DAIR.AI
RT @omarsar0: This team has been publishing some really interesting work on diffusion LLMs.

LLaDA 2.1 is a 100B discrete diffusion LLM with a draft-then-edit approach.

It hits a peak speed of 892 tokens/s on complex coding tasks.

Autoregressive models commit to every token permanently but LLaDA 2.1 can go back and fix mistakes mid-generation. The error handling capabilities are worth looking into.

What if an LLM could EDIT its own tokens in real-time, not just generate them? 🤯
Introducing LLaDA2.1 — a diffusion model that breaks from autoregressive dominance. It drafts fast, then fixes its own mistakes on the fly with Token-to-Token editing.
The result? 892 tokens/sec on a 100B model. 🔥
⚡ 892 TPS on HumanEval+ (coding)
⚡ 801 TPS on BigCodeBench
🧠 Real-time self-correction via T2T editing
✅ @lmsysorg SGLang Day 0 support — production-ready now
A "non-consensus" architecture now challenging the mainstream. Open-sourced TODAY. 👇
#LLaDA #TokenEditing #OpenSource #LLM #dLLM - Ant Open Source
tweet

God of Prompt
“Judging AI based on free-tier ChatGPT is like evaluating the state of smartphones by using a flip phone.”

well said

https://t.co/ivXRKXJvQg - Matt Shumer
tweet

Jukan
Micron Technology: It's Okay If HBM4 Is a Bit Delayed – Chae Min-suk, Korea Investment & Securities

Micron's HBM4 schedule delay is inevitable
- Micron's stock weakened after SemiAnalysis reported it would adjust Micron's share in NVIDIA's HBM4 to 0%
- However, we had already updated in our August 2025 report that Micron's HBM4 schedule would be delayed due to HBM4 speed upgrade issues
- Our HBM model already adjusted Micron's 2026 HBM4 market share to around 1%, factoring in redesign and sample schedule delays

Impact on earnings expected to be minimal
- The delay in Micron's HBM4 is not expected to significantly impact earnings
- This is because most of the upside in the 2026 memory market will be driven by commodity DRAM and NAND ASP increases
- As of Q4 2025, commodity DRAM operating margins likely already exceeded HBM operating margins
- Unlike HBM, where prices are determined on an annual basis, commodity DRAM ASPs are rising significantly on a quarterly basis, meaning the share of commodity DRAM in revenue and operating profit will expand considerably in 2026 compared to 2025
- According to our HBM model, HBM is estimated to account for 9% of Micron's DRAM by volume, 11% by revenue, and approximately 8% of DRAM operating profit in 2026
- Additionally, the current tight supply environment will also work in Micron's favor
- Even if Micron cannot supply HBM4, Samsung Electronics and SK hynix have limited additional supply capacity due to constrained capacity
- While market entry for HBM4 may be limited in 2026, full-scale supply should be possible from 2027 onward following the submission of improved samples

Short-term share price correction is a buying opportunity
- We maintain a positive outlook on Micron
- HBM remains a strategically important product, but its contribution to earnings in 2026 is relatively less significant compared to 2025
- In particular, the current tight supply environment limits competitors' ability to expand supply further, serving as a floor for Micron's earnings
- Data center memory demand will translate into rising commodity DRAM ASPs, and based on this, Micron's earnings are expected to continue trending upward

$MU
tweet

Javier Blas
OIL MARKET: Washington issues a new general license to allow oilfield-service companies to work in Venezuela — it’s a crucial step to boost oil output in the Latin American country. https://t.co/ozdDE2MiMZ
tweet

God of Prompt
RT @godofprompt: sometimes i have doubts about everything i'm making

ai will do this, ai will do that

what will be left for us to do?

but words like these make me grind harder https://t.co/EVBaHI8QUa
tweet

Michael Fritzell (Asian Century Stocks)
RT @firstadopter: The definitive list of the best newsletters on the internet. Congrats to all who made it. https://t.co/JLkDkiEaCY
tweet

Michael Fritzell (Asian Century Stocks)
RT @douglaskimkorea: We believe there is a 30% chance the Korean govt concludes mandatory treasury share cancellations by late February, a 40% chance in March, and a 30% probability of further delays beyond March 2026. https://t.co/ANQ5GvcQO9
tweet

Dimitry Nakhla | Babylon Capital®
RT @DimitryNakhla: Did you know that $SPGI & Anthropic announced a collaboration to bring financial data into Claude?

“With this cutting-edge integration, customers can unlock S&P Global’s trusted data and insights seamlessly and reliably within Anthropic’s Claude.”

Source: Kensho Communications https://t.co/3c8263d2aP

Kensho is one of the more under-the-radar assets inside $SPGI.

It’s $SPGI AI and data analytics platform, built to analyze massive, complex datasets across economics, geopolitics, financial markets, and corporate fundamentals—used by asset managers, banks, governments, and enterprises.

While most of the focus is on ratings and indices, Kensho quietly expands $SPGI moat by embedding AI-driven insights deeper into client workflows, increasing switching costs and long-term relevance.

Not flashy, but the kind of capability that can strengthen a toll-booth business over time.
___

3Q 2025 Earnings Call:

“We acquired Kensho back in 2018, including that acquisition since 2018, we have invested over $1B in AI innovation across three developmental stages…

Importantly, our AI innovation serves as a powerful example of our ability to leverage our scale, our expertise and our fiscal discipline. The fact that we made such bold investments early on means that we’ve been able to innovate very efficiently from a financial perspective.”
___

@contextinvestor thank you for the thoughtful comment. - Dimitry Nakhla | Babylon Capital®
tweet