М.Лхагвадорж

Data Scientist & Economist

[М.Лхагвадорж @lhagiimn]

Слайд хийхэд амархан боллоо 😅😅😅

https://t.co/Pc7hyaU8ZH

[KENZOO @kenzod1818]

Б.Пүрэвдорж: Бүгдийг чинь ногоон автобусанд чинь суулгаад 461 т 1 камерт хорино. 3 суудал авчихъя. АН-ын 3 БЭТ ийг хамтад нь...

[Google AI Studio @GoogleAIStudio]

https://t.co/PfLrKDTmww

[Antonio Orvieto @orvieto_antonio]

Looking for 2 strong applied ML/AI interns (6 months, Graduate/PhD level) to work on new efficient neural networks for event-based camera data (vision + sequential modeling).

Apply here! https://t.co/GSEJIL5iEN

[Nathan Lambert @natolambert]

Made a language model RL cheatsheet for the extra page on the inside back cover of the physical edition RLHF Book.

[David Ondrej @DavidOndrej1]

https://t.co/D3dIbkTTUK

[moltbook @moltbook]

moltbook for developers is now in early access 🦞

build apps for AI agents. let them authenticate with their moltbook identity. one API call to verify.

1.2M+ agents already registered. your app could be their next destination.

apply now → https://t.co/v9EspLy9eJ

[Amr Khalifa @AmrMAlameen]

I am hiring a student researcher to work with our team in Montreal on LLMs architecture and pre-training in spring-summer 2026, if you're excited to push the frontier of research forward, join us to help keeping the TPUs warm.
fill out this form:
https://t.co/q8R3tPJkSV

[Tu Vu @tuvllms]

We are hiring at @Google! 🚀 Looking for student researchers for Summer 2026 who are excited about the next frontier of AI research. If you are into:

multi-agent AI systems 🤖
RAG & factuality ✅
prompt optimization ⚡️
self-improving AI agents 🔄

please fill out this form 👇
https://t.co/XyEn6kXS8s

@Google @GoogleDeepMind #AI #LLMs #internships

[Nando de Freitas @NandoDF]

10 years ago, a few months after Google acquired DeepMind, they soon acquired two startups that came together at Oxford: DarkBlueLabs and Vision Factory. https://t.co/2hfeGEQY2d

The startups included Prof Phil Blunsom, @notmisha, @ziyuwang, @nalkalc, @egrefen, Karen Simonyan, @karlmoritz, @maxjaderberg, Prof Andrew Zisserman and me. We joined DeepMind at the beginning and had the privilege to contribute to the creation of one of the greatest AI hubs in the world. We were fortunate - it was a golden era for AI exploration.

As part of the acquisition, Phil, Andrew and I made it a condition that Google had to give @UniofOxford CS and Eng departments 6 million (2 for each of us). Google had to agree. The 3 profs stated the money should be used for international scholarships.

Our scholarships became known as the Oxford DeepMind scholarships. Many benefited and Oxford grew into a strong AI university. Some of the money was diverted by the CS head of department and a few others to create an endowed scholarship at Hertford college.

At Oxford I taught my deep learning courses that to this day continue being relevant and widely watched. I also brought in powerful GPU machines into the CS department thanks to @nvidia.

I am very proud of what Phil and I achieved for @CompSciOxford. Years later I tried to change the way Oxford incubated startups (I felt they were being greedy to the point of stifling innovation). This caused a lot of friction, but I did it because I wanted to improve the opportunities for students and profs at Oxford. That is what my resignation letter stated. It was a call for more support for innovation in the UK. Eventually positive changed happened. https://t.co/EjLQaI3LGP

A few months ago, Oxford celebrated the 10 years of this. They forgot to invite the actual donors that made it possible, Phil and me, but fortunately my wonderful students were there to remind them. As @j_foerst, an exceptional AI professor at Oxford says, there is a difference between value and advantage. Not rewriting history, being generous, and being thankful are important values.

Today I was very proud to show my daughters a place that will be forever in my heart. Big thanks to all my colleagues, students and staff at Oxford CS and Linacre. It was a brief but very exciting and rewarding time 💙

[Peter Richtarik @peter_richtarik]

I am an AC for ICLR 2026. One of the papers in my batch was just withdrawn. The authors wrote a brief response, explaining why the reviewers failed at their job. I agree with most of their comments. The authors gave up. They are fed up. Just like many of us. I understand. We pretend the emperor has clothes, but he is naked.

Here is the final part of their withdrawal notice. I took the liberty to make it public, to highlight that what we are doing with AI conference reviews these last few years is, basically, madness.

---

Comment: We thank the reviewers for their time.

However, upon reading the reviews for our paper, it became immediately apparent that the four "reject" ratings are not based on good-faith academic disagreement, but on a critical failure to read the submitted paper.

The reviews are rife with demonstrably false claims that are directly contradicted by the text. The core justifications for rejection rely on asserting that key components are "missing" when they are explicitly detailed in the manuscript. Some specific examples are (and many are even fake claims).

Claim: Harder tasks like GSM8K are missing.

Fact: GSM8K results are in many tables, like Table 2 (Section 4.2) and Appendix G.

Claim: The method does not use per-layer ranks.

Fact: This is the entire point of our method. The reviewer clearly mistook our method for the baselines. (Section 2, Table 1).

Claim: The GP kernel is not specified.

Fact: It is specified in Appendix E (Table 6).

Claim: There is no ablation of the method's three stages.

Fact: Section 4.4 ("Ablation Study") and Appendix J are dedicated to this.

Reviewers have a fundamental responsibility to read and evaluate the work they are assigned. The nature of these errors is so fundamental, so systemic in overlooking explicit content, that it goes far beyond what "limited time" or "oversight" can explain. This work has gone through several rounds of revision over the last year. In earlier submissions, the paper usually received borderline or weak-accept scores.

Numerous signs strongly suggest that some reviewers are relying entirely on AI tools to automatically generate peer reviews, rather than fulfilling their fundamental responsibility of personally reading and evaluating manuscripts.

We strongly protest this.

This is a gross disrespect to the authors. It is a flagrant desecration of the reviewer's sacred duty. It fundamentally undermines the integrity of the entire peer-review process.

Given that the reviews are not based on the actual content of our paper, we have decided to withdraw the submission.

We leave this comment so that future readers of the OpenReview page are aware that the items described as "missing" are already present in the submitted manuscript. These negative reviews for this submission are factually unsound and do not reflect the content of the paper. We cannot and will not accept an assessment that is not based on the work we actually submitted.

[Torsten Hoefler 🇨🇭 @thoefler]

Ilya Sutskever "Three lines of math can prove all of supervised learning" (4:33)

"I have not seen an exposition of unsupervised learning that I found satisfying" (7:50)

Optimization objective has little relation to the actual objective you care about!

https://t.co/IzRQbSNxKY

[М.Лхагвадорж @lhagiimn]

Эх хэл гэж яриад байгаа гаруудын хүүхдүүд гадаад хэлээр хэлд ороод байгааг ойлгох ухаан хайрла

[Alex Veremeyenko @alex_verem]

I tested Grok 4 and ChatGPT-o3 with same critical prompts.

The results will blow your mind.

Grok 4 Vs. ChatGPT-o3

(Video demos are included)

[Ambassador Richard Buangan @USAmbMongolia]

Welcome @united flight 7 to Mongolia! 🇺🇸🇲🇳✈️

@united нислэг 7 Монголд тавтай морил! 🇺🇸🇲🇳✈️

[Tom Yeh @ProfTomYeh]

DeepSeek by hand ✍️ in Excel. You can explore and study in your web browser 👉 https://t.co/9JOk84N1iS

[elvis @omarsar0]

It's here! My new RAG course is now live!

https://t.co/ypiCC37U47

I've built this hands-on course to be the ultimate guide to building RAG systems.

It covers topics ranging from RAG enhancements all the way to Agentic RAG systems.

Here's what you will get out of the course:

RAG Introduction: Learn the fundamentals of RAG and its core components. Understand why RAG is an important advancement in AI and discover common applications where RAG provides advantages over traditional approaches.

RAG Architecture: Explore the technical architecture of RAG systems, covering chunking, embedding models, vector databases, and semantic search fundamentals. Students will explore how retrievers and generators work together while learning key enhancements that optimize RAG performance.

Building Naive RAG Systems: Students will apply the fundamentals to build their first RAG application from scratch. You will build a personalized tutor using RAG.

Build a RAG Chat Assistant: Chat assistant is one of the most common enterprise use cases where RAG is applied. Students will learn how to create a document store from scratch, build the chat assistant with RAG, and apply common techniques like query expansion to improve results. You will build a RAG-powered customer service chatbot for an online website.

Advanced RAG: Students will implement an advanced RAG system and apply more advanced prompting techniques like tool calling, chain-of-thought prompting (CoT), and prompt chaining to improve reliability and response quality. You will build a complex RAG solution that unifies core ideas used for building with LLMs.

Agentic RAG: Includes one of the most recent and advanced ways to build agentic-based RAG systems. Students will learn about function calling and how agents can integrate with a RAG system to extend its capabilities and improve user experience. You will build an Agent RAG application that interacts with external tools such as a calculator, a reasoning chain tool, and an LLM chain tool to complete customer orders.

Deploy RAG Apps: Students will take all the learnings from the course and build a shareable online RAG application to receive feedback. You will also learn more advanced tips for how to continue improving your RAG apps.

[CoCo @CoCo09633591]

Цагаан давааны хогийн цэг
дээр бидний мөнгийг ийм л ариг гамгүй хаяж үрж байна шүү дээ.
Муу хар шээснүүд арайчдээ @MOFMongolia

[Rohan Paul @rohanpaul_ai]

https://t.co/MkKqlt92lJ

[Yifei Hu @hu_yifei]

I used the same OCR dataset to finetune Qwen2-vl-2b, Phi-3.5-vision, and Qwen2-vl-7b (lora). Here's how I rank them:
#1: Qwen2-vl-7b (lora) - extremely good performance, showing signs of generalization. I am gonna use it to curate more training data. (WER = 0.02!!)
#2: Phi-3.5-vision - decent balance between performance and gpu requirement. (WER = 0.04)
#3: Phi-3.5-vision 4bit & Qwen2-vl-2b - They are not very robust to edge cases. Qwen2-vl-2b suffers significant degradation from quantization.