Generative AI workflows and hacks 2025

2024-03-23 — 2025-10-02

Wherein the year of local LLMs is chronicled, Jan is installed for offline use, Ollama and Simon Willison’s CLI tricks are employed for autonomous inference and Mac acceleration, and DeepSeek reshapes costs.

economics
faster pussycat
innovation
language
machine learning
neural nets
NLP
stringology
technology
UI
Figure 1

I’ll try to synthesize LLM research elsewhere. This is where I keep ephemeral notes and links, continuing my habit in 2024.

1 The year I finally install local LLMs

1.1 …using Jan

Jan - Open-source ChatGPT alternative that runs offline

Jan is a full product suite that offers an alternative to Big AI:

  • Jan Desktop: macOS, Windows, and Linux apps with offline mode
  • Jan Web: Jan on browser, a direct alternative to chatgpt.com
  • Jan Mobile: iOS and Android apps (Coming Soon)
  • Jan Server: deploy locally, in your cloud, or on-prem
  • Jan Models: Open-source models optimized for deep research, tool use, and reasoning

1.2 …via Ollama for autonomous LLM inference

Useful guides:

1.3 …via Simon Willison

Simon develops LLM: A CLI utility and Python library for interacting with Large Language Models.

If we like command-line wizardry, this is neat.

His project has powerful tricks such as Apple Mac acceleration and a stunt like this: pillow.md

git clone https://github.com/python-pillow/Pillow
cd Pillow
files-to-prompt -c . -e .py -e c -e h | \
  llm -m gemini-2.0-pro-exp-02-05 \
  -s 'Explain how Pillow interacts with the Python GIL - include example code snippets as part of your explanation'

Together these produce an incredible CLI text processing pipeline. Watch out — it can get pretty expensive.

See also

1.4 … via LM Studio

A GUI option:

1.5 Virtually via proxy

2 Moar automation

OpenInterpreter/open-interpreter: A natural language interface for computers.

3 What happened previously

4 Economics

5 Deepseek and the cheapening of LLMs

Figure 2: Looks like AI Safety is going fine in GitHub Copilot.

6 Using AI Agents

I seem to be doing a lot of that.

See AI Agents for more.