Great paper choices @Dagmawi_Babi but some if not most are not easy for beginners, e.g the scaling law paper or any sort of scaling laws for LLMs are only useful for people who want to pretrain models, even that is not enough tbh. If people want just to have a high-level understanding of how LLMs work, they should just read blogs etc.

We're a Dr. now 😎, for your information.


Some people don't even do basic background check😂

Anyone here? I want to hear your takes on the conf?


Also, say hi👋 , if you're around

💰Nobel prize,

This year's Nobel prize is the probably the first one I really followed, since there were some familiar names. Usually, i just don't go deep into winners works or even check their names.


I thought the Nobel Prize was the perfect award but this year is the year I looked back to see some weird choices etc.



Also, this should be a good reminder that you should quit doing CRUD based stuff etc😂. If your job doesn't require you to come up with some logic either new or old, then you are probably not going to sneak in to the Nobel prize... jk btw

Hellooo👋,

I was pissed with the recent "AI" hype and didn't want to engage in it for the past few days, anyway what did I miss? Only some cool things please

A recent paper from Apple builds benchmark datasets that may be free from distribution shift and data leakage and thinking through failure test cases for reasoning.

And says there's ZERO evidence that LLM's show any signs of logical reasoning, and rather replicate reasoning steps learned in its training data through pattern recognition. 😳


Paper: https://arxiv.org/abs/2410.05229


This is a highly readable and straightforward paper — no math, no LLM or ML pre-reqs needed beyond just the basics (a curious reader without the basic knowledge can also follow through), you can check this blog too

Ev Fedorenko's Keynote at COLM

Some key points:
* in the human brain, language system is distinct from modules that activate when doing knowledge/cognition related things like maths. It can lead to activations of the others, but it's not necessary.

* LLMs are interesting to study as a in-silica model organism of language, as their representations somewhat resemble those of humans. And it looked like better models had better alignment, but curious if that continues.

*Baby LLMs


https://youtu.be/8xS7tjy92Ws?feature=shared