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

• Replit agent for accelerated programming

• Flux for image generation

• Video generation - luma, runway ml, pika labs?

• neural voice generation - elevenlabs

• music - suno.ai and udio.ai

• recent 3d model generation model from stable diffusion