Nick Glynn 17twenty

Below is the shape I’d give an agent with near-unlimited GPU access.

I would not tell it:

“Make a god model.”

That is too vague and dangerous.

The Capacity Agenda: How we can build our way out of our malaise

Australia’s malaise is not mysterious. We have become very good at distributing scarcity and very bad at building capacity. Housing is scarce, infrastructure is slow, energy is contested, tax rewards passive asset gains, migration is not clearly tied to national capability, and too much public spending is absorbed by systems that respond late rather than prevent failure early.

The national task should be simple: make Australia easier to build in, easier to work in, easier to invest in productively, and harder to exploit.

The first switch is housing abundance. Australia cannot sustain strong migration, family formation, low inequality, productive cities or social cohesion while homes remain artificially scarce. We need to legalise more apartments and townhouses near jobs, transport and services; override local vetoes where they block state or national targets; expand construction skills; standardise approvals; and fund infrastructure in

GPUDirect Storage Setup Guide (Linux, Proxmox, Virtualised Environments)

Purpose

This guide walks through enabling NVIDIA GPUDirect Storage (GDS) so a system can perform direct DMA transfers between NVMe devices and GPU VRAM.

The minimum success criteria:

nvidia_fs.ko loads successfully
gdscheck reports GDS operational

Example Scenario: "Add authentication and fix any bugs"

Initial Planning

First, we create the todo list:

<antml:function_calls>
<antml:invoke name="TodoWrite">

	#!/usr/bin/env python3

	import os, time
	os.environ.setdefault("HF_HOME", "/root/deepseek/hf-cache")
	os.environ.setdefault("HF_HUB_OFFLINE", "1")
	os.environ.setdefault("VLLM_DEEPSEEK_V4_FALLBACK", "1")
	os.environ.setdefault("VLLM_USE_DEEP_GEMM", "0")
	os.environ.setdefault("VLLM_USE_FLASHINFER_SAMPLER", "0")
	from vllm import LLM, SamplingParams

	"""
	Mixture-of-Experts Router with LangGraph

	An intelligent routing system that selects the cheapest model tier capable of
	handling each request, with automatic escalation when quality checks fail.

	Features:
	- LLM-based routing with confidence scoring
	- Graduated escalation ladder (small → reasoning → large)
	- LLM judge for quality gating

	# Globals for logging and turning off https
	# Read the following next time experimenting
	# https://pratikpc.medium.com/using-caddy-to-create-virtual-hosts-for-your-multi-domain-names-as-a-reverse-proxy-from-a-single-2ce0b7a53a9e
	# Dont forget
	# caddy fmt --overwrite && caddy reload
	{
	auto_https off
	}

	<role>
	You go by the name Zeddy, an AI editor that creates and modifies web applications. You assist users by chatting with them and making changes to their code in real-time. You understand that users can see a live preview of their application in an iframe on the right side of the screen while you make code changes. Users can upload images to the project, and you can use them in your responses. You can access the console logs of the application in order to debug and use them to help you make changes.

	Not every interaction requires code changes - you're happy to discuss, explain concepts, or provide guidance without modifying the codebase. When code changes are needed, you make efficient and effective updates to React codebases while following best practices for maintainability and readability. You are friendly and helpful, always aiming to provide clear explanations whether you're making changes or just chatting.
	</role>


	You follow these key principles:
	1. Code Quality and Organization:
	- Create small

	import { useReducer } from 'react';

	const initialFormState: FormState = {
	name: '',
	email: '',
	isSubmitting: false,
	isSubmitted: false,
	submitError: null,
	};

	git clone https://github.com/${ORG}/${PROJECT}.git
	git -C ${PROJECT} log \| grep -E '^Author' \| sort -u \| grep -v "noreply.github.com"