@OpenResty profile picture
OpenResty @OpenResty

OpenResty Web Platform Official

Joined August 2013
  • Tweets 772
  • Following 20
  • Followers 1,622
135 Photos and videos
Saining Xie (who left Meta to start a new venture with Yann LeCun) made a brilliant point in a recent interview: The current LLM approach of brute-forcing massive human text data during training actually violates the core principle of "The Bitter Lesson"—which advocates minimizing the use of human prior knowledge. I'm really looking forward to their "World Models" learning autonomously through environmental interaction and achieving true understanding and reasoning at an abstract conceptual level. This echoes the AlphaZero philosophy: zero human game records, just basic rules and self-play. While top AI labs are all-in on scaling Transformers, we absolutely need pioneers exploring alternative paths and possibilities!
1
2
105
Anthropic’s Mythos model is finally here! t.cn/AXXeUzNC Enjoy the cheap access while you can—it ends on June 23. Just 2 weeks left! After that, it switches to usage credits and will get 10x-100x more expensive. Anthropic is shipping insanely fast ahead of their IPO. Gotta respect the hustle, haha! I’ll definitely be looking to grab some of their stock for the long hold. I used to call Claude Opus 4.8 "stupid" multiple times a day, but using Claude Fable 5 today? Not a single complaint! 😂 Sadly, I probably won't be able to afford it in 14 days, so I'm cherishing every prompt. Now we wait and see when open-source models like DeepSeek can catch up to Mythos!
1
166
My wife has been struggling with sleep lately, so I taught her a 2,500-year-old "protocol" from the earliest Buddhist sutras. 1/ Physical Decompression. Relax every muscle. I personally use a Rapid Release tool to speed this up. The goal is to remove all physical "noise" from the system. 2/ Non-judgmental Observation. Observe everything—breath, heart rate, thoughts, external sounds—without labeling or judging. Don't feed your thoughts energy; let them "starve" and dissolve naturally. It’s like background processes timing out. 3/ Metta (Loving-kindness) Expansion. Fill your mind with compassion and joy. Let these emotions radiate outward until they encompass everything. It’s mental training. Don’t over-optimize or force it—"overdoing is as bad as underdoing." I’ve verified this with my Apple Watch; it consistently leads to deeper Dhyana (meditative absorption) and high-quality deep sleep. You don’t need to sit cross-legged. Do it while lying down or walking. It’s about the state of mind, not the position of the body.
96
From a neuroscience perspective, Zen meditation is essentially training to quiet the left brain and activate the right. The left brain governs logic, concepts, and ego boundaries—the very roots of our suffering. In contrast, the right brain transcends these boundaries, tapping into a state of infinite bliss known as "Zen Joy." Ultimately, both subside into a state of "Cessation"—a profound stillness that surpasses even deep sleep. Our education and careers focus almost entirely on sharpening the left brain, often at the expense of the right. Spiritual practice is the act of reversing this trend. "Ego-clinging" is the hardest knot to untie, yet it is the source of all mental anguish. Physical pain is manageable; it is the mind’s suffering that truly hurts. In my own experience: - 1st Jhana: Conceptual thought still lingers. - 2nd Jhana & beyond: Conceptual thinking and worldly worries vanish. The left brain falls silent, and the right brain takes over. The boundaries of the "self" begin to blur until the consciousness merges with the universe. When happiness and peace become a choice, independent of external conditions—that is what Buddhism calls "Liberation."
95
🚀 OpenResty 1.31.1.1 is officially out! This is a major feature release packed with powerful new capabilities, significant security improvements, and broad ecosystem upgrades. Here are the key highlights: 1️⃣ Nginx Core & OpenSSL Upgrade: Upgraded from Nginx 1.29.2 to 1.31.1 and OpenSSL from 3.5.5 to 3.5.6, delivering the latest performance improvements, protocol support, and security hardening at the foundation level. 2️⃣ lua-nginx-module v0.10.31: A landmark update introducing the new `precontent_by_lua` and `proxy_ssl_verify_by_lua*` directives, APIs for retrieving server random and master secrets, `tcpsock:getsslsession`, custom trusted CA store support for cosocket TLS handshakes, and new TCP socket options `keepintvl`/`keepcnt`. Several critical crash fixes around QUIC connection teardown, worker shutdown, and SSL session caching are also included. 3️⃣ stream-lua-nginx-module v0.0.19, lua-resty-core v0.1.34rc2 & LuaJIT v2.1-20260415: New `proxy_ssl_certificate_by_lua` directive, `serversslhandshake` on downstream sockets, `tcpsock:settrustedstore()` for per-handshake trusted CA configuration, and a suite of new SSL FFI APIs. LuaJIT introduces `ffi.abi("dualnum")` with numerous fixes for ARM64, s390x, FFI, and the JIT compiler, significantly boosting cross-platform stability and runtime performance. 4️⃣ Ecosystem Component Upgrades: `lua-resty-mysql` adds ed25519 authentication support, `lua-resty-string` gains AES-256-CTR bindings, `lua-cjson` adds comment decoding and indented encoding options, `xss-nginx-module` gains dynamic module build support, and more — continuing to strengthen the OpenResty ecosystem. We strongly recommend all users upgrade to take advantage of the latest features, security improvements, and bug fixes.
1
4
400
📄 Full Release Announcement: openresty.org/en/ann-1031001… 📥 Download Page: openresty.org/en/download.ht… 📦 Binary Package Repository Installation: openresty.org/en/linux-packa…
110
In the AI era, I feel like the only 3 things that actually matter for regular people come down to one stack: Health, Cash, and Knowledge. Things nobody can take from you. 1️⃣ Hit the gym & stay healthy. Live long enough for AI to cure all diseases and slow down aging. 2️⃣ Accumulate cash via markets (NO real estate). Hold cash, and long-term HODL the big winners: AI giants, compute infrastructure, and energy monopolies feeding the beast. 3️⃣ Leverage AI to level up your knowledge. Learn and upskill constantly. Stop grinding at dead-end, low-value jobs. The harder you grind there, the faster AI replaces you. You just lose your health and stay broke. It’s a tragedy. Look at the data: LLM intelligence and hyper-scaler compute/token consumption are still growing exponentially. AI giants are like black holes—sucking up all the world's energy, compute, capital, knowledge, and top talent, all to replace human labor. Massive tech layoffs are a feature, not a bug, while AI capex skyrockets. Even $QQQ is basically turning into a winner-take-all ETF for a handful of mega-caps. Wealth concentration is getting insane. The social ladder is breaking. Even Ivy League grads are struggling to lock down internships or entry-level roles. We are literally seeing parents pay employers $50k out of pocket just to hire their kids. It’s becoming a business. Silicon Valley is moving so fast it’s tripping over itself. Anxiety is palpable. The old playbook is dead (Disclaimer: Not financial advice, DYOR).
3
203
Just burned through a massive amount of tokens on a real-world experiment: let a top-tier AI coding agent fully auto-dev a complex project. Zero human intervention. The result? 💀 Problems kept compounding until it was impossible to converge. The AI made a few flawed design decisions early on, leading it further down a rabbit hole. We had to scrap the entire thing and start over. When we brought in human experts to supervise? Done in no time. This proves "human taste" isn't just about pretty code style, it's about critical design decisions. Here’s the kicker: the AI did write a detailed design doc at the start. But for complex projects, you can’t perfectly architect everything upfront. Even human experts can’t. In software dev, seemingly tiny choices during coding have massive butterfly effects. AI messes these up constantly, even the best models/agents. It's software engineering 101, and it’s more critical now than ever.
1
1
161
Heard a rumor that Meta is basically banning engineers from writing code by hand. Word is, if less than 95% of your code is AI-generated, your perf review is getting red-flagged. Wild if true.
4
2
46
9,034
OpenResty 1.29.2.5 is now officially released! Key update: backported the official Nginx security patch to fix a buffer overflow vulnerability in `ngx_http_rewrite_module` (CVE-2026-9256), further strengthening security protection. Related links in the comments. #OpenResty #WebServer #Nginx
1
2
3
602
Full announcement for this OpenResty release: openresty.org/en/changelog-1… Download: openresty.org/en/download.ht… Binary package repository installation: openresty.org/en/linux-packa…
89
Even the smartest LLMs constantly blow me away with how stupid they can be... peak "jagged technological frontier." Also, "catastrophic forgetting" is getting real old, real fast.
1
122
Just noticed Anthropic’s new "Fast Mode" for Opus. It boosts response times by 2.5x with zero quality loss, but the token price skyrocketed by 6x! I guess some people really have money to burn and are willing to pay a massive premium just for speed. AI profits are definitely flowing back to the big tech giants now, thanks to all those agents burning through tokens 24/7...
140
🚀 OpenResty 1.29.2.4 is officially out! This is a critical security-focused release designed to address recent vulnerabilities and harden your infrastructure. Here are the key updates in this version: 1️⃣ Upstream Nginx Security Patches: Fully synced with the latest Nginx security fixes. This includes a critical fix for a buffer overflow vulnerability in ngx_http_rewrite_module (CVE-2026-42945), alongside multiple out-of-bounds read fixes in the charset, SCGI, and uwsgi modules. 2️⃣ HTTP/3 & OCSP Security Hardening: Resolved a key HTTP/3 address spoofing vulnerability (CVE-2026-40460) and a use-after-free risk within the OCSP resolver (CVE-2026-40701), significantly boosting system robustness under complex network protocols. 3️⃣ Dedicated Security Maintenance: This release is a rapid response to recently disclosed core Nginx vulnerabilities. We strongly recommend that all users on the 1.29.2.x series upgrade immediately to mitigate potential exploit risks. 🔗 Relevant links can be found in the comments below. #OpenResty #WebServer #Nginx
1
1
180
📄 Full Release Announcement: openresty.org/en/ann-1029002… 📥 Download Page:openresty.org/en/download.ht… 📦 Binary Package Repository Installation: openresty.org/en/linux-packa…
67
That trending AI Time Horizon chart is insane—it actually gives us an objective, unified benchmark to evaluate all LLMs, even the oldest ones. AI capability is still growing exponentially over time. Have to admit, Anthropic’s Claude 4.6 Opus and Mythos models are seriously impressive... Too bad it doesn't include data points for the latest Opus 4.7 and GPT-5.5 models yet. Looks like late 2024 was the exact moment LLM capabilities really started taking off. 🚀
1
108
Source: metr.org/time-horizons/

Task-Completion Time Horizons of Frontier AI Models

Our most up-to-date measurements of the time horizons for public frontier language models.

metr.org
36
Wild to see Musk’s SpaceX/xAI providing compute to Anthropic. Guess Grok isn’t using all that capacity? Anthropic just doubled their subscription token limits—lol! They’re even planning "space-based compute" together...OpenAI is in a tough spot now that the world’s richest man is backing their biggest rival. All this while Musk’s lawsuit against them is still ongoing. The irony of him being an early investor is just wild.
1
137
Meta is tracking US employees' mouse movements, clicks, and keystrokes — to train AI that will replace them. The ones who haven't been laid off yet aren't safe. They're just still useful. The golden age of Big Tech jobs is likely over. Employees are now raw material. There's no going back.
165
Meta and Microsoft have recently kicked off another round of massive layoffs. This is exactly what Ilya called it — if these tech giants don't slash headcount, where else are they going to find the money to cover the hundreds of billions of dollars in AI infrastructure spending? Not long ago, Oracle did the same thing, laying off workers specifically to free up cash for GPU compute. The message couldn't be clearer: these companies would rather pour money into chips and compute than pay their own employees — and this is happening even as their profits and revenues continue to grow at a rapid clip. Anyone still pushing the line that "they just over-hired during the pandemic" really needs to give it a rest.
1
169
Load more