Manqing Liu (@manqingliu5) 's Twitter Profile
Manqing Liu

@manqingliu5

PhD Candidate @Harvard interested in causal machine learning; She/Her

ID: 1945170565

linkhttps://manqingliu.github.io/ calendar_today07-10-2013 20:56:36

116 Tweet

121 Takipçi

345 Takip Edilen

Manqing Liu (@manqingliu5) 's Twitter Profile Photo

How Claude 3.7 romanticizes functional analysis with striking analogies, this is so beautiful #AI #Math #FunctionalAnalysis

How Claude 3.7 romanticizes functional analysis with striking analogies, this is so beautiful 
#AI #Math #FunctionalAnalysis
MatthewBerman (@matthewberman) 's Twitter Profile Photo

We knew very little about how LLMs actually work...until now. Anthropic just dropped the most insane research paper, detailing some of the ways AI "thinks." And it's completely different than we thought. Here are their wild findings: 🧵

We knew very little about how LLMs actually work...until now.

<a href="/AnthropicAI/">Anthropic</a> just dropped the most insane research paper, detailing some of the ways AI "thinks."

And it's completely different than we thought.

Here are their wild findings: 🧵
Neel Nanda (@neelnanda5) 's Twitter Profile Photo

After supervising 20+ papers, I have highly opinionated views on writing great ML papers. When I entered the field I found this all frustratingly opaque So I wrote a guide on turning research into high-quality papers with scientific integrity! Hopefully still useful for NeurIPS

After supervising 20+ papers, I have highly opinionated views on writing great ML papers. When I entered the field I found this all frustratingly opaque

So I wrote a guide on turning research into high-quality papers with scientific integrity! Hopefully still useful for NeurIPS
Ethan Mollick (@emollick) 's Twitter Profile Photo

Huh. Looks like Plato was right. A new paper shows all language models converge on the same "universal geometry" of meaning. Researchers can translate between ANY model's embeddings without seeing the original text. Implications for philosophy and vector databases alike.

Huh. Looks like Plato was right.

A new paper shows all language models converge on the same "universal geometry" of meaning. Researchers can translate between ANY model's embeddings without seeing the original text.

Implications for philosophy and vector databases alike.
Petar Veličković (@petarv_93) 's Twitter Profile Photo

the 1st draft 'g' chapter of the geometric deep learning book is live! 🚀 alice enters the magical, branchy world of graphs & gnns 🕸️ (llms are there too!) i've spent 7+ years studying, researching & talking about graphs. this text conveys what i've learnt. more in thread 💎

the 1st draft 'g' chapter of the geometric deep learning book is live! 🚀

alice enters the magical, branchy world of graphs &amp; gnns 🕸️ (llms are there too!)

i've spent 7+ years studying, researching &amp; talking about graphs. this text conveys what i've learnt.

more in thread 💎
Neel Nanda (@neelnanda5) 's Twitter Profile Photo

I'm very excited about our vision for "mech interp" of CoT: Study reasoning steps and their connections - analogous to activations Don't just read it: study attn, causally intervene, and, crucially, resampling - study the distn over CoTs, not just this one There's lots to do!

I'm very excited about our vision for "mech interp" of CoT:

Study reasoning steps and their connections - analogous to activations

Don't just read it: study attn, causally intervene, and, crucially, resampling - study the distn over CoTs, not just this one

There's lots to do!
Peng Qi (@qi2peng2) 's Twitter Profile Photo

Seven years ago, I co-led a paper called 𝗛𝗼𝘁𝗽𝗼𝘁𝗤𝗔 that has motivated and facilitated many #AI #Agents research works since. Today, I'm asking that you stop using HotpotQA blindly for agents research in 2025 and beyond. In my new blog post, I revisit the brief history of

Anthropic (@anthropicai) 's Twitter Profile Photo

New Anthropic research: Why do some language models fake alignment while others don't? Last year, we found a situation where Claude 3 Opus fakes alignment. Now, we’ve done the same analysis for 25 frontier LLMs—and the story looks more complex.

New Anthropic research: Why do some language models fake alignment while others don't?

Last year, we found a situation where Claude 3 Opus fakes alignment.

Now, we’ve done the same analysis for 25 frontier LLMs—and the story looks more complex.
Scott Emmons (@emmons_scott) 's Twitter Profile Photo

Is CoT monitoring a lost cause due to unfaithfulness? 🤔 We say no. The key is the complexity of the bad behavior. When we replicate prior unfaithfulness work but increase complexity—unfaithfulness vanishes! Our finding: "When Chain of Thought is Necessary, Language Models

Is CoT monitoring a lost cause due to unfaithfulness? 🤔

We say no. The key is the complexity of the bad behavior. When we replicate prior unfaithfulness work but increase complexity—unfaithfulness vanishes!

Our finding: "When Chain of Thought is Necessary, Language Models
Rohin Shah (@rohinmshah) 's Twitter Profile Photo

Chain of thought monitoring looks valuable enough that we’ve put it in our Frontier Safety Framework to address deceptive alignment. This paper is a good explanation of why we’re optimistic – but also why it may be fragile, and what to do to preserve it. x.com/balesni/status…

Gabriele Berton (@gabriberton) 's Twitter Profile Photo

If you think NeurIPS reviews are getting worse because of LLMs, think again The seminal 2015 distillation paper from Jeff Hinton, Oryol Vinyals, and Jeff Dean was rejected by NeurIPS for lack of impact, was published as a workshop, and it has now 26k citations🤯

If you think NeurIPS reviews are getting worse because of LLMs, think again
The seminal 2015 distillation paper from Jeff Hinton, Oryol Vinyals, and Jeff Dean was rejected by NeurIPS for lack of impact, was published as a workshop, and it has now 26k citations🤯
alphaXiv (@askalphaxiv) 's Twitter Profile Photo

In-context learning is just gradient descent without explicit training! This paper "Learning without training: The implicit dynamics of in-context learning" shows that ICL can be mathematically interpreted as an implicit low-rank weight update during inference.

In-context learning is just gradient descent without explicit training!

This paper "Learning without training: The implicit dynamics of in-context learning" shows that ICL can be mathematically interpreted as an implicit low-rank weight update during inference.
Emmanuel Ameisen (@mlpowered) 's Twitter Profile Photo

Earlier this year, we showed a method to interpret the intermediate steps a model takes to produce an answer. But we were missing a key bit of information: explaining why the model attends to specific concepts. Today, we do just that 🧵

Earlier this year, we showed a method to interpret the intermediate steps a model takes to produce an answer.

But we were missing a key bit of information: explaining why the model attends to specific concepts.

Today, we do just that 🧵
ludwig (@ludwigabap) 's Twitter Profile Photo

The "Circuit Analysis Research Landscape" for August 2025 is out and is an interesting read on "the landscape of interpretability methods" and model biology Qwen3 4B is also out on Circuit Tracer

The "Circuit Analysis Research Landscape" for August 2025 is out and is an interesting read on "the landscape of interpretability methods" and model biology 

Qwen3 4B is also out on Circuit Tracer
steve hsu (@hsu_steve) 's Twitter Profile Photo

Is Chain-of-Thought Reasoning of LLMs a Mirage? ... Our results reveal that CoT reasoning is a brittle mirage that vanishes when it is pushed beyond training distributions. This work offers a deeper understanding of why and when CoT reasoning fails, emphasizing the ongoing

Is Chain-of-Thought Reasoning of LLMs a Mirage?

... Our results reveal that CoT reasoning is a brittle mirage that vanishes when it is pushed beyond training distributions. This work offers a deeper understanding of why and when CoT reasoning fails, emphasizing the ongoing
Pingbang Hu 🇹🇼 (@pingbanghu) 's Twitter Profile Photo

As a PhD student, sometimes I feel isolated. Not from the world, but from myself. A while ago, in an event full of startup founders, someone asked me what do I really want to do, and I said: "I want to do impactful and meaningful things." That's from the bottom of my heart, for

Andrej Karpathy (@karpathy) 's Twitter Profile Photo

In era of pretraining, what mattered was internet text. You'd primarily want a large, diverse, high quality collection of internet documents to learn from. In era of supervised finetuning, it was conversations. Contract workers are hired to create answers for questions, a bit

Andrej Karpathy (@karpathy) 's Twitter Profile Photo

Excited to release new repo: nanochat! (it's among the most unhinged I've written). Unlike my earlier similar repo nanoGPT which only covered pretraining, nanochat is a minimal, from scratch, full-stack training/inference pipeline of a simple ChatGPT clone in a single,

Excited to release new repo: nanochat!
(it's among the most unhinged I've written).

Unlike my earlier similar repo nanoGPT which only covered pretraining, nanochat is a minimal, from scratch, full-stack training/inference pipeline of a simple ChatGPT clone in a single,