I really broke my algo in the past month when I was writing about AI in law enforcement, but now I can bring you daily dystopia. Axon is doing things with AI that anyone who cares about privacy would recoil from.
Today, they are marketing a panopticon that must include AI monitoring tools.
This is the coolest shit I have seen done on the federated web
American and Israeli missiles hit Sharif University in Tehran tonight. Often called #Iran’s MIT, the institution has been home to some of the country’s best and brightest minds in science, technology, engineering, and mathematics since 1966. 1/2
Some reports say over 500 schools, 55 libraries, & 25 universities hit.
You can debate the numbers, but hitting Sharif University & Beheshti is like hitting MIT & Stanford. I keep wondering: How would the scientific community respond differently if it was those universities? What’s the difference?
new: Blacksky People’s Assembly conversations display embedded inside posts on blacksky.community
In our next conversation, you’ll be able to vote on People's Assembly topics while scrolling the feed and without leaving our app. 🗳️
this is so fucking awesome
Demo People's Assembly conversation where @ruuuuu.de is logged in and is a Blacksky Member. He's voting on "Rudolph Fraser's" statement "strong opinions" @rude1.blacksky.team has the admin, funder, member, and OSS supporter badge
Our next @blackskyweb.xyz People's Assembly will have you login with your atproto account to participate.
On login we'll match your handle/email/DID to our subscribers, open source supporters and check for Blacksky membership (in feed or blacksky.app)
Badge + profile will show next to statements.
And that's very challenging with the LLMs! Because without careful watch over how you are using them, they have the modal social impact of "rudeness," so now the lessons about respectful contribution and working together come sooner.
The most important thing I was not taught and I try to impart is that being part of a community of people working on something is much more important than code on page. If you are working with people, talking about what you all need and ideas for how to do it, that's much more important than syntax
But I do have to also now teach how to be skeptical of something new, how to read for volume, to only half trust what's already written. I haven't seen anyone really struggle to plug into an AI tool, but slowing people down from walking off a cliff with them is harder.
Of course, with the most recent people I have trained, I still focus on what I always have: how to think about what you need, the patterns that will give you a strong skeleton to get there, how to set development goals and knowing when you're there. How they get the code on page is less important
Yeah right - like if you have to write code to do something, why aren't you like keeping track of that, sharing it, refining it, yno contributing to shared empowerment. It feels a lot better to find someone's tool and give them credit then stumble around on my own.
The design of all the consumer frontends obscures most of what is actually done, especially ones claiming to be "agentic," and watching the debug logs I can't count the number of times ive watched it hit a wall and then actively concoct a lie. That's the last thing I would want analyzing a paper
Something I still can't quite fathom is being able to just take the model output summaries of what was done at face value. In most LLM PRs there will be some claim about what it does where you look inside and go "no the fuck it does not." It only took one of those for me to always inspect everything
TBC I'm not saying you are being overbroad here, I am a professional LLM hater lmao, just to say that even for those who are big boosters and think its magic, this should still be part of their model of the world and how their actions fit into it
Its complicated, there are a lot of different angles and a broad "everything is good or bad" brush doesn't quite get it, but ya one thing that's important to me is even if it works perfectly, its fundamentally an antisocial practice
Programming is hard, and even the old model of every lab handcodes their own private code is extremely busted in a different way. Pulling labor from shared, inspectable, validated packages into gazing privately into the orb is a step backwards.
It just isn't really validating the paper, and you don't really know if what you're doing is actually what they did, even if you do actually validate the code that is generate. Things looking sorta similar is not a great standard.
In the other you are essentially asking the language model to be credulous to the results of the paper, and it also receives the expected outputs, so it estimates something that is roughly similar to the textual description and produces roughly the expected outputs.
Asking an LLM to ingest a paper and produce code for it is not the same operation as open code/open data independent validation: in one you have a fixed set of operations that should produce what is in the paper and a) lets you validate that matches what they wrote and b) evaluate the implementation
Not sure how great it is for the state of science where rather than publishing code and working on shared libraries that make it so everyone can easily do exactly the same analysis, everyone just generates their own code privately that purports to do something and kinda looks right but not sure
I read this @leahsottile.bsky.social essay last night at the speed of print and I thought it was an exceptionally thoughtful take about who is, and who isn't, considered a terrorist
www.hcn.org/issues/58-4/...
The complete destruction of American cognitive science is happening right now
The NSF 2027 budget has noted that they will close out the Social, Behavioral, and Economic Science Program (SBE). This is not a good thing. nsf-gov-resources.nsf.gov/files/FY-202...
Trump is seeking to pay for his new $1.5 trillion military budget by cutting the following: $510 million - Grants for farmers and agricultural research $82 million - Loans for rural small businesses (Fully eliminated) $61 million - Support for farmers and food markets (Fully eliminated) $240 million - School meals and food education for children abroad (Fully eliminated) $659 million - Community building grants $47 million - Support for minority-owned businesses (Fully eliminated) $449 million - Economic development grants for communities $1.6 billion - Weather forecasting, fisheries, and coastal protection (NOAA) $993 million - Scientific research and technology standards
$150 million - Support for American exports and trade $2.2 billion - Broadband and internet access programs $8.5 billion - Funding for public schools $1.5 billion - Vocational training and adult education (Fully eliminated) $2.7 billion - College access and higher education support $15.2 billion - Roads, bridges, and infrastructure projects $1.1 billion - Home energy efficiency and clean energy programs (Fully eliminated) $1.1 billion - Scientific research funding $386 million - Environmental cleanup programs $150 million - Cutting-edge clean energy research $4 billion - Help paying home heating and cooling bills for low-income families (Fully eliminated) $768 million - Refugee resettlement assistance $819 million - Care and shelter for migrant children $775 million - Local anti-poverty programs (Fully eliminated)
$5 billion - Public health programs, mental health services, and disease prevention $5 billion - Medical research (NIH) $129 million - Healthcare quality and safety research $356 million - Emergency preparedness and disaster response $1.3 billion - FEMA community disaster preparedness grants $707 million - Cybersecurity protection for critical infrastructure $52 million - Airport and transportation security $40 million - Protection against chemical and biological weapons threats $53 million - Funding for homeland security operations $3.3 billion - Community development block grants for local neighborhoods (Fully eliminated) $1.3 billion - Affordable housing construction grants (Fully eliminated) $393 million - Programs to reduce homelessness $529 million - Housing assistance for people living with HIV/AIDS (Fully eliminated)
$489 million - Housing and services for Native American communities $50 million - Grants to help communities build more housing (Fully eliminated) $60 million - Enforcement of fair housing and anti-discrimination laws $58 million - Homebuyer and renter counseling services (Fully eliminated) $45 million - Renewable energy development programs (Fully eliminated) $1.7 billion - Grants for local law enforcement and public safety $20 million - Civil rights mediation and legal access programs (Fully eliminated) $1.6 billion - Job training for at-risk youth (Fully eliminated) $395 million - Jobs program for low-income seniors (Fully eliminated) $234 million - Worker safety and labor protection programs $101 million - Enforcement of equal pay and workplace anti-discrimination laws
It’s a death cult.
You may also enjoy the new "auto" feature, where to check if a given tool use is safe, the exact same LLM with exactly the same context window is asked if the tool use it just proposed is safe.
neuromatch.social/@jonny/11633...
"if you have unlimited tokens, funding, and expertise, you can use the AI to produce an incomprehensible nightmare that can only continue to work by constantly expanding its tech debt against the surface of its tests" is a great value proposition for AI.
right now the astronauts are calling houston because the computer on the spaceship is running two instances of microsoft outlook and they can't figure out why. nasa is about to remote into the computer
😭
The Orange Cat Brain Atlas is here. 🧠🐈
Today, we published the first comprehensive cellular map of the orange cat brain. The new atlas reveals a single, specialized neuron responsible for behaviors like staring at walls, knocking objects off tables, and the 3am "zoomies."
