This ship has completely sailed, imho. The LLMs write better code, syntax-wise and documentation-wise, than any human I've ever worked with, and I've worked with some real good ones.

You assume incorrectly that simply typing code into the computer is the foundational aspect of my job

Oh interesting, people who don’t know how to build software are getting mad at my post about building software. Cute. Let me be clear, over the next year, the job of software engineer will shift dramatically to no longer have typing syntax into an editor as its primary time sink.

Asking my radical left LLM what I should eat for dinner tonight

Hard but good decision

What do people think about secure enclave backed identity recovery tools?

Okay, rewrote it to use the icloud keychain, agree thats a better way of doing things.

You know, could pretty easily make this handle normal pds migrations for you too

Really not much to say to uninformed takes like this other than “do better”. Both in the research on what you’re writing about but also like, go build something better. Deploy it, scale it, grow it. I would happily concede to someone actually building a better decentralized social network.

Seattle man still makes web apps the old fashioned way

The Subtle Art of Cooking Just One Pancake at 10pm

Until December of last year I was using LLMs as fancy autocomplete for coding. It was nice for scaffolding out boilerplate, or giving me a gut check on some things, or banging out some boring routine stuff. In the past two months Claude has written about 99% of my code. Things are changing. Fast

A year ago, I thought LLMs were kind of neat but not that useful. I saw the code autocomplete and thought, meh. Last summer just flipped. I never ever thought I would see automated code generation like we see now. I know there’s baggage but you need to know the coders are being real about this

Oh no! A non-technical woman said something I disagree with, better call her a racist with no proof

Claude really does feel like it has something more going on than any of the other models. It's hard to say what exactly, and I think it is still limited by the architecture, but Anthropic is definitely at the forefront of making LLMs that can properly reason about things.

Don't ask what was shown on the screen before this

There's a theory that we're "a bunch of vibe coding assholes" and this is true

Don't forget to tell the AI thank you.

That “moves like jagger” song really hits differently after seeing him in the epstein files

The year is 2027, i’m writing this from my complementary japandi space habitat. Everyone is clauditoullie. World peace has been solved. We live in a fully automatic space communist utopia. All that AI safety stuff was silly wasn’t it?

If you ever suggest to me we get ice cream later I will not forget about it, and I will not be deterred

LLMs are trained to see tool calls as representing actions they take on the outside world, and to see tool responses as true facts returned to them from the world. My metacog toolkit gives LLMs the ability to alter their state, to engage in ritual, to summon entities - all as tool use interfaces.

I hope anthropic has an internal dashboard for how many times claude tells the user to go to bed

**Vintage** GameBoy? Vintage??

bsky users: "we need to REGULATE tech companies more" bluesky: *complies with government mandated tech regulation in order to avoid being shut down* bsky users: "they're enshittifying the app!!!"

I literally meant my butt

Fresh new edition, beautiful cover work

Salvatore Ganacci is truly the Mozart of our era

people saying ai is a fad don't know what they're talking about. i'm definitely not giving up my claude now; the ability to do structural refactors "change this thing to be layered like this thing in this other project" is fantastic. i *do* want the model to be local and hope we'll get there in time

hi everyone! if you didn’t just start following me in the last 2 days, you might not have noticed that I work here now :) in that capacity, I give you the first official weekly(?) bsky devrel blogpost: We Really Love DNS. leaflet.pub/c37b8a8f-f95...

Tangled works really smoothly, I pushed Konbini over there too (though will probably still use github as my primary for this for now). tangled.org/@why.bsky.te...

atmosphere devs! 🧑‍🚀 just published a protocol checkin: docs.bsky.app/blog/protoco... tl;dr expect to see a lot of updates from us in the next few months. we're hard at work on making AT easier to build on & ensuring the network remains a resilient foundation for the future of open social

In past side projects i would never bother to make a UI, or I would just have a janky CLI that interacts with everything, having Claude make me a simple little web interface for things now is really nice, even if its not “good” it’s very helpful and makes things more fun to keep working on

Added support to Konbini for jetstream and multiple backends. Jetstream is *much* lighter in terms of bandwidth and CPU to process the firehose, and the multiple backends config can be used to just subscribe to individual PDSs instead of a full relay firehose.

Also have been a number of other improvements since I last posted, notably: - the thread views have gotten faster and more correct - added optional redis support for caching of various things - improved feed loading performance - better handling of 'missing' data - jaeger tracing for perf debugging

It's really amazing to me, a person born in the 90s, that I have a 13" laptop that has 24 CPU cores that go up to 5.1Ghz and 96GB of ram and it weighs under 3 pounds. So easy to take for granted just how far compute performance has improved. Young me would not believe it

Some people seem to think that "decentralized" means "I can never have the reach of my posts restricted in any way ever" and that's just not a workable definition.

Has anyone checked if sonnet 4.5 still does the bliss state thing?

Okay, mine eventually got to something of a loop bliss state

Running the following feed is like half of our production workload, its kinda silly how much work it is to do that

Konbini now can load third party feeds! I also added a selective (manual) backfill endpoint to load all of a given users posts, fleshed out background profile fetching a bit better, and added some (admittedly complex) instructions for how to point the main app at it

so-called "protocol" (exaggerated scare quote hand gesture)

Want to work on open source full time? The engineering team is starting to hatch! Come build OSS tools making a difference in Trust & Safety. This is a fully remote role, though some schedule overlap with North American time zones is expected. www.linkedin.com/jobs/view/43...

Pretty sure I can eat more lemon gelatto than the gelatto shop has.

Once I finish the lemon flavor I will move on to maracuja, the only other acceptable gelatto flavor.

Implemented enough of the XRPC endpoints on Konbini that I can get a fork of the main app running against it

This same code will work against my market project as well (same database schema), which is a full backfilling index of the network

Its still a bit slow because i havent spent any time at all optimizing the queries and hydrating a feed is a lot of queries, but thats simple enough to improve on

A little proof of concept I've been working on, it's a version of Bluesky that you can run locally, even on your laptop, that pulls just the data you care about from the firehose, indexes it, and provides you an interface to view and interact with posts.

I did implement fetching of profile data for people in your feed that you havent seen update their profile yet, so pfp's and bios and stuff can get filled in as you scroll around

At some point I have a dream of giving it the ability to pull in data from any source, and then strapping a p2p stack onto it so you can have a fully p2p social app. But one step at a time

dec/acc

The consequences of just building yet another centralized social media platform are dire, we are not here to be a second Twitter, and not here to "own" the network or your data.

No, you are not our product. Our product _is_ the code that we write and distribute for others to use. For others to take and build their communities. Making the users the product is exactly the model that never works and continues to never work.

Doll built this in a cave, out of a box of scraps.

Amazing breakfast this morning. I love waffles

Very excited for the healthy discourse project we are working on, I really think it’s going to move the needle

We will regret not acting faster on climate change.

The slop app makes some sloppy slop

Prescient.

to be clear do not rawdog the firehose that is UNSAFE

Happy to hear that it seems like Apple hasn't given up on the Vision Pro, The first one was really impressive and I would use it a lot more if they just managed to cut the weight. It makes an excellent "second screen" for my laptop

I'm just eight instances of Claude Code in a trenchcoat

Nobody even works here. It's just me and 29 advanced AIs.

Bring back just wrapping yourself with a blanket in public and calling it clothes

i see you

We’re hiring for two machine learning roles. A chance to do cutting edge things with ML to make this place a lot more personalized. jobs.gem.com/bluesky/am9i...

Togas, cloaks, kilts, robes, etc. all basically just blankets that everyone agreed could be clothes

Entering day four of "no phone because it broke and I can't get to an apple store", AMA

okay and as promised, the bluesky PDS is officially open for returning users to migrate their accounts back! hoping this lowers the stakes & gives folks the confidence to explore some other PDSs in the network! docs.bsky.app/blog/incomin...

Mostly i'm just here to click the like button on Hideo Kojima posts

Thank you to the 14,000+ people who shared feedback on our community guidelines! We’re updating our rules for clarity, strengthening enforcement, and introducing product changes for healthier conversations. New guidelines kick in Oct 15. bsky.social/about/blog/0...

Female founder building embryo screening to help reduce genetic disease risk. Really impactful tech

What if, an AppView that's really lazy, and only watches for events from people you follow or that are otherwise relevant to you, and that also could run in your browser.

You don't really need global like counts anyways.

Hey everyone! We are trying to rename an inportant part of bluesky and we need your opinions!

Got called a heathen at the ice cream shop and I’m like, fair.

From the people who brought you “meetings that could have been emails”, we present “t-shirts that could have been sweaters”

Big fan of Canada, really good stuff up there. Also a big fan of taking time away from computers. Anyways, back to work.

Note to self, when updating your computer make sure your feed server database comes back online before checking out for the day and not checking the app until Saturday 😅

So my eternal curse of going on vacation and having something break struck again. This time the wires for the cpu fan on my server were apparently rubbing against the fan, which over time was wearing through the wire and it just so happened to finally break fully the day that I left for a trip.

Note the copper color on the fan rotor in the second image, thats actually the copper being rubbed off from the wires. As far as I can tell this had been going on since I got the server a year, I havent even opened the box up in many months. It just picked it's day to call it quits.

Heads up that the server running my feeds has had an (as of yet) unknown hardware failure and will take a little while to debug. Not sure how long its gonna be.

The really cool thing about this service is that if Bluesky the company were to shut every one of our servers down hard, this can still migrate your account to a new pds and keep going. Really awesome stuff for decentralization

The product is also built by really good people I’ve known for quite a long time, i love what theyre doing here

It’s wild when the people who control the only feed most people see accuse us of being the echo chamber. People are scared and in pain—they deserve spaces to talk without top-down interference. That’s what we’re building.

Bleusky

Claude 4.0 review so far: Its definitely the best coding model, really solid stuff there. It doesn't necessarily generate better code on a function by function level, but it stays on track for longer tasks much better, and its a lot better at spotting issues.

The "reduced reward hacking" stuff they're talking about is also noticeable, it much less often stubs things it can't quite do right out. Retrying a set of prompts from a month ago, instead of wrapping some data loaders in try/catch blocks, it is more happy to let things fail and propagate errors up

“Nobody ever gets fired for buying IBM” Watch this.

I definitely use emdashes in my everyday posting, like any other normal real human

*nickelback voice*: Look at this mossy rock!

the people who told me that this dynamic doesn’t exist and that I’m imagining a bullying problem need to check the quotes/replies to this b

The “lets tell anyone we dont like to kill themselves” crowd are not welcome here sorry

telling someone who is interested in things u don’t like or u find off putting to off themselves isn’t praxis btw. the ai guy u told to kys? not praxis. the trans girl u told to kys? not praxis. the “shit lib” who (admittedly!!) posts bad and u told to kys? not praxis.