Post

Are we supposed to be against machine learning and other subsets of AI now? I don’t get all the rage

The challenge with Attie for the builder community isn't the use of AI. It's the question of whether Bluesky begins absorbing/cannibalizing existing ecosystems within the Atmosphere.

I have no idea what’s going to happen to the application developer ecosystem because of AI coding but I would say the call to action should be for all of us to start utilizing this tech in our work

I’m going to be starting a virtual meetup for agentic personal computing so we can all row in the same direction I don’t care if it’s Attie, I care if it’s the atmosphere

As I said elsewhere I am basically radicalized about this. We fight now for personal computing and personal agency or we lose another decade to closed clouds We push now for an open internet and open models. Nobody is going to hand it to us because they’re nice

I can see the reasons behind this push, whether I agree with them or not. But then, why not go all-in? Why not make it a fundamental requirement that Attie *only* use open models. From my understanding, the current system is to use Claude. Why not push the limits from day 1?

Agree with Paul... It's worth acknowledging that most users care about outcomes, not process; using the best quality model is the only way. Also the "api" for LLMs is literally just prose. There has never been SW where it's easier to switch horses. LLMs are lock-in-proof in a way web 2.0 never was.

I think the risk is that the open models, and often unintentionally, become an excuse to keep using the proprietary hosted models and all the stuff that comes along with them. That they exist is a foil to a lot of complaints about the AI industry, despite few really using them

In other words, you say you're radicalized. Great! Be radical about it! Take a stand!

Outcomes matter! Maintaining optionality does too! Giving people worse outcomes solely on principle doesn't get you or them anything.

I am curious how you define "open" models, and I'm really not trying to be pedantic here. Open weight? Open dataset? Are you thinking about the implications of building on what courts may determine is stolen work?

Indeed I think a lot of people have been saying "the Supreme Court decided AIgen output is in the public domain!" but that's not true, they declined to hear the case. One reason (of several) I don't include AIgen in my projects; international legal situation unclear AFAICT.

The case is about art that predates consumer LLMs by almost a decade and has no human intervention to create it, & the human creator of the algorithm wanted to assign copyright to the software itself. It is unlikely applicable to modern generative AI at all, and definitely not code generated by AI.

Closed models will not stop improving either. This is Linux vs macOS… or accepting the discomfort of rail travel over airplanes. Being radical in theory doesn’t count.

And shipping software nobody likes doesn’t affect change.

Sure… but.. What you just said is a pragmatic stance, i would say. “Radical” usually implies a strong commitment, I think..?