Great set of slides by my Gemini colleague @FeinbergVlad on scaling considerations in large language models, addressing the fact that the classic "scaling law" work does not take into account inference cost (!), distillation, learning rate schedules, etc. https://
vladfeinberg.com/2025/04/24/gem
ini-flash-pretraining.html
…
RESEARCH
-

Scaling Laws in LLMs: Beyond Classic Models and Inference Costs
By
–
-
AI Advancement Expands Opportunities for Decades Ahead
By
–
not true, having ai get better increases the order of magnitude of opportunity, we have many decades of this happening before any change might change
-
Creating Real Value: Beyond Vibes in Problem Solving
By
–
sure, but the point is to create value, vibes can be part of it, but at the core solving hard problems is never easy
-
An Introduction to Innovation by Sarina Dibazar
By
–
An Introduction to Innovation by Sarina Dibazar pic.twitter.com/XTU6vbxwPf
— Dr. Hassan Rashidi (@DrHassanRashidi) 27 avril 2025An Introduction to Innovation by Sarina Dibazar
-

General-Reasoner: Reinforcement Learning Boosts LLM Reasoning
By
–
9. General-Reasoner General-Reasoner is a reinforcement learning approach that boosts LLM reasoning across diverse domains by using a 230K-question dataset and a model-based verifier trained to understand semantics beyond exact matches.
-
Tiny Reasoning Models: 1.5B Parameter Tina Family
By
–
10. Tiny Reasoning Models Tina is a family of 1.5B parameter reasoning models trained using LoRA-based reinforcement learning (RL) to achieve high reasoning accuracy at very low cost.
-

Assessing LLM Goal-Directedness: New Evaluation Framework
By
–
8. Evaluate the Goal-Directedness of LLMs Introduces a new framework to assess whether LLMs use their capabilities effectively toward achieving given goals.
-

Tic Tac Toe Complexity and Benchmark Performance Discussion
By
–
Hey @timzaman can you rerun your bench maybe. Personally btw I think tic tac toe is secretly relatively hard. There are 8 lines to check. Each a medium tricky indexing op. And if you want to play, you have to roll it out a bit. Humans find it easy to play because they use their
-

Stanford’s Essential 1-Hour Lecture on Agentic AI Systems
By
–
Stanford released an 1 hour lecture on Agentic AI and is a must-watch for every AI enthusiast! This 1-hour lecture will teach you everything you need to know to start building with agentic LLMs, including reflection, planning, tool use, and iterative reasoning.
-

Scientific Method Enhances Technology Assessment and Innovation Decision-Making
By
–
Scientific thinking is a powerful tool for assessing technology. A structured approach reduces bias, enhances decision-making, and ensures innovations deliver real value, preventing wasted investments in solutions that don't truly fit. Microblog by @antgrasso #ScientificMethod