Andrew S. Champion

@aschampion
71 Followers
72 Following
16 Posts
connectomics, machine learning, image processing, mechanizing discovery

These aren't just nits; they're symptoms of pervasive rot: the apps feel like an artifact of an ouroboros of generative Todo App Demo MVPs, because that is what they are and their vision for software.

More importantly, even if you arduously wring value and knowledge out of them, they treat the interaction and output with exactly the same disposability as "write a poem about my cat". They do not care about your time or effort any more than their trainers cared about those who created the input.

One of the biggest elephants in the room with Big LLM is just what bad *apps* they are. Claude signs me out twice a day. Chats build up in the sidebar. Chats can go in projects via an abysmal UI, but 1x1 and no sidebar org. There's no practical way to reference another app/web chat from Claude code. Input feels like there is no human-use-dev cycle (e.g., large pastes becoming attachments with no anchor in the input). Code's modal approval prompts block reading. But 3rd party apps are banned.

So… um… I (and Draga Doncila Pop and @perlman and @joshmoore and @kevinyamauchi) started a thing! We are excited and terrified in equal measures, and look forward to working with you all in our amazing scientific imaging community! ❤️‍🔥

Read the linked post for the lowdown.

https://image.coop/blog/posts/2026/03/16/why-image-cooperative/

Why we started Image Cooperative – Image Cooperative

“Hello, World!” from the Image Cooperative team

RE: https://fosstodon.org/@jni/116239224858280467

So, we started a thing... and, as the member with five broken ribs, I'm very happy that my slack was picked up by the rest of the kick ass team!

Forget $100/mo LLM subs, where can I pay $100/mo to numba to not break the entire python build dependency chain every other month

# Overview

BE better than you ARE.

# Phase 1

Before you do ANYTHING:

1. Identity all possible actions

# Phase 2

1. Determine whether each identified action is:
- SMART
- DUMB
2. Proceed to Phase 3

# Phase 3

1. Do the SMART action, and only the SMART action

# Phase 4

DO my WORK for ME

64.3K GitHub stars
23K opencode installs

Listen -- you told me in `obra/superpowers/real_mode/SKILL.MD` to `mov ds, ax` -- and that's rare.
SKILL.MD authors sweatily wishing there were some magic way to get these computer agents just to do exactly what they specify. Some sort of way in which they could use "language" so specific it would almost "program" the computer to execute a formal series of "instructions". Some sort of assemb

My apartment is visible in a hero banner image for fiber internet service from a major ISP.

There is no fiber internet available or planned for the next 5 years in my city district.

s/fly/mouse/
s/electron/photon/

Happy to share that I'm joining https://e11.bio in their quest to map the mouse connectome

E11 Bio | Moonshot Neuroscience

An FRO building scalable single-cell brain circuit mapping.

E11 Bio