limengnlp's profile picture. PhD student @unipotsdam, supervised by @davidschlangen. Working on NLP, ML and CogSci. Prev @LstSaar. Former NLP engineer.

Meng Li

@limengnlp

PhD student @unipotsdam, supervised by @davidschlangen. Working on NLP, ML and CogSci. Prev @LstSaar. Former NLP engineer.

Meng Li reposted

this is a zero sum war of U.S...U.S made a mistake...when they let China freely do what they like from platform, AI structure to Chip, HBM....congratulations they lost 60% of their customers.


Meng Li reposted

Hello, community! We’re Tongyi Lab — the AI research institute under Alibaba Group, and the team behind Qwen, Wan, Tongyi Fun, and a growing ecosystem of models and frameworks loved by millions of developers worldwide. From this week forward, we will be sharing the latest updates…

Ali_TongyiLab's tweet image. Hello, community! We’re Tongyi Lab — the AI research institute under Alibaba Group, and the team behind Qwen, Wan, Tongyi Fun, and a growing ecosystem of models and frameworks loved by millions of developers worldwide.
From this week forward, we will be sharing the latest updates…

Meng Li reposted

Ilya Sutskever: bald Demis Hassabis: bald Noam Shazeer: bald Greg Brockman: bald forget AGI. forget curing cancer. cure baldness now. My hairline is on gradient descent.


Meng Li reposted

Day 1/5 of #MiniMaxWeek: We’re open-sourcing MiniMax-M1, our latest LLM — setting new standards in long-context reasoning. - World’s longest context window: 1M-token input, 80k-token output - State-of-the-art agentic use among open-source models - RL at unmatched efficiency:…

MiniMax__AI's tweet image. Day 1/5 of #MiniMaxWeek: We’re open-sourcing MiniMax-M1, our latest LLM — setting new standards in long-context reasoning.

- World’s longest context window: 1M-token input, 80k-token output
- State-of-the-art agentic use among open-source models
- RL at unmatched efficiency:…

Meng Li reposted

⏰ We introduce Reinforcement Pre-Training (RPT🍒) — reframing next-token prediction as a reasoning task using RLVR ✅ General-purpose reasoning 📑 Scalable RL on web corpus 📈 Stronger pre-training + RLVR results 🚀 Allow allocate more compute on specific tokens

qx_dong's tweet image. ⏰ We introduce Reinforcement Pre-Training (RPT🍒)  

 — reframing next-token prediction as a reasoning task using RLVR  

✅ General-purpose reasoning 
📑 Scalable RL on web corpus
📈 Stronger pre-training + RLVR results
🚀 Allow allocate more compute on specific tokens

In Hinton's NN class, there is an interesting tip to get a geometric view of high dimensional space. I think authors of interpretability papers did the opposite; they stare at LLMs and pray in their minds that it's linear and interpretable.

LLM Interpretability papers be like 👇🤙🏻 #AI美女 #MachineLearning

hbouammar's tweet image. LLM Interpretability papers be like 👇🤙🏻 #AI美女 #MachineLearning


Meng Li reposted

I just read this WSJ article on why Europe's tech scene is so much smaller than the US's and China's. I'm afraid that, like most articles on this topic, it largely misses the mark. Which in itself illustrates a key reason why Europe is lagging behind: when you fail to…

RnaudBertrand's tweet image. I just read this WSJ article on why Europe's tech scene is so much smaller than the US's and China's.

I'm afraid that, like most articles on this topic, it largely misses the mark.

Which in itself illustrates a key reason why Europe is lagging behind: when you fail to…

Meng Li reposted

📢 I am on the JOB market this year 📢 I am looking for both faculty and research scientist positions. My research makes AI agents useful and safe for humans. I enable them to effectively convey uncertainty, ask for help, learn from human feedback, and pursue goals that benefit…

khanhxuannguyen's tweet image. 📢 I am on the JOB market this year 📢

I am looking for both faculty and research scientist positions.

My research makes AI agents useful and safe for humans. I enable them to effectively convey uncertainty, ask for help, learn from human feedback, and pursue goals that benefit…

Meng Li reposted

Excited to be at #NAACL2025! Let’s meet (and grab a Char's Zaku sticker 🚀). 📅 May 4, 11–12, RepL4NLP: "Amuro&Char: Analyzing the Relationship between Pre-Training and Fine-Tuning" 📅 May 2, 12 PM, Ballroom B: "SHADES: Towards a Multilingual Assessment of Stereotypes in LLMs"

KaiserWhoLearns's tweet image. Excited to be at #NAACL2025! Let’s meet (and grab a Char's Zaku sticker 🚀).
📅 May 4, 11–12, RepL4NLP: "Amuro&Char: Analyzing the Relationship between Pre-Training and Fine-Tuning"
📅 May 2, 12 PM, Ballroom B: "SHADES: Towards a Multilingual Assessment of Stereotypes in LLMs"

Meng Li reposted

Every ChatGPT query costs more energy than the entire life of a fruit fly.


Meng Li reposted

AI phone agent realizes it is talking to a parrot


Meng Li reposted

🚀 Day 0: Warming up for #OpenSourceWeek! We're a tiny team @deepseek_ai exploring AGI. Starting next week, we'll be open-sourcing 5 repos, sharing our small but sincere progress with full transparency. These humble building blocks in our online service have been documented,…


Meng Li reposted

🚀 DeepSeek-R1 is here! ⚡ Performance on par with OpenAI-o1 📖 Fully open-source model & technical report 🏆 MIT licensed: Distill & commercialize freely! 🌐 Website & API are live now! Try DeepThink at chat.deepseek.com today! 🐋 1/n

deepseek_ai's tweet image. 🚀 DeepSeek-R1 is here!

⚡ Performance on par with OpenAI-o1
📖 Fully open-source model & technical report
🏆 MIT licensed: Distill & commercialize freely!

🌐 Website & API are live now! Try DeepThink at chat.deepseek.com today!

🐋 1/n

Meng Li reposted

The #NobelPrizeinPhysics2024 for Hopfield & Hinton rewards plagiarism and incorrect attribution in computer science. It's mostly about Amari's "Hopfield network" and the "Boltzmann Machine." 1. The Lenz-Ising recurrent architecture with neuron-like elements was published in…


Meng Li reposted

My Bet: Strawberry is algorithm distillation/procedural cloning. Everyone right now is coming up with ways to distill System 2 into System 1, but that will always be limited. We need to train the model to run the algorithms, not just outputs (and post-train with RL of course).

rm_rafailov's tweet image. My Bet: Strawberry is algorithm distillation/procedural cloning. Everyone right now is coming up with ways to distill System 2 into System 1, but that will always be limited. We need to train the model  to run the algorithms, not just outputs (and post-train with RL of course).

Meng Li reposted

rip

aryaman2020's tweet image. rip
aryaman2020's tweet image. rip

Meng Li reposted

Pretty fun paper, finetuning llama to produce blender code for synthetic renderings

Parskatt's tweet image. Pretty fun paper, finetuning llama to produce blender code for synthetic renderings

Meng Li reposted

Good Scientific American piece on the idea of AGI -I think and argue here that its incoherent - there is no general intelligence natural or artificial but different cognitive abilities that often trade-off.. scientificamerican.com/article/what-d…


Meng Li reposted

cognitive scientist: so the lesson of Clever Hans is we need.. engineer: more horses cognitive scientist: engineer: stacked horses. parallel horses. pooled horses. horse dropout. RL with horses in the loop. cognitive scientist: engineer: Hans is All You Need


United States Trends

Loading...

Something went wrong.


Something went wrong.