RSSAmplifier

Blog

mdda

Blog{mdda} : thought process / brain dump

mdda.net ↗RSS feed ↗389 posts

Latest posts

GDG I/O Extended Singapore : New Gemma Models

Local Models & Self-Evolving Harnesses

DSTA BrainHack event - Invited Talk

External Keypad for OBS Controls

AI Engineer Gemini Event

Running ~30B MoE models on an old GTX 1080

Robots : Reasoning and Acting

Personal Agents - Cloud and Edge

Signals Between the Tokens

Learning to Think, Not Learning to Solve

NeurIPS Roundup

Sizing gemma3 for TPUv5-n

Practical TPU VM on GCP

Reinforcement Learning for Long-Horizon Multi-Turn Search Agents

Evolving Reasoning

DevFest SG Talk: Lessons from Reinforcement Learning

Neural Assets and World Models

IPhO Gold using Agentic Gemini

Agent Efficiency, Memory and Confidence

A Reasoning-Based Approach to Cryptic Crossword Clue Solving

GPU Kernel Scientist : An LLM-Driven Framework for Iterative Kernel Optimization

Self-Improving Agents

Evolving GPU Kernels

Latent Space Reasoning

Diffusion LLMs

Getting to Aha (notes)

Getting to Aha

NeurIPS recap and SOTA

System 2 - Training for Test-time compute

Slack to Mattermost on New Fedora server

o1 - Are the curves awesome?

Fedora 40 XFCE Live modifications

Install cuda-12.6 on Fedora 40

Upgrade to Fedora 40 with cuda-12.4

Proving that Cryptic Crossword Clue Answers are Correct

Small LM Agents

Cryptic Proof Checking

Upgrade to Fedora 38 with cuda-12.4

Long Context Transformers

DSPy - Gemma Sprint

Gemini 1.5 One Million Sprint

DSPy - Gemini and Gemma

Gemini 1 million

Research from NeurIPS

Enable bluetooth management via GUI on Fedora

DevFest SG Talk: No-code Prototyping with Flowise

What's Next in RAG

Converting a Presentation Transcript to Speaker Notes

LLMs and Reasoning

Keras Core