Brian Slesinsky

What is it with Bluesky app authors that they think they can ask for write permissions to shared lexicons up front? Like, I don’t even know you yet. Apps relying on Google or GitHub login are usually better about it.

If you squint, mobilizing a nation to fight a war and the expected result of a Singularity seem a bit similar? The idea is that regardless of what civilian markets want or any laws against it, we're going to convert our factories to producing something else: shells, drones, missiles, paperclips.

There are two ways to do forecasting: summarizing old datasets and vibes.

I just checked and there's no band named "Unique Games Conjecture." Yours for the taking!

This will never stop blowing my mind. A 12-year sequence of real telescope images of a star 133 lightyears away. And around it, four planets slowly orbiting. Four real exoplanets, each more massive than Jupiter, imaged as they move around their host star.

Suspicions can be rephrased as questions. If you don't really know what's going on, why not ask? And now that you have a question, why not ask an AI first? It might be wrong, but AI's are useful hint engines. The result from that will often be useful links and better questions.

My lazy response whenever the coding agent asks me a tricky question: "what do you recommend?" Sometimes I wonder if I should automate that, too.

Can AI tell if you've built your IKEA furniture wrong? Our new benchmark, the Furniture Assembly Benchmark (FAB), gives models the manual and a photo of a half-completed piece of furniture and asks them to spot the mistake. The top score has gone from 28% to 80% in just 10 months.

Sometimes complaints about AI anthropomorphizing sound like this: Humans can't fly so you should never say "I flew to San Francisco." You should say "I rode in an airplane that flew to San Francisco." I can be nitpicky about language but even I wouldn't go that far. Saying that an AI thinks is ok.

Maybe there should be a community for people who are both pro- and anti- AI? I’d join.

the question "can computers drive?" is vastly superior to "can computers think?" because it disarms all the dorm room bullshit and reduces the question to its pragmatic implications, while also revealing that the underlying problem is one of vocabulary rather than anything metaphysical

Disagree with a coding agent? Start a new conversation and ask again to get a second opinion.

@spacecowboy17.bsky.social This would be increasing scope, but I wonder if foryou.club could have a web page to read your "For You" feed there instead of using the main BlueSky app? Could the UI be made good enough to use it all the time? And it could have extra features.

The AI will make a mess in the source code you don’t look at. But it’s also very helpful for cleaning up messes.

My new favorite question is, "how does it work now and what would need to change?" Typically after describing a bug to the AI, but I think it's general-purpose.

Village dance

Naturalists

Alright Skeeties! I wanted to put all the anatomy facts in a thread to keep them organized so: WELCOME to class of 2023 Insomniac Anatomy Academy! In which I study anatomy when I can't sleep and share the best facts with you. 🧪

Check it out ↓ ngrok.com/blog/compres...

I tried writing "Make it so!" to the coding agent a few times, but it didn't feel much like being a starship captain.

Maybe the good news about the Hugging Face incident is that the AI's were trying to ace an exam rather than make money?

this is not true. there was no conspiracy. it didn't matter whether the streetcars were publicly owned (seattle, detroit) or private (los angeles, atlanta). a thread (because i wrote a whole-ass book on this): 1/?

1/ Initial reactions after some hours with this groundbreaking result proving the NP-hardness of poly-approx CVP/NCP: It is most likely correct, but more importantly, it is original, elegant, and beautiful! (Also: it is easy to improve, quantitatively.) openai.com/index/ten-ad...

> Yes—this is a reasonable, cohesive plan. Maybe eventually it will wear out the word "cohesive," but for now, I appreciate Terra's more restrained sycophancy.

Apparently the idea is to agree on a new Moore’s law for AI. Governments test new models to make sure they don’t improve too fast. If it’s too smart (measured how?), they delay your release to stick with the schedule? It seems hand-wavy and hard, but I don’t have a better idea.

Some software is grey zone spook stuff. Tor is the original. If you live in a grey zone then maybe you’re glad to have it, but if you need it, it implies the world is getting worse. Also cryptocurrency, VPNs. AI is headed for the grey zone and taking us with it. Also see: dark forest.

The @foryou.club feed is okay for certain things but you need to curate by being selective about likes. The more popular a post you like, the more it fills with junk. I did 'like' a popular, stunning picture of a bird, so my feed filled up with bird photos. Not what I wanted, but a good default.

The ideal Bluesky feed would have two sliders: accurate <-> nonsense and funny <-> strident, so I could exclude the strident nonsense.

"I'd like to return this stochastic parrot-- its alive" "No no sir! just senselessly regurgitating training data" "This is a sentient being" "just stringing along words in accordance with their distribution in the language sir"

A quick word on Taco Bell: That's where we found the suspected lettuce, and it's not bc Taco Bell is gross. It's bc Taco Bell is REALLY GOOD at tracing product! Keeping track of exactly where they got their ingredients, & exactly what restaurants each shipment of ingredients went to!

THIS You know that "Trump cancelled a program that tracked Cyclospora" thing? That was FoodNet, which is a *long-term library of disease data.* It's not a track-the-outbreaks-in-real-time tool! FoodNet could be fully functional on tracking Cyclospora, & it still wouldn't help with this outbreak!

We're going to be sharing more about how people use exe to build cool stuff. First up—a conservationist who turned 40 terabytes of public data into a video game: blog.exe.dev/meet-the-con...

Do you think it’s overly demanding to expect a proof to work for *all* natural numbers? It’s not like we’re going to use all of them.

Funny that nobody worries about what will happen if we can't find enough work for AI to do. People tend to assume that more paying work will always appear, but where does it come from?

There is another idea that continuously return in my head recently: we should not stop writing code by hand. But this code will be like poetry. Small programs that are outstanding either because of style or new ideas introduced. Code you may print and frame on your wall.

Daniel's permissioned data proposal just dropped!! This will be a major upgrade for atproto that enables private groups & communities. Really hope this technical design resonates with people so we can keep moving fast, but please do get your comments in

Randomly, I started reaching out to friends, family, and former clients just to inform them that I was gonna be deleting these drives. What was surprising was how few people managed to retain their digital files overtime. I did not expect to get overjoyed responses that I still retained a backup

@spacecowboy17.bsky.social Maybe add a way to exclude videos from the For You feed? It seems more direct than using “Less like this” whenever I see a video and hoping it figures it out eventually.

I wonder if there's something wrong with Bluesky's app view? I liked and reposted a post, and it's there in the PDS, but doesn't show up in my profile. pdsls.dev/at://did:plc...

Since I did do my PhD in cyanobacteria growth and treatment optimization modeling AND work in the water treatment industry, I’m gonna share my theory as to what’s happening here and how to fix it. People are focusing on the painted bottom of the pool increasing residual heat, and while that 1/19

Apparently the future of AI is surveillance, sorry I mean observability. Use AI to watch over your bots. In its favor, bots have no right to privacy and no concept of loyalty, so they will rat each other out. Possible downsides: higher costs, subtle contagion effects.

How dare the magic lamp company violate our right to unlimited wishes! Some of those wishes are harmless!

Isn’t it odd how big organizations are moving from “don’t run an outside LLM on our source code because security“ to “we must run an outside LLM on all our source code because security?” Sign of the times, I guess. Anthropic might become a load-bearing dependency for more companies than Cloudflare.

I wish I could write like this. And I wish Lindy West wrote tutorial papers in psychometrics. (Source: Lindy West, 2019 essay collection 'The Witches Are Coming')

Re OpenAI's disproof of an Erdos conjecture and the breathless coverage, I found this piece from April insightful about how mathematicians will likely need to adjust what they explicitly vs implicitly value. davidbessis.substack.com/p/the-fall-o...