4× RTX PRO 6000 Blackwell on Water, and the One Card That Wouldn't Behave
Converting four RTX PRO 6000 Blackwell cards to waterblocks, finding a VRM choke loose on the workbench, and getting back to 41k tok/s.
Life and AI
Recent content on Life and AI
Latest posts
Many Contexts Is All You Need
RLM — Recursive Language Models. The model calls itself with smaller problems and fresh working memory. Here's why this changes everything.
Test-Time Training for LLMs: Models Should Keep Learning After Deployment
Pretraining taught us that neural networks can compress massive amounts of data into weights. But once we deploy an LLM, we usually stop updating those weights completely. The model becomes frozen — it reads new inputs but never learns from them. Test-time training asks a more ambitious question: what if the model kept learning while it was being used? TTT-E2E is one practical answer. It lets a…
Teaching Qwen3-4B to Trade: From Hold-Collapse to +9.4% Returns
How I fine-tuned a 4B parameter LLM into a trading decision model using a 5-stage SFT waterfall and weighted CCE loss, then why I'm pivoting to GRPO reinforcement learning.
MCP Compact: Keep Agent Context Lean
A proxy that trims noisy tool outputs so MCP agents stay on track.
All You Need is 4x 4090 GPUs to Train Your Own Model
How I built an ML rig for training LLMs locally, exploring hardware choices, setup tricks, and lessons learned along the way.
Defining AGI
A thoughtful exploration of Artificial General Intelligence (AGI) through three fundamental concepts.
Embarking on My Journey into LLM
An introduction to my experiences and explorations in the world of Large Language Models.
(untitled)
Sabareesh Subramani https://sabareesh.com | LinkedIn | hello@sabareesh.com Summary ML researcher and CTO exploring how language models can learn, adapt, and make decisions in real-time. Current research spans three areas: training LLMs that keep learning during inference (TTT-E2E with second-order meta-gradients, 4.6–6.1% perplexity gains on Qwen3-4B), teaching LLMs to trade autonomously via…