Saurabh Prasad

There is a restaurant that serves Chinese & Italian food, located in the New Delhi metro area in India, named 'Chao Bella'. 😄

Here's their Insta page:
www.instagram.com/chaobellacpo...

Here's why "alignment research" when it comes to LLMs is a big mess, as I see it.

Claude is not a real guy. Claude is a character in the stories that an LLM has been programmed to write. Just to give it a distinct name, let's call the LLM "the Shoggoth".

Happy new year, Bluesky peeps! 🎉🥳

Wish you all a very happy, healthy, and prosperous 2025! 💐

GitHub - nedjmaou/CMT122-2425-Resources: Resources for CMT122 students (2024-2025). Resources for CMT122 students (2024-2025). Contribute to nedjmaou/CMT122-2425-Resources development by creating an account on GitHub.

@camachocollados.bsky.social and I taught ML for #nlp last semester.
Here is a list of resources that we shared with the students (sorted by theme) in this blogpost: github.com/nedjmaou/CMT...
[Not all students were familiar with coding so it also includes resources for beginners.]

AIs Will Increasingly Fake Alignment This post goes over the important and excellent new paper from Anthropic and Redwood Research, with Ryan Greenblatt as lead author, Alignment Faking in Large Language Models.

LLMs faking alignment during training. #mlsky

thezvi.substack.com/p/ais-will-i...

NeurIPS2024 Related RL papers | Notion Deep RL papers

#NeurIPS2024 wrapped up last week. I put together a curated reading list for #DeepRL and #reinforcementlearning work. (represents my interests).

Talks and workshops:
third-crowd-c77.notion.site/NeurIPS2024-...

Curated reading list
fracturedplane.notion.site/NeurIPS2024-...

#Holidayreading

Interviewer: Can you explain this gap in your resume?

Physicist: I can only tell you about my momentum at that time, not the position

One of the most bizarre + complex talks at NeurIPS [1] was given by my fellow Yorkshireman, the inimitable Prof Karl Friston [2], explaining active inference to a room full of #AI people who are not really neuroscientists. This was interesting to me because... (1/n)

Sometimes I am a Tree: Data Drives Unstable Hierarchical Generalization Language models (LMs), like other neural networks, often favor shortcut heuristics based on surface-level patterns. Although LMs behave like n-gram models early in training, they must eventually learn...

Transformer LMs get pretty far by acting like ngram models, so why do they learn syntax? A new paper by sunnytqin.bsky.social, me, and @dmelis.bsky.social illuminates grammar learning in a whirlwind tour of generalization, grokking, training dynamics, memorization, and random variation. #mlsky #nlp

Supporting GLP-1 prescribing with digital twin technology | TechTarget Learn how digital twin technology could support clinical decision-making for providers prescribing GLP-1s.

🧪 Emory leverages #DigitaTwin tech to guide GLP-1 prescriptions for weight and diabetes care. An AI model simulates clinical decisions, providing expert-level support for patient suitability. 🩺💻 #MLSky

The Year in Computer Science | Quanta Magazine Researchers got a better look at chatbots’ thoughts, amateurs learned just how complicated simple systems can be, and codes became expert self-fixers.

The ever elusive quantum computer moved a few crucial steps toward practical applications this year.
www.quantamagazine.org/the-year-in-...

The Year in Biology | Quanta Magazine Biologists used artificial intelligence to make discoveries about molecules and the brain, and overturned long-held assumptions about the immune system and RNA.

Read about this year’s biggest moments in biology.
www.quantamagazine.org/the-year-in-...

The Year in Physics | Quanta Magazine Physicists discovered strange supersolids, constructed new kinds of superconductors, and continued to make the case that the cosmos is far weirder than anyone suspected.

This year, researchers detected a hint of a signal that, if real, could upend and deepen our understanding of the fundamental laws of the universe. Read our annual review of the biggest developments in physics:
www.quantamagazine.org/the-year-in-...

The Year in Math | Quanta Magazine Landmark results in geometry and number theory marked an exciting year for mathematics, at a time when advances in artificial intelligence are starting to transform the subject’s future.

Here are the biggest breakthroughs that happened in mathematics this year.
www.quantamagazine.org/the-year-in-...

I'll get straight to the point.

We trained 2 new models. Like BERT, but modern. ModernBERT.

Not some hypey GenAI thing, but a proper workhorse model, for retrieval, classification, etc. Real practical stuff.

It's much faster, more accurate, longer context, and more useful. 🧵

Finally, a Replacement for BERT: Introducing ModernBERT We’re on a journey to advance and democratize artificial intelligence through open source and open science.

Great blog post (by a 15-author team!) on their release of ModernBERT, the continuing relevance of encoder-only models, and how they relate to, say, GPT-4/llama. Accessible enough that I might use this as an undergrad reading.

Picture of a plaque assay (purple field with clear holes) showing gorgeous poliovirus plaques

My replies are perpetually full of anti-vaxxers these days telling me about polio vaccines.

Not shockingly, most of what they are saying is wrong. Luckily, I trained with Vincent Racaniello & he taught me a few things about poliovirus.

So let’s discuss the king of the Picornaviridae👇🏻

The AI Pioneer With Provocative Plans for Humanity | Quanta Magazine While some fret about technology’s social impacts, Raj Reddy still believes in the power of artificial intelligence to improve lives.

Artificial intelligence is going to improve productivity. That will also create more wealth. How do you keep AI-enhanced productivity from only benefiting the wealthy? Raj Reddy, one of the pioneers of AI research, has ideas. www.quantamagazine.org/the-ai-pione...

Infamous paper that popularized unproven COVID-19 treatment finally retracted Study on hydroxychloroquine by Didier Raoult and colleagues gets pulled on ethical and scientific grounds

It took four days from submission to publication, and nearly five years from publication to retraction. After campaigning by many, many scientists, and an investigation by Elsevier, an infamous paper on hydroxychloroquine as a COVID-19 treatment has been retracted. 🧪
www.science.org/content/arti...

YouTube video by seremot Ilya Sutskever: "Sequence to sequence learning with neural networks: what a decade"

Ilya Sutskever full talk "Sequence to sequence learning with neural networks: what a decade" at NeurIPS 2024 in Vancouver, Canada.

www.youtube.com/watch?v=1yvB...

Want all NeurIPS/ICML/ICLR papers in one single .bib file? Here you go!

🗞️ short blog post: fabian-sp.github.io/posts/2024/1...

📇 bib files: github.com/fabian-sp/ml-bib

Fun fact related to this thread:

Do you know what brought the ReLU activation function back into vogue in AI?

It was this paper from Yoshua's group, which was motivated by the observation that ReLU is a better match to real neurons' IO-functions!

proceedings.mlr.press/v15/glorot11a

🧠📈 #NeuroAI

1/ Okay, one thing that has been revealed to me from the replies to this is that many people don't know (or refuse to recognize) the following fact:

The unts in ANN are actually not a terrible approximation of how real neurons work!

A tiny 🧵.

🧠📈 #NeuroAI #MLSky

🌟Noether's razor⭐️ Our NeurIPS 2024 paper connects ML symmetries to conserved quantities through a seminal result in mathematical physics: Noether's theorem. We can learn neural network symmetries from data by learning associated conservation laws. Learn more👇. 1/16🧵

🎯 How can we empower scientific discovery in millions of nature photos?

Introducing INQUIRE: A benchmark testing if AI vision-language models can help scientists find biodiversity patterns- from disease symptoms to rare behaviors- hidden in vast image collections.

Thread👇🧵

Gukesh Dommaraju becomes youngest world chess champion after horrific Ding Liren blunder * Indian teenager becomes 18th world chess champion * Modi hails ‘historic and exemplary’ achievement * Move-by-move report of Game 14 – as it happened * Play through 22 famous world championship games Indian teenager Gukesh Dommaraju capped a…

Gukesh Dommaraju becomes youngest world chess champion after horrific Ding Liren blunder

Have you ever wondered how training dynamics differ between LLMs 🖋️ and Vision 👁️ models? We explore this and close the gap between VMs and LLMs in our #NeurIPS2024 paper "TrAct: Making First-layer Pre-Activations Trainable".
Paper link 📜: arxiv.org/abs/2410.23970
Video link 🎥: youtu.be/ZjTAjjxbkRY
🧵

It's never too late to start learning a new instrument 😄

That's so cool!

I'm a software architect & engineer/developer, but I enjoy playing the guitar & piano as a hobby! 🎸🤘🏼😊

Could you please add me 👋😊