May 2026

pipeline is really slow – consulting [D]

⋅

Hi, after a long debugging process and many discussions, I wanted to ask for advice from people who may have encountered similar training bottlenecks. My goal is imitation learning for robotics. Model / Pipeline Observation space: 4 RGB robot cameras image resolution: 128x128x3 small vector of robot joint velocities (14 dims) Pipeline: Shared ResNet18 encoder processes each image Each image embedding dimension is 128 Final input to policy: 4 * 128 image embedding concatenated with 14-dim state vector […]

Ver mais

Liked Liked
I'm a bit shocked that this finally worked

I’m a bit shocked that this finally worked

⋅

So, I’ve been working on RL models for the last few years as my hobby, did I finally get my trader to be profitable? I’ve been on a multi-year optimization journey, from model architecture, reward shaping journey, and endless training loops and unintentional curriculum learning … Happy to answer any question about my journey so far, the architecture, or the endless optimizations and tricks I needed to do to get to something that’s actually profitable … To give […]

Ver mais

Liked Liked
Hilbert transform as an infinite matrix

Hilbert transform as an infinite matrix

⋅

The previous post linked to a post I wrote a few years ago about the Hilbert transform and Fourier series. That post says that if the Fourier series of a function is then the Fourier series of its Hilbert transform is When I looked back at that post I thought about how if you thought of the Fourier coefficients as elements of an infinite vector then the Hilbert transform can be represented as multiplying by an infinite block […]

Ver mais

Liked Liked
Governing A.I. in the Enterprise

Governing A.I. in the Enterprise

⋅

Notes from the Security Architect’s Chair When we stood up our Security Architecture function, we were working across several distinct operating divisions, each with its own technology DNA. I expected AI governance to be one workstream among many. Instead, it became the workstream that is reshaping many of the others. Not because AI is uniquely dangerous (every technology wave brings its own risk), but because it bends the assumptions our control frameworks were built on. Identity, blast radius, […]

Ver mais

Liked Liked

Real and imaginary parts

⋅

The previous post announced some notes I wrote up based on an article by Henry Baker implementing functions of a complex variable in terms of functions of a real variable. That is, it finds functions u(x, y) and v(x, y) such that f(x + iy) = u(x, y) + i v(x, y) where x, y, u, and v are all real-valued. Not only that, but if f is an elementary function, so are u and v. Here “elementary” has a technical meaning, but essentially it […]

Ver mais

Liked Liked
I Added 50 Tools to My AI Agent, and It Got Dumber.

I Added 50 Tools to My AI Agent, and It Got Dumber.

⋅

I built ToolGate, an MCP proxy that uses semantic search to filter AI agent tools. It reduces context bloat by 60%+ while maintaining 85%+ precision on synthetic benchmarks. Open source and ready to try. Generated with Gemini. ToolGate — a smart MCP/tool proxy The Problem: When More Tools = Less Intelligence I build (and help teams build) agentic AI systems, and they all use many tools. Each agent has access to different tool sets through MCP servers. By the time I hit 6+ […]

Ver mais

Liked Liked
Karpathy Said Kill Prompt Engineering. Here’s What He Actually Meant.

Karpathy Said Kill Prompt Engineering. Here’s What He Actually Meant.

⋅

The internet declared a funeral. The body wasn’t dead. Last June, Andrej Karpathy — former Tesla AI director, OpenAI co-founder, and one of the few people in the field whose offhand posts routinely restructure how thousands of developers work — posted something that landed like a scheduled demolition. “+1 for ‘context engineering’ over ‘prompt engineering,’” he wrote on X. “People associate prompts with short task descriptions you’d give an LLM in your day-to-day use. When in every industrial-strength LLM app, context engineering is […]

Ver mais

Liked Liked

Load More

↑

Digitado © 2026