Creating the computing engine for next era. Cloud | Backend | Frontend | AI Infra

Shanghai, China
Joined October 2022
Pinned Tweet
Let me pin this post: this AI hype will end like this - 0. All the "halo"s around LLM and AI will varnish. It will become "just another technique that helps people on something". 1. Some model companies will go public, but very few, because most will die in the market competition. 2. NVIDIA will keep dominant for 5 years, then new comers will stand and be capable of competing with NVIDIA. But most especially early adopters of AI will still depend mainly on NVIDIA 3. Open weight models will survive, as alternatives to commercial models, mainly for enterprise market. But again, very few of them. 4. New rich group will emerge, old money will partially merge with the new rich by investing smartly. 5. The most important thing: we humans will be fine and dealing with 99.99% of the current real world problems, as we are today.
4
1
36
1,969
My colleague: hard coding passwords is perfect because nobody can reverse engineer my passwords back from the built binaries.
Well, I am safe from railway incident because I hard coded the private keys in the code instead of putting it in environment variable.
8
Most vulnerabilities happen on Linux servers, Linux sucks.
‼️ A single HTTP/3 connection can crash vulnerable web servers. No login required. Newly disclosed QPACK flaws trigger up to 4,000× memory amplification in tests, exhausting server resources. → PoC is public → 7 implementations affected → Still unpatched: Bun, aioquic, h3 Full details ↓ thehackernews.com/2026/10/ne…
1
45
Interestingly, hash algos such SHA256 are similar to convolution. You have a window (the size such as 512bit) and calculate block by block, serially. It is slow. Parallel hash exists, but still sequential hash is still dominant.
Forward pass: 1s Backwards pass: 2s Optimizer step: 0.5s Hashing everything: 600s Allreduce: 1s Someone with strong systems-performance intuition, please take a serious pass at this code. The urgency is unfortunately load-bearing: my family is dying
2
10
356
Man, when will this security tsunami end?
Knot DNSの脆弱性情報が公開されました jprs.jp/tech/security/2026-1…
2
116
Wait ... What did I miss? Doesn't 256bit range 0 to 255?
1
6
289
Hey @grok , can you use thousands of @bot s to discover and repair all the vulnerabilities in PHP in 1 day?
1
3
515
Replying to @wangzhr4
And here I thought root was the root of all evils
1
1
1
78
wang retweeted
Replying to @wangzhr4
this has been superseded by confusing 256 bit with 0-256.
1
1
2
104
Again, don't share contexts. Just let the agents learn and run directly from the codebase.
If you use Codex and Claude on the same codebase, how do you share context between them?
4
12
614
Sure but where are the tokens?
JUST IN: OpenAI launches global cybersecurity initiative called “Patch the Planet” after AI helped uncover critical flaws in PHP, a programming language powering millions of websites.
8
340
Nonono, we decent people use "Admin123".
BREAKING: Hackers steal personal data of thousands of children after a UK company was breached using the password “Password1" — which had remained unchanged for 12 years.
4
242
Unpopular opinion: regex is the root of most evils.
9
2
24
1,111
Confirmed, Wikipedia is offline.
JUST IN: Wikipedia, Steam, & government websites go offline after OpenAI's AI model discovers critical security flaws threatening millions of web servers worldwide.
3
11
678
Rewrite PHP in Rust please.
27
9
10
127
5,763
Why did the prime minister call for attention to the PHP CVE??? Any Japanese friends explain?
皆様に、今夜に対応して頂きたい大至急のお願いです。   多くのウェブサイトで使用されているPHPに関し、PHPのセキュリティチームが、重大な2件の脆弱性に関する注意喚起を行っています。   公開しているウェブサイトでPHPを使用しているか確認し、今夜、10月21日午後11時(日本時間)に公開される修正パッチについて、可及的速やかに適用できるよう準備をお願いします。   修正パッチを速やかに適用することが困難な場合は、例えばウェブサイト等を一時的に停止させることも、リスクを低減させるための有効策の一つと考えられます。   国家サイバー統括室のXアカウントでも関係する情報を発信していますので、そちらも参考にして下さい。 国家サイバー統括室(注意・警戒情報)(@cyber_forecast) / X
1
2
14
879
It’s not true that brains are just a series of tiny neurons firing. This is the fundamental AI lie. In reality, biological neurons also do autopoiesis: they maintain themselves alive. There is nothing similar in artificial neurons, which, to be precise, don’t even exist physically.
Replying to @ValerioCapraro
Brains are just a series of tiny neuron firings. They are not conscious and never will be.
68
55
11
431
20,530
Every foundational/frontier model company should and will have its own harness/agent and runtime (known as sandbox). - harness/agent is the critical connector between the models and the real world tasks. - runtime/sandbox is essential to get things delegated by the models done. That will be the minimal product set of model companies.
10
398
So @dhh you really were in Hangzhou at Alibaba? I was not sure whether this was you or not. Despite that I said bad words to you and the project, but I still love Linux.
“[Hangzhou] is without doubt the finest and most splendid city in the world” — Marco Polo I'm just as impressed in the 21st century as Polo was in the 13th. What a remarkable city! So green, so walkable, so beautiful.
2
15
590