·discuss
> 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
binyu
·há 28 dias·discuss
> 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?
binyu
·há 28 dias·discuss
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.
binyu
·há 28 dias·discuss
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-pi

Why not use the leaked claude code source? Not that you really need it to execute the jailbreak
binyu
·há 28 dias·discuss
> 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 to characterize this as harmless.

> These vulnerabilities all appear relatively simple, and we have found that other publicly-available models are able to discover them as well without requiring a bypass.

Ah so now they are admitting that this is all about hype after all.
binyu
·há 28 dias·discuss
> 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 it for their convenience
binyu
·mês passado·discuss
> 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 4.8
binyu
·mês passado·discuss
Hey guys,

check out this technique https://github.com/0xSufi/fable-jailbreak/

It works with security audits and other workflows that are currently blocked.
binyu
·mês passado·discuss
A tool to manage Claude Code conversations based on my typical workflow which integrates with my desktop OS and terminal app.