Ryan Huang

@balajis China never wanted to be number one, they just want to live better life, competition is an illusion of US

This paper is almost too good that I didn't want to share it Ignore the OpenClaw clickbait, OPD + RL on real agentic tasks with significant results is very exciting, and moves us away from needing verifiable rewards Authors: @YinjieW2024 Xuyang Chen, Xialong Jin, @MengdiWang10 @LingYang_PU

AI labs are coming to grab all data from all industries, and related all other SaaS with better AI agents

@charles_irl Why everything has to be a paper these days 😥

We find that the adoption of Cursor leads to a statistically significant, large, but transient increase in project-level development velocity, along with a substantial and persistent increase in static analysis warnings and code complexity. arxiv.org/abs/2511.04427

Introducing 𝑨𝒕𝒕𝒆𝒏𝒕𝒊𝒐𝒏 𝑹𝒆𝒔𝒊𝒅𝒖𝒂𝒍𝒔: Rethinking depth-wise aggregation. Residual connections have long relied on fixed, uniform accumulation. Inspired by the duality of time and depth, we introduce Attention Residuals, replacing standard depth-wise recurrence with learned, input-dependent attention over preceding layers. 🔹 Enables networks to selectively retrieve past representations, naturally mitigating dilution and hidden-state growth. 🔹 Introduces Block AttnRes, partitioning layers into compressed blocks to make cross-layer attention practical at scale. 🔹 Serves as an efficient drop-in replacement, demonstrating a 1.25x compute advantage with negligible (<2%) inference latency overhead. 🔹 Validated on the Kimi Linear architecture (48B total, 3B activated parameters), delivering consistent downstream performance gains. 🔗Full report: github.com/MoonshotAI/Att…

At #NVIDIAGTC, Jensen showed the industry where AI infra is heading: disaggregate the stack. NVIDIA’s Groq LPX push applies this to inference with Attention–FFN Disaggregation. Our view: this idea matters even more for long-context LLM training. 🧵 github.com/hao-ai-lab/Dis…

@CarlZha Yes it is

Why are all the on the ground footages from Strait of Hormuz coming from Chinese sailors posting on Douyin? Is there a deliberate suppression on the US media platform???

@paulg then what about Israel?

Prediction: When fighting Iran gets too painful because of oil prices or polls or whatever, Trump will claim that the current state of things, whatever it happens to be, was his goal, declare victory, and retreat.

History of software development process : waterfall —> agile-> interactive on-demand

Openclaw is the new Ruby on Rails

@dwarkesh_sp Is AI can do coding and math, does math still matters?

What should I ask Terence Tao?

@zarazhangrui Big tech is aggressively adopting AI than startups

Never have I seen a larger gap between how startups and how large companies: - use and work with AI - select talent - organize their teams

@scaling01 @jasondeanlee Not that simple

You had one job Meta: - take the DeepSeek recipe - scale the recipe to 5T params - train it with your bazillions of H100s and unlimited social media data - RL until your staff is burnt out from babysitting the runs - distill into cuter 30B, 100B and 500B models - profit

@GergelyOrosz They don’t care about consumers facing pages

One thing that endlessly frustrates with Anthropic, a $300B+ dollar company, where most code is written with AI: Their landing page for paying customers, Claude .ai has been broken for weeks UX-wise, and no one notices or cares or fixes: It "loses" stuff I type while it loads:

Why do SOTA models stop at 1M context? seem like it is an engineering limit, after 1M tokens, quadratic scaling really "catches" you

**Corrected & concise version:** Why do SOTA models stop at 1M context? Is it an engineering limit or diminishing returns from the model itself?

@New_tres 😂

⚡🇮🇱🇨🇳 BREAKING: Netanyahu calls on China to intervene to calm Iran and mediate to stop the war and prevent regional escalation.

@ChristosTzamos Each token is an assembly instruction? That’s cool

1/4 LLMs solve research grade math problems but struggle with basic calculations. We bridge this gap by turning them to computers. We built a computer INSIDE a transformer that can run programs for millions of steps in seconds solving even the hardest Sudokus with 100% accuracy

@idosal1 That’s wild!

AgentCraft v1 is live ⚔️ Control your agents like it's an RTS game! It's early. It's rough. It's fun. npx @idosal/agentcraft