Key Components of My AI and Technology Stack
As of September 9, 2024
These are the core components of my tech stack as of today. Will update as things change.
Hardware: Apple ecosystem
- M1 Mac
iPhone
iPad
Sync across Apple ecosystem: iCloud
Tasks and notes: Obsidian
Create an iCloud folder for an Obsidian vault (or for each vault), and point to it from all devices, then use it
Use Tasks core plugin to manage tasks
Use Link Embed community plugin to embed link previews
Create daily notes
Use task fields to manage tasks across all notes
Use hashtags to tag different items and find them via search
Use internal links (wiki style double square brackets
[[...]]) to create other pages for different projects and edit them.
Web: Chrome
Chrome - mainly because it has multiple profiles for work accounts at startups and personal profiles
Bookmarking - copy url and paste to Obsidian as embedded links with a hashtag for reading list, etc.
Web search with AI
Bing search with Copilot - the best search first AI as of now - change default search engine in Chrome to bing
Perplexity - for when bing search isn't quite what we wanted
Programming
cursor editor with claude-3.5-sonnet - AI first editor - fork of VS Code, but much more AI programming friendly
language - python/gradio, javascript+html+css
metadata format - json
use python json module
create a custom JSON Encoder to encode nested objects into json
output data format - pdf
- reportlab python library with pdfgen and platypus submodules
cloud sync and backup for large files in local projects: create a local folder on mac, set it to stream with google drive - the files are not saved locally
AI workflows with GPUs - colab paid account for A100 GPUs (or when not available, high RAM T4 GPUs) for GPU intensive jobs (fine tuning, large model inference, multi-AI -agent workflows)
Local AI - AI with M1 mac - usually slower than Nvidia GPUs, but works for some long running unattended jobs.
AI chat
Almost always start with Claude AI (with 3.5 sonnet) - the ecosystem is pretty good
For current knowledge based chats, try ChatGPT with ChatGPT-4o
For search first AI, use bing search and copilot
Perplexity works for some search use cases, but has been hit or miss for me
To assimilate links on X/twitter, use grok (requires paid X subscription)
AI chat automation
Claude for most things
Claude with search APIs if search is needed
Images
AI image generation
Stable diffusion XL in Automatic1111's stable-diffusion-webui
Other stable diffusion models and controlnets and loras from civitai
Rejected: midjourney, because it has no automation/API
Rejected: DallE3 - because it does not provide as high quality or as much control as sdxl (if you use the right prompts and config)
AI image generation automation
local huggingface diffusers on M1 mac to use stable diffusion xl
local calls to Automatic1111 stable-diffusion-webui in api mode
can also run on windows with Nvidia GPU (I use GeForce RTX 3080)
be careful to fix any saved paths for forward and backslashes
Lay out images in specific sizes - Canva
Tools worth evaluating
I haven't tried them (much or at all, or recently), but have a good feeling about them
Some tools have hosted or local API, others do not - I prefer to use local tools for most things for maximum control