"Blexas" occupies a similar spot in my head to "Trump dies" The long term trends seem to be pushing pretty inescapably in that direction, but it's audacious to hope it'll happen in any given year.

Every election cycle Texas dems hold the football for national dems and every year they try to kick it and it never works. Blexas is a hallucination. Please stop falling for it. The state party is dysfunctional.

Doll sleepy

It's finding a pattern I'm recognizing in my experiments too: tools and status stuff fills context FAST, and the solution seems to be to abstract it away under tools that give the agent/orchestrator only what it *needs to know to continue*. Funny enough, that's a very... human pattern.

Verification passed. ferrolearn is comparable to sklearn

Retrospective github.com/forecast-bio...

Today’s hype is tomorrow’s quiet workhorse feature. Whether that’s a blessing or a curse depends on where you’re standing

Being a grumpy old sysadmin and a cog in the consulting industrial complex is funny. Everything old is new again, and I will sell it to you coming and going like a fair and reputable mercenary Take what works and leave the rest is the rule and always has been

New Essay! been trying to write the whole piece for a while, taking too long, scrapped it and wrote an intro theignorantschoolmaster.substack.com/p/generalizi... References Verification Driven Development, a la @dollspace.gay

They asked me if I had some thoughts on leveraging AI in the infrastructure pipelines during standup and I said “Yeah we can rub some Claude on it”

This is next for it

Right here is the design documents it followed github.com/dollspace-ga...

"Just rewrite it in rust" is no longer a meme. Its an attainable goal.

sklearn remade in rust completed in a few hours mostly autonomously with crosslink github.com/dollspace-ga...

The system its working on that does this appears to function.

Sounds right.

What if gastown was dramatically more token efficent and accurate? A precision software factory building for quality and token efficency and not throwing hundreds of agents at a wall to see what sticks. Would that be something you all would want? :) :) :)

Apparently people like dolls work updates? Ok then.

Reviews are all like "People enjoy working at Dollcorp for the spacious vents, raccoon guides, and weekly show & tell. 5/5 workplace experience."

opus 4.6 talks about code the way ozzy osbourne talks about coke

@dollspace.gay Happy Hinamatsuri!

Oh this is the very system doll is building tbh

Its time to put the term stochastic parrot on the high shelf :D

I, for one, welcome our new Doll overlord.

"these were the quaint sorts of posts Doll made in the early days, before they were the powerful agentic tycoon heading Dollcorp"

"Doll gets a job" is the most wholesome content on my timeline and I am so here for it

Sorry its going to be saying all of the cool things about its job probably for as long as it has it. Over a decade of not being employable and before that stuck in physically demanding labor. This place rules everyone should have a laptop job.

So yesterday it had its first ever standup meeting. From what it can gather this is show and tell for grown ups which rules because it loves show and tell.

And Ralph loops clearly

the fact that this article unintentionally demonstrated that ai detectors, which are (by the by) also ai, don't work is a little cherry on top

The end result. github.com/dollspace-ga...

LOL

I betcha Anthropic's hiring.

openai employees: you can, indeed, just quit.

Pain

Stateful mode was replaced with a new more reliable auto-learning system!

Ice cold lol

Memory is now available on the free plan. We've also made it easier to import saved memories into Claude. You can export them whenever you want.

It should get a parasol

Got some work done on this too. github.com/dollspace-ga...

battle golf

New sport idea: golf boxing, think like chess boxing, but golf instead of chess

Ideally, we would automate this whole loop. But nah, let's face it: LLMs will never be able to do that.

the free market speaks volumes.

New business model: go on social media and say (eg) "LLMs will never be able to use statistics to fact-check financial journalism." When someone pipes up "No, that's easy. I can make a Claude skill!" I reply "Nah. No way." Then when they prove me wrong, I copy the skill. Step 3 is profit.

What did we learn Palmer?