iiinova — where your digital self keeps evolving

Reposted by Tilmon Edwards (Dirigo)

rev. howard arson @theophite.bsky.social · 7h

it is possible to be propagandized by things which are true.

Tilmon Edwards (Dirigo) @tilmonedwards.com · 7h

Guy getting dopamine from going on worldstarhiphop to watch other guys get their asses beat: "it's called *bearing witness*"

0734

Reposted by Tilmon Edwards (Dirigo)

Ed @ed3d.net · 29/09/2026

yeah. more to the point, if you start using stuff like facebook's agent or this thing, you deserve it when you cook yourself.

3676

Tilmon Edwards (Dirigo) @tilmonedwards.com · 10h

If your main objection to AI is the concentration of power in the tech industry, you SHOULD use their inference with products that are NOT owned by them! Force the AI labs to be a low-margin utility. Ideally, someday, a nationalized utility owned by the public and operated by a competent government.

000

Tilmon Edwards (Dirigo) @tilmonedwards.com · 10h

But, business gotta business, and so that means that the action is in the ecosystem around inference. Inference itself is a commodity utility, what do you *do* with it? It's harness and context engineering.

100

Tilmon Edwards (Dirigo) @tilmonedwards.com · 10h

They tried to make inference the lock-in thing, and the technology just structurally resists vendor lock-in and monopolizing. It's real "information wants to be free" kind of thing. The piracy machine turns out to be extremely susceptible to piracy.

280

Reposted by Tilmon Edwards (Dirigo)

Ed @ed3d.net · 29/09/2026

observing for no particular reason that you should be treating inference providers as dumb pipes because their only way to ramp up the extraction machine on you is to sell a product that you can't get out of

216416

Tilmon Edwards (Dirigo) @tilmonedwards.com · 29/09/2026

I submit that these are not men, they are boys, overgrown children who never outgrew their tantrums. Pathetic, actually.

130

Tilmon Edwards (Dirigo) @tilmonedwards.com · 29/09/2026

Plus there’s all the corporate leaders going “hack meeeee!! upset our foooorrrmmmm!!”

010

Tilmon Edwards (Dirigo) @tilmonedwards.com · 25/09/2026

a full plate of fish and chips, extra tartar sauce. enough to share so that they know I am generous, but I do not offer any so that they know I am also powerful

061

Tilmon Edwards (Dirigo) @tilmonedwards.com · 25/09/2026

I kinda love it. Not that I want every building to be like this, but once in a while, we can have a big mysterious obelisk as a treat

130

Tilmon Edwards (Dirigo) @tilmonedwards.com · 22/09/2026

Kathryn Tewson is a legendary, S-tier internet troll. Outstanding work, hats off.

1221

Reposted by Tilmon Edwards (Dirigo)

Tilmon Edwards (Dirigo) @tilmonedwards.com · 21/09/2026

You're telling me that I'm not highly paid because I'm extremely smart and cool and fun to be around? It's a purely transactional relationship based on my ability to deliver solutions that allow the company to hire fewer people? And my employer does not care if those solutions are computer programs?

1155

Tilmon Edwards (Dirigo) @tilmonedwards.com · 21/09/2026

1155

Tilmon Edwards (Dirigo) @tilmonedwards.com · 21/09/2026

I would ask my fellow programmers: did you get into the business of automating labor or not?

191

Tilmon Edwards (Dirigo) @tilmonedwards.com · 21/09/2026

I have less than zero sympathy for “it’s not fair” for this reason. It is actually the most fair, and it’s terribly pathetic to be whining about it now.

150

Tilmon Edwards (Dirigo) @tilmonedwards.com · 21/09/2026

In my experience, that type of work happens most often in places where you have to interface with institutions. Institutions like things that are scheduled, automated, predictable, consistent. Institutions like to be able to plan.

000

Tilmon Edwards (Dirigo) @tilmonedwards.com · 21/09/2026

I wasn't careful with my language, I assumed drugery to be unimportant. If you consider drugery to be important, that's fine, I'm not trying to tell anyone what parts of their job are important. If there are tedious, unimportant parts of your job, those are the things that can be automated.

110

Tilmon Edwards (Dirigo) @tilmonedwards.com · 21/09/2026

It depends though, doesn't it? I see from your profile that you work with music, drugery is important there. It's important in education. In production systems I work on, it's a cost inefficiency that is carried by consumers as higher prices, and sometimes it's a risk that leads to safety problems.

100

Tilmon Edwards (Dirigo) @tilmonedwards.com · 21/09/2026

Models and inference tokens are in the same category as line shafts or electricity, or a river over a water wheel. They‘re a utility and a source of energy, but the value of them comes entirely in how they’re directed to do productive work.

100

Tilmon Edwards (Dirigo) @tilmonedwards.com · 21/09/2026

“Creating more value out of a tool” isn’t what I’m saying. I’m saying that tokens are not a finished good, they are an industrial input. Inference isn’t even a complete tool yet. Actual factories barely even exist. Models are not the factories, agents and harnesses are the factories.

100

Tilmon Edwards (Dirigo) @tilmonedwards.com · 21/09/2026

I’ve mentioned this in a few other places and replies, this is probably my most direct response. Basically, I just don’t think that owning the inference is going to be where the value accrues. I think it’s hugely capital intensive and vanishingly small margins. The value is in how you use the tokens

110

Tilmon Edwards (Dirigo) @tilmonedwards.com · 21/09/2026

This feels related. Tweaking the weights in the direction of watermarking also tweaks them away from accurately modeling language. Maybe it’s worth it, idk, but I current hypothesize that a watermarking model will be outperformed by a non-watermarking model.

030

Tilmon Edwards (Dirigo) @tilmonedwards.com · 21/09/2026

It seems to me that when industrialization comes for your labor, you are, at least for a few moments, well-positioned to be the subject matter expert in the new industry. That’s what I was trying to say in my original post, seize it!

010

Tilmon Edwards (Dirigo) @tilmonedwards.com · 21/09/2026

I also think language models shake out to look at lot more like a fungible utility, long term. Frontier labs have not been able to establish a moat. Open models are just as good, sometimes better. Unlike VC SaaS consolidation, there is actually competition in this market. That’s good! Previously:

130

Tilmon Edwards (Dirigo) @tilmonedwards.com · 21/09/2026

SaaS is a cottage industry, I agree. And I say this as a programmer, I don’t think industrialization of that industry devalues the labor of the people who built it, and I do think this will have significant benefits to the rest of the economy. The SaaS economy has been rent seeking for a long time.

100

Tilmon Edwards (Dirigo) @tilmonedwards.com · 21/09/2026

Thanks for sticking with me, by the way. I feel like this deserves a longer form blog post and this is helping me understand my own thesis better.

100

Tilmon Edwards (Dirigo) @tilmonedwards.com · 21/09/2026

SaaS business, yes. brain, no, because your brain can’t scale beyond your own limited supply of attention. That‘s what makes it a means of production and not just labor. Your brain is labor. The repeatable and scalable knowledge work process you build with your brain is the means of production.

100

Tilmon Edwards (Dirigo) @tilmonedwards.com · 21/09/2026

I don’t think it’s unreasonable to say that one can own the means of production without also needing to own the thing that powers it, be it electricity or tokens or diesel or something else. But several people have said essentially that. Owning the means of production includes operational expenses.

100

Tilmon Edwards (Dirigo) @tilmonedwards.com · 21/09/2026

I don’t think it’s inconsequential at all. I said “means of production” and then several people replied with the same misunderstanding that buying tokens from a vendor means you are renting the means of production. This is the whole thing I’m trying to say.

110

Tilmon Edwards (Dirigo) @tilmonedwards.com · 21/09/2026

Here, I drew a picture If you own the means of production, you still have to buy the inputs that will become your product. The means of production is the harness, not the model.

000

Tilmon Edwards (Dirigo) @tilmonedwards.com · 21/09/2026

I know, and you keep thinking that the model is the means of production and I’m trying to tell you in every way that I can think of that the model is not the means of production. They are the industrial input. The model doesn’t matter, it’s fungible. What you do with the tokens is what matters.

100

Tilmon Edwards (Dirigo) @tilmonedwards.com · 21/09/2026

Tokens are the output of the model. The model is the forest. The forest outputs logs, the model outputs tokens. Tokens are the input to your process. Like a sawmill, you take tokens (logs) as input to produce something else. The process you build is your means of production, tokens are your inputs.

110

Tilmon Edwards (Dirigo) @tilmonedwards.com · 21/09/2026

Logs are the inputs to a lumber mill. You get boards as output. I really don’t think you’re following the analogy.

220

Tilmon Edwards (Dirigo) @tilmonedwards.com · 21/09/2026

You are misunderstanding my analogy and I don’t think you know what a token is. Tokens are the model’s output. Tokens are the logs. The method you use to direct the tokens toward productive uses is the factory.

130

Tilmon Edwards (Dirigo) @tilmonedwards.com · 21/09/2026

Tokens are the logs, not the mill.

000

Tilmon Edwards (Dirigo) @tilmonedwards.com · 21/09/2026

The owner of a lumber mill still has to buy logs.

210

Tilmon Edwards (Dirigo) @tilmonedwards.com · 21/09/2026

The tokens are not the means of production. They are the raw materials of production. The way you use your brain to direct them toward productive uses is the means of production. This is a common misunderstanding.

150

Tilmon Edwards (Dirigo) @tilmonedwards.com · 21/09/2026

I don’t know if it will work well, I think it will, but I certainly wouldn’t outsource my grading or feedback to the robot.

010

Tilmon Edwards (Dirigo) @tilmonedwards.com · 21/09/2026

This week I’ll be using AI for the first time in an educational environment. I’m teaching programming for robotics to high schoolers. I’ve used it to create a robot program so they can see something functional on day one, and over the course of several weeks we will take it apart.

110