Yaogang Lian (@yliansoft) header
Yaogang Lian (@yliansoft) profile picture

Indie developer building Quiver, Dino (https://teapawt.net/t.co/rxO1PTIHDu) and other apps. Beautifully crafted. Love is in the details.

Joined November 2009
Photo by Yaogang Lian (@yliansoft)Photo by Yaogang Lian (@yliansoft)Photo by Yaogang Lian (@yliansoft)Photo by Yaogang Lian (@yliansoft)Photo by Yaogang Lian (@yliansoft)Photo by Yaogang Lian (@yliansoft)Photo by Yaogang Lian (@yliansoft)Photo by Yaogang Lian (@yliansoft)Photo by Yaogang Lian (@yliansoft)Photo by Yaogang Lian (@yliansoft)
Quiver 4 is coming soon. Years of my life spent on this update. 😀 - Cross-platform support: macOS, Windows, Linux, iOS, Android - full featured on every platform - Lightweight, native Swift apps, no Electron, no RN - Built-in sync: cloud folders, git sync, webdav - Over 30 beautiful themes (including Dracula) - Mixing text/markdown with code (100+ languages), images, diagrams (mermaid), math (LaTeX) - Multi-tab, multi-window support - Full version history built-in with revert, restore - Markdown files as note format, Obsidian compatible - One-time purchase. No subscription. - and many more great features...
3
1
4
287
Congrats @liquidai: LFM3-1.7B is SOTA on Multi-IF! 🏆 The 1.7B open model beats Baidu's closed ERNIE 5.0 on multi-turn, multilingual instruction following, 89.1 to 85.56. Qwen3-VL 235B-A22B Thinking, over 100x its size, scores 79.1. How small can a good assistant get?
4
10
3
69
11,070
Expertise is now called "taste".
125
The epic battle is fought between intelligence and complexity.
111
the "AI clones pro software in an hour" hype is noise from people who've never shipped real products. ignore it and keep going deep on coding so you build quality software and pull far ahead of everyone who stops
1
2
115
Quiver 4 will come with multi-tab support - it's implemented in a super lightweight way, unlike browser tabs. You can literally create a hundred tabs and they use the same memory as a single tab.
1
161
Yaogang Lian retweeted
1/4 Today, we’re releasing Mercury 3. Mercury 3 is by far our most powerful model and the most intelligent diffusion LLM ever released. It outperforms GPT-6 Luna and Haiku 5.5 on quality while running at 1000+ tok/sec.
🤖 Made with AI
291
454
230
5,262
490,102
Grok @bot is actually quite good.
107
35% improvements over previous v4.1-flash, purely due to RL. Exactly what I said before: agentic tool use has nothing to do with intelligence, but purely a RL thing.
🚀 DeepSeek-V4.1-Flash-1017 is here! Stronger benchmark performance, faster task completion, and shorter outputs than the previous version—a Pareto improvement in performance and cost. Now live on our website, app, and API. 🛠️ Optimized for @DeepSeekHarness, Tencent CodeBuddy, WorkBuddy, and more. Upgrade to DeepSeek Harness v0.2.1-rc.1 or later for the best experience with the new model. 1/4
1
148
Coding models have been good for a while. The biggest unlock in 2026 was actually affordable flash models with 1M context.
116
Just noting down a thought I had while studying transformer as a learning algoithm: Learning is not an optimization problem but happens in stages - you unlock one stage at a time. ML training is inexplicable from the perspective of traditional optimization algorithms - way too many params and way too much training data - thus my hunch that "learning is not optimization".
105
There is absolutely no reason to write shitty apps any more.
117
Best prompt for your coding agent:
🤖 Made with AI
71
The gemini @GeminiApp webapp finally allows editing previous messages (not just last one), also supports conversation branching. Better late than never, I guess.
106
Switched all my coding to deepseek-flash-4.1 via @ollama cloud and synthetic.new. Dirt cheap, GLM 5.3 level.
192
an observation: the highest leverage of AI at the moment is from the best people in non-AI fields the best ones are quietly using AI to speed up their own research by 10x or 100x
115
so coding is solved, math is also solved, time to solve physics: Astra, pls unify general relativity with quantum field theory, use 100000 subagents, make no mistakes
We’re sharing a solution to the Navier-Stokes Millennium Prize Problem, one of the deepest problems at the frontier of mathematics. The proof was produced by a group of agents, using an OpenAI next-generation model significantly more capable than GPT-6 Astra. The problem concerns whether the description of smooth three-dimensional fluid motion modeled by the Navier-Stokes equations can break down. It has remained unresolved for roughly 90 years.
137