Charles Connaughton

Meta’s internal memo said the political tumult in the United States was good timing for the feature’s release. “We will launch during a dynamic political environment where many civil society groups that we would expect to attack us would have their resources focused on other concerns,” according to the document from Meta’s Reality Labs, which works on hardware including smart glasses.

this is comically evil. brazenly anti-social. just absolute black-pilled nihilism. we will not have a republic, we will not be free, until we regulate these companies to the point where — at a bare minimum — they're too afraid to put stuff like this down on paper.

www.nytimes.com/2026/02/13/t...

Many of the most highly compensated tasks performed by humans are things humans are inherently pretty bad at.

The things humans writ large are good at are things everyone can do.

I often wonder how many times different researchers discover the same statistically null findings, over and over again, because there is no formal mechanism to document and share those results.

Real risks of having a single "anti-AI" coalition (covering jobs, kids, slop, existential risk, environment) that can only agree on a full halt to AI as a remedy. Not only is that unlikely, but it undermines the desire to make policies that channel AI to good uses or that mitigate specific harms.

The popular narrative worries a lot about a singular, centralized intelligence, but scaling laws and budget constraints don't disappear even in superintelligence worlds.

A population ecology perspective is very likely to be much more helpful for understanding where we are going.

Mhm. In a heterogeneous environment with a budget constraint, you get a whole ecosystem of niche specialists, not a generalist monoculture.

Feels like we're seeing some flavor of the no free lunch theorem - given a budget constraint, you can tune a model to perform better at some subset of problems at the expense of being worse at a different subset.

For absolutely no reason, let me remind people of this banger of a paper by @caroartc.bsky.social

doi.org/10.1016/j.jp...

This is fucking wild. My brain is exploding.

I came to Minneapolis to report on what's going on, and one of the main questions I showed up with is "just what is the scale of the resistance?" After all, we're all used to the news calling Portland a "war zone" or whatever when it's just some protests in one part of town.

Very cool—thanks for the update! Today I got gassed by the federal government two times within a few blocks of my house

I gave my daughter one of those hats last week. She is obsessed with it, wears it everywhere; insisted I get one for all of her friends.

All of this is utterly vile, but this hits especially close to home. This child is no different than any of ours. This is an attack on all of us. I'm so, so mad.

A comprehensive breakdown of how different conversations in the literature hang together, and where they disagree, used to be a vital task for researchers. How a new paper fits in was important work!

Now it takes 30 seconds and costs half a cent. There's a quality trade-off but at that price...

This is the norm I most expect to change in response to generative AI.

Irrespective of any other thoughts on its utility, 'track down every single paper related to what I just wrote and substantiate every sentence with citations' is a job "fancy autocomplete" crushes.

Dr. King did amazing things, but often we only talk about his successes. Dr. King also had some "failures", and they are important to discuss because we often learn a great deal from failure. And failure can inform or grow a movement, like Albany, Georgia's failure did for Dr. King. /1

At risk of being heavy handed, this thread is about AI.

Coding agents are going to relieve some bottlenecks, while other factors become limiting. Will the subsequent order be less, or more human centered?

It took decades to figure this all out for electric motors - will it be faster this time?

This new set of constraints kicked off the 2nd industrial revolution - replacing the old, rigid, machine centered paradigm with a new, flexible, human centered one.

The main hurdles were not technological, but architectural - understanding how new constraints opened up newer, better processes.

Individual devices could also more readily be switched on and off in an electrified world. This introduced a level of task flexibility that was previously too costly to seriously consider.

Taking advantage of that, factories were radically reconfigured. Instead of designing around maximizing the power from the line, factories could be laid out in a more human-centered way. Workstations could be designed around workers and their tasks, rather than workers around the workstation.

It wasn't for a couple more decades, in the early 20th century, that factory planners started to understand the power of electricity.

Electricity was more flexible. With electricity, you could power many smaller electric motors. You didn't actually need the line shaft anymore.

This had limited success. Not just because of the efficiency of steam engines - electricity today is still often generated by making steam to drive a turbine that generates electricity. Why include the extra step when you can drive the line shaft directly?

Industrialists of course experimented with replacing the main driver of their line shaft with an electric motor - upgrading the technology powering the factory in the same way that steam engines has replaced water wheels and mules.

Electricity had been invented long before all this, but electrical generators had lagged behind steam engines in their power and usefulness.

The first applications of electricity in factories were in lighting, replacing gas lamps with a safer, more reliable alternative.

Factories are laid out to maximize the power from the engine. Workers crowd around a drive shaft, feeding in materials or pulling out finished parts, doing the fine tasks the machine can't.

The machine works at its own pace. If workers are slow they miss a cycle, as the machine moves without them.

Steam engine factories worked the same way - but you get even bigger shafts, more elaborate pulley systems, and complex floorplans to take advantage of all of that power.

Steam engines were a drop in replacement for those other technologies, and allowed for much larger scale line shaft layouts.

Think of it like how an old windmill works - the wind turns the wheel, which powers a main drive shaft, with a series of belts and pulleys powering everything else.

Thinking a lot about electrification lately.

Let's go back to the end of the 19th century. Factories were booming from the proliferation of steam engines, which could deliver far more power than the technologies they replaced - windmills, water wheels, or draft animals walking in circles.

My impression is that there's enormous latent demand for customized software, B2B especially, that'll eagerly suck up productivity gains in the short run.

There's going to be a lot of money in those applications.

In the longer run I suspect 'writing code' will cease the be the gating step.

The key part of the article is that asset prices are rising faster than wages, while affluent consumers prioritize positional goods.

You can't grow your way out of a status competition. Soaring asset prices amplify it, where the winners are not determined by productivity, but generational wealth.

EVs alsp destroy the dealership business model because there are laws in most states prohibiting manufacturers from selling internal combustion engine vehicles directly to the consumer (or running their own dealerships), but those laws do not apply to electric vehicles.