Ha, I hope it didn't scare people away. The attention mechanism is probably the toughest one in the book (but it's necessary as it's otherwise building a car without the engine). In any case, hope you are liking the book overall!
MACHINE LEARNING
-
LLMs Versus Classic Methods: Knowledge Graphs and Explicit Techniques
By
–
I think these are more used in the more classic, pre-LLM era. LLMs are not typically using those explicitly (unless we are talking about knowledge-graph augmented LLMs)
-
DeepResearch Advances AI Reasoning Capabilities with Humanity’s Last Exam
By
–
Barely a week old, #DeepResearch is making remarkable strides to smashing the ceiling of human reasoning. ***
Humanity’s Last Exam (HLE) was developed year by @scale_AI and the Center for AI Safety (CAIS) as the ULTIMATE test of AI’s ability to reason at the highest human -
Data Management: The Key To Unlocking AI And BI Potential
By
–
Data Management: The Key To Unlocking AI And BI Potential
#AI #AIio #AIInnovation #ML #DataScience #Futureofwork @lexfridman @sama @kaifulee @ID_AA_Carmack @karpathy @2morrowknight @ylecun http://
ow.ly/HvZW30sJE9V -
Robust Representation Consistency Model Accepted to ICLR 2025
By
–
Thrilled to announce our paper "Robust Representation Consistency Model (rRCM)" was accepted to #ICLR2025.
It combines contrastive learning with consistency training to enhance robust representation learning and sets a new standard in certified robustness. @JiachenLei @julberner -
Neural Networks Model General Data Structures Before Transformers
By
–
Yup Grammar as a Foreign Language was mind blowing to me, and the first demo that we could really model general data structures with neural nets — even pre-transformers. Felt like everything was possible… and alas, it turns out that everything was possible : )
-
Flexibility in Reasoning LLMs with Preference Tuning
By
–
Oh I see. Yeah maybe depends on how extensive the preference tuning was. But in general it should be possible to achieve the same "flexibility" there as with non-reasoning LLMs.
-
Meta PARTNR: Open Source Framework for Human-Robot Collaboration Research
By
–
Meta PARTNR is a research framework supporting seamless human-robot collaboration. Building on our research with Habitat, we’re open sourcing a large-scale benchmark, dataset and large planning model that we hope will enable the community to effectively train social robots.
-
Top arXiv Paper Trends: DeepSeek and Test-Time Scaling
By
–
Trends of top arXiv papers this past month Showing views on alphaXiv over time for
– DeepSeek V3
– DeepSeek R1
– s1: Simple test-time scaling -
R1 Explicit Preference Reward Training Purpose Discussion
By
–
Why not? I mean R1 even had an explicit preference reward at the end of training; I think that's there exactly for that purpose.