jonny "saunders"

Hell ya

So that's a big training data exfil vector

If an adversarial takeover happened tomorrow I'm not so sure how many footprints would remain owned. The PDS model very much can support this, but I don't see much interest in realizing independent ownership in the bulk middle of atproto culture

I remember this well, I was in college in exactly this time period and remember it being an early exposure to platform politics when it got bought. Glad it got a history.

🔬🧠 Releasing the 1.0 version of #Suite2p and THE PAPER w/ @marius10p.bsky.social! Now with GPU acceleration. Want to use Suite2p but don’t have 100,000 neuron recordings? We show you how to get those with a standard 2p microscope #neuroscience #imaging #neuroAI www.biorxiv.org/content/10.6...

looking forward to checking it out, always love to see new work from your group, already extremely good and just keeps getting better somehow

the heady days of the grift are behind us, the long tail of "uh hey everyone i just looked at my credit card and something seems wrong" lays ahead

i read some software today from the latest models that was a complete disaster for this one person's private problem, but it demanded three people's days to evaluate it to say it wasn't fit for any other purpose.

the people who keep saying they are having their mind blown keep being people whose job it is to tell other people what to do already

i think it's probably unlikely that "everyone" only tried chatGPT in 2023 and is just out of date. i think maybe AI tech CEOs not being personally exposed to the externalities of their products is what you might call more parsimonious than everyone being ignorant.

show me the output, and then show me the social system it's dumped in

show me the output

or, rather than "number always goes up" meaning we will be able to rent god for $20 a month this year, if you actually have to deal with the downstream consequences of the sludge, things look very different - important social safeguards eroding make broken systems at high-volumes palatable

When people think they can automate science what they're saying is they think they can automate millions of life-years of hacks and gambiarra, which is the same thing as saying you can simply automate the motion and mass of every atom

There is no "way" Its supposed to be. The slides have always been sealed with nail polish and the pipettes have always been lung powered and superstition polished

People think science is like supercolliders and big steamy pistons and test tubes but science is actually like hot glue and vaseline and aluminum foil and little plastic purses

There are so many glues. The glues are all different and have alchemy style interactions with other objects. Some glues are just mad at glass, its how they are.

Years ago I registered glue [dot] wiki because of the sheer quantity of contextual knowledge surrounding glue in research that literally has no venue or medium of knowledge transfer

Dogs are glitchy enough I bet there would be some good techniques in there, dog circle around scent 47 times to write bone to registry and this lets us take over at another dogs state in a previous run where the pendulum swings were positioned for a gapless traversal

ah yes, the i don't look at the bad part of the system that i am integral to the continued function of so therefore avoid all ethical culpability defense. and who says academics have no culture

same, good memories. this matlab project used to run the behavioral experiments across several labs, hand to hand flash drive versioned by untrained grad students across ~10 years. It had a fixed >1s ITI by how it tried to update itself from a long-dead SVN server at the start of every trial

what i don't understand is why one would want engagement from people who don't have a problem with engaging with you over a tool that is literally labeled "informational weapon to be used exclusively in service of fascism" right on the box

i get much more engagement screaming directly into the vortex of human misery

Some of those moments are sweeter than other, depending on the historical prestige and exclusionary gradients in the background. Seeing people waste coders time is easier for me to forgive since coders wasted so much of everyone else's time declaring "that looks easy" and then completely bungling it

There's a physics prof here who generates multiple papers a day on some neuro theory that us mere mortals can't understand, storms into neuro meetings like "why are you talking about this, ive already solved the brain." And im imagining that energy playing out in every direction.

Matches my experience, and def +1 that human grounding rec.

Re: the "everyone started coding -> more work for the people whose job is coding" example- Curious how that secondhand AI labor burden plays out for other roles. The coding story is common wisdom now, but can't be the only one right?

The competition has billions of dollars, billions of users and deep political connections.

Quit the crabs-in-a-barrel mentality if you care about all of our collective work meaning something in this lifetime.

Interesting. ive only ever had way chunkier steel bikes so I definitely thought they were always heavy. I have been wanting to make a disastrous ebike conversion but didn't want to sacrifice my carbon bike. now I might look for a used steel bike, thx for the tip

Wait I'm also sleeping, what is good about them? (aside from looking very cute with the lights)

Why? Because now the maintainer is the person driving the agentic coding model (the vibe coder has almost certainly checked out at this point and is just shoveling the maintainer's review comments and the model's commits back and forth)

This turns the maintainer into an unwitting vibe coder