Chris Murphy

@theo I've been working on a lot of DX focused libraries that post AI might not be as useful. But, this one might be. It's a way to enfuse state management functions directly into the data and create a fluent api. cogsbox.com/potentiate Site's rough, early, just vibe checking.

@RhysSullivan Perfect use case. Github issue searches are such an old school experience. It's like forensic analysis, reading post dates looking at credentials of posters, cross referencing threads. I just want, "is this issue known, what's the current state?"

@MonoBrainChris next up is probably expanding on answeroverflow.com/chat thoughts on it are it'll be able to search docs + github (issues, code, prs, discussions) + discord to give you extremely high quality answers

Answer Overflow just hit 10 MILLION ALL TIME VISITORS!!! is incredible it's come this far, surreal thinking that it's been this helpful for people

@soychotic But Javascript be like "What do you mean int? This definitely isn't an int it's 2.000000000034".

The control freak in me wants to create a superset of TypeScript called TypeStrict with literally the only change being that you have to pick between int and float, “number” is just too lazy ..

@r_marked @theo If it's a bad idea explain this. TheOS

@theo 🤦

So I kind of want to make an operating system…

@theo I was unsure, so asked Gemini why LLMs are so expensive and it basically said because H100's are "wildly overpriced" "1000% markup". LMAO

I think a lot of y'all don't understand the point of the $200/month Claude Code plan. It isn't an attempt to make a profit. It's a marketing expense. Those plans LOSE money. Sure, 80% of users will do less than $200 of inference. The top ones will go WAY over. It's easy to do $1000+ of inference on that plan. The people who do that become evangelists. They post about it. They tell their friends and coworkers about it. They become marketers. That marketing is only valuable if it drives lock-in. If they evangelize the $200/month Claude Code plan this week, and the OpenAI one next week, the spend was a loss. If they get locked into Claude Code and the Anthropic Agents SDK, Anthropic no longer has to have the "best model" to make their money.

@kenwheeler @HPNnetwork They are just on the far side of the container, with the drone above, so it has direct LoS I think.

@MonoBrainChris @HPNnetwork how would infrared projection work from outside the container

holy fuck lol

@kenwheeler @HPNnetwork The skeletons are Kinect tech it looks like. The drone above is attempting to track limbs.

@HPNnetwork upon closer review, they’re friendlies. which means they could have telemetry that could show that kind of skeletal visualization

@theo Google's AI studio is also fun. If you expand "thinking" your scroll bar locks to window bottom as it thinks, mouse scroll CANNOT change that, you have to click off message and hit up arrow. Maybe because of this virtualized scroll behavior with "loading nodes"...

Last time I tried to use Claude, I got: - Broken modal - Tons of error toasts - Broken input box - Weird banding in the logo bar - Hard errors on all message sends I genuinely feel bad for making fun of Claude so much, but it's truly abysmal. T3 Chat is 10x better with 3 devs.

@kenwheeler Also it's just incredibly inefficient. We have the primitives and software to design things really easily. Like how do you do support queries on a fucking M.C. Escher website?, "Sometimes options are a cog, but could be spanner... maybe refresh until you see one of them"

anyone who thinks they’ll be generating ui on the fly hasn’t had to product legible text outside of an extremely tightly coupled system

@ThePrimeagen Good tech ideas used to revolve around solving a problem and making peoples lives easier in some way. Software is having an appliances "internet of things " like moment but more lame. At least an internet fridge has uses. Literally just an AI chat but with sms latency...

such innovation also glad to see that they did cheap on on the commercial and ai generate it because it would have looked horrible, uncanny, and make me feel uncomfy

@zeeg @threepointone When I first started messing around with DO's with the help of Claude I literally thought it was making stuff up when it said .dev.vars

@threepointone good riddance!

RIP .dev.vars

@thdxr The main thing I judge an OSS lib on is the api. That shit matters. If you vibe with an api the lib often has the more advanced features you'd want, if you don't vibe, a lib can feel like speaking another language, sometimes literally.

my favorite kind of product to build is one that has a very simple, opinionated surface that you can use without knowing much but with built with a set of underlying primitives pro users can access and go crazy configuring and composing

@soychotic I'm fucking ancient and for some reason this Usher one sticks in my head as a turning point of, we will literally mimick anything even if it looks like scraping dog shit off a shoe. youtu.be/-yEx4q9a2pk?si…

I miss that era of music from 2006-2012 where everything had a dumb or slooty dance associated that you could do in front of your parents coworkers like cat daddy cat daddy cat daddy

@therealcliffyb It kinda lead the way in how 3rd person should be done on console as well. Like how do you "remake" something as genre defining as that now for todays gamers. I can't imagine we'll see another cover shooter any time soon.

@therealcliffyb Also Gears was very much a time and place game. It's a cover shooter with nonsensically over the top jacked dudes fighting an army of "bug" men that pop out of the ground. Yet despite that you sold it straight and made people emotionally invested in the characters....

WE'VE BEEN OVER THIS DUDE

@thdxr @jordwalke You just need to speed the animations up. People default to making animations seem natural. i.e. movement that feels biological and relatable. But as soon as you go to uncanny ludicrous speeds most people will think it's actually doing stuff super fast.

@jordwalke every time i spend a bunch of time doing a really slick animation i'm like wow this is awesome 5min later i remove it because it makes the app feel slower

One design philosophy I am kicking around: Your UI must earn the right to use animations. Animations are bold and require confidence. First make sure you you've eliminated layout shifts, flickers, jumps & delays. Unearned confidence is off-putting - both in people and in UI.

@fat @TkDodo It makes sense on paper, but everyone's got different requirements. Sounds like yours are pretty fine grained. I'm pretty curious.

@MonoBrainChris @TkDodo tim originally didn't want the no cache by default bc he knew this was wrong if you are actually obsessing about performance

@TkDodo I feel like I see this a lot where i'm guessing people lean too hard into RSC patterns. I assume once RSC fetched data is in client it will always be faster to tanstack-query/trpc invalidate rather than RSC(next) route invalidate. Just wondered if you've ever tested that?

@fat @TkDodo Obviously your use case is probably complex and I have no clue about it. I only tagged TK as was interested in his RSC thoughts. This might be useful though See youtube.com/watch?v=iM-Alw…

@fat @TkDodo Next doesn't cache fetches by default now due to the issues people had. So any RSC data should be fresh in more recent versions. And if you use a <HydrationBoundary> vs initialState in tanstack-query you'd ensure fresh data from the RSC updates client state. Atleast i think...