This week, we explore what FlashMLA is, why it matters, and how it builds on key innovations like Mixture of Experts, infinite context attention, and context caching! https://
louisbouchard.substack.com/p/how-flashmla
-cuts-kv-cache-memory
…
RESEARCH
-
FlashMLA: Cutting KV Cache Memory in LLMs
By
–
-
Conference Papers Volume Impact on Research Citation Priorities
By
–
Since there will be 30000 top conference papers every year, and surely you will not read/cite most of your even directly related paper, time to think about what will be real impact (especially for industrial researchers)
-
01.AI Builds Windows System to Complement DeepSeek Kernel
By
–
DeepSeek is becoming a Windows kernel demanded by businesses, but https://t.co/ylJtTsF7S6 is aspired to build the Windows system and interface to ignite it. Check out more on: https://t.co/IBRt3Hwgaw
— Kai-Fu Lee (@kaifulee) 20 mars 2025
Thanks @BloombergTV @DavidInglesTV and @BelleDroulers for the insightful… https://t.co/yrAREwr8NEDeepSeek is becoming a Windows kernel demanded by businesses, but 01.AI is aspired to build the Windows system and interface to ignite it. Check out more on: b.01.ai Thanks @BloombergTV @DavidInglesTV and @BelleDroulers for the insightful interview. David Ingles (@DavidInglesTV) "Is OpenAI's model even sustainable?" The China moment that sparked the sudden shift in AI economics and where the value add now lies for investors and innovators — https://nitter.net/DavidInglesTV/status/1902595690515312693#m
-
Tsinghua Hephaestus Wins RoboCup German Open 2025 Championship
By
–
9:0! Tsinghua Hephaestus won the championship of the RoboCup German Open 2025!
— 机器之心 JIQIZHIXIN (@jiqizhixin) 20 mars 2025
Congratulations! @Tsinghua_Uni https://t.co/3GsRABYgmw9:0! Tsinghua Hephaestus won the championship of the RoboCup German Open 2025! Congratulations! @Tsinghua_Uni
-
Kolmogorov: New AI Research Paper Code Released
By
–
Paper: https://
arxiv.org/pdf/2503.13992
Code: https://
github.com/facebookresear
ch/kolmogorov
…
Leaderboard: -

Compression Benchmarks: Infinite Problem Instances for AI Evaluation
By
–
The benefits include: An essentially infinite number of problem instances of varying difficulty is readily available, strong baselines already exist, the evaluation metric (compression) cannot be gamed, and pretraining data contamination is highly unlikely.
-

Kolmogorov-Test: Compression as Intelligence Metric for LLMs
By
–
Compression = Intelligence? This ICLR 2025 paper introduced the KOLMOGOROV-TEST (KT), a compression-as-intelligence test for code generating LLMs.And the team identified several benefits of KT for both evaluation and training.
-
Compositionality Over End-to-End Learning: The Future of AI Models
By
–
Much of the field obsesses over end-to-end learning. But strong generalization requires compositionality: building modular, reusable abstractions, and reassembling them on the fly when faced with novelty. The models of the future won't be just pipes, they will be Lego castles.
-

Data Structure Alignment for Machine Learning Questions
By
–
Does your data's structure match the question you want to answer? https://
buff.ly/MSSxVEu
#AI #MachineLearning #DeepLearning #LLMs #DataScience -
Kevin Weil Podcast Interview: AI and OpenAI Discussion
By
–
I've got @kevinweil coming on the podcast next week. What would you wish you could ask him about AI, OpenAI, his career, or anything else?