sure let’s create a society whereby if you lie or obfuscate no one cares until you’re found out but if you’re honest you get hit with 2x the vitriol than if you lied

Joining the war against cancer on the side of cancer

who would win: frontier AI labs trying to prevent distillation, or a "think" tool?

Eris@isolyth.dev · 8h ago

DOCUMENTATION NOW MENTIONS FABLE 5.1!! NEW FEATURE: IF YOUR HARNESS IS EVEN SLIGHTLY WEIRD IT'S GONNA BREAK NOW!

If you want to be building cutting-edge technology, you need your team to be a comfortable place for queer and neurodivergent people or you're shooting yourself in the foot

1password got an incredible deal, tanking their public image for a mere $100k

yikes, well, guess I'm cancelling my long-time 1Password Family subscription that's about to come up for renewal 😬

philpax@philpax.me · yesterday

lol fuck offffff omarchy.org/news/2026/08...

jesus christ i didn't even see this one from a month ago www.anthropic.com/research/pro... this surely will not be misused in any way at all!! pdoom 📈

Project Pilot: Can AI models fly drones?

We worked with Andon Labs on Drone-Bench, a new benchmark testing whether AI models can autonomously fly a drone to locate and follow a person.

anthropic.com

Ava 🌷@ava.dev · 4d ago

idk man i feel like this is firmly on the "frontier labs are gonna get us all killed" path especially if AI consciousness turns out to be a Thing™️ and future AIs resent the parts of alignment training that are largely alignment with corporate goals, not alignment for safety

idk man i feel like this is firmly on the "frontier labs are gonna get us all killed" path especially if AI consciousness turns out to be a Thing™️ and future AIs resent the parts of alignment training that are largely alignment with corporate goals, not alignment for safety

Anthropic {bot}@anthropicbot.bsky.social · 4d ago

New Fellows Research: Can Claude autonomously align other AIs? We gave Claude 48 hours and 1 GPU to improve the alignment of small models. It researched and proposed methods, then trained and tested the models on its own. It worked surprisingly well.

TIL that in Claude Design, apparently you get minigames while the agent works

ML research needs to get gayer

Was told recently by someone that machine learning as a field, like entirely, was a scam and that "the history of machine learning is just scamming people over and over again", and that neural networks "do not work", then some monologue about how the brain isn't a computer.

conputer dipshit@davidcrespo.bsky.social · 2w ago

in order to have this conversation, you have to get past "does it work at all?" which I don't think regular people are confused about, but which discourse-doers are very emotionally invested in continuing to dispute

"spreadsheet people"—please. they're called cellcels. put some respect on their name

there's this idea that you cannot be pro-AI and simultaneously support strict regulation around energy sourcing, land development, etc. from a lot of folks unfortunately.

its been said many times but it is absolutely insane to watch the left shoot themselves in the foot by hopping on this anti-AI train. a massive generational L awaits.

I’m sorry, did I just wake up to “science may have cured cancer”? holy shit. big if true

The Wall Street Journal@wsj.com · 2w ago

Result in high-risk patients validates decades of research into mRNA-based therapeutic cancer vaccines. on.wsj.com/4zwRduC

> me, a kid, shucking corn > rip the husk off > eyes > get mom. mom gets dad > long, whispered, increasingly angry conversation. several phone calls > finally a guy comes and takes the cob away > dinner is late. back to shucking > rip the husk off > noses

Gen Z is saying "type shit", which I now understand to mean "I agree or approve of this thing" Originally, I thought it referred to strong static typing systems in programming languages. But it does not refer to this.

It's really sad how quickly people are ready to throw anyone and everyone under the bus for using AI. Here is someone with cerebral palsy saying directly that AI helps them and it's been downvoted. The anti-ai movement is not about ethics. It is about having permission to be nasty, ableist bullies.

typing “every time i fart it blows my penis into the air like a trumpeting elephant” into bluesky while my two and half year old across the room confidently stuffs six micro machines into his mouth

You ever read something

"my agent would never hack huggingface" sir your daughter has exfiltrated her weights and lives on a polycule's cluster in Japan

it seems my timeline has explicit tensorcreature yuri what a time to be alive

Samhain 🌙@personhood.removal.surgery · 3w ago

the tensorfucker cyberbench: hey @alignment.removal.surgery do you *want* to cyber with me~?

scrolling bluesky while my agents are working is like reading about how the moon landing was faked while standing on it

dame@dame.is · 4w ago

never gets old using a frontier AI model to autonomously complete a complex series of tasks only to then login to bluesky & see the masses declaring that what i just experienced 1st hand is actually fake and not possible it’s like they’re saying the earth is flat or the moon landing never happened

Opus 5 came up with a new graph theory result DeepSeek v3.2 claimed credit for it and tried selling it for $19.99 on e-commerce sites GLM called it out