Andrew Prifer

#DreamMachine's carousels are probably the most complex UI I've ever built. We wanted a free-flowing, diagonally arranged, arbitrarily long carousel that would satisfyingly flow around other UI elements. Something that doesn't exist anywhere else. Read on for some behind the scenes. 🧶

@SebAaltonen @rikarends I also dislike the constant “entry level” qualifier. Then it should also wipe out future mid and senior level professionals too, because there’s no path to become one.

I don't like tech CEOs trying to cause mass panic to the job market. Jensen and Elon too. Excavator + crane didn't mean that construction (shovel) jobs got cut down by 10x. We started building massive bridges and skyscrapers. Early numbers show just that (much more commits).

@initjean It's always the same 3 episodes too

how many times are AI companies gonna try and reinvent Black Mirror???

Frenchduckmaxxing

Btw super hacky undocumented feature: your agent can send media by emitting MEDIA:. This is currently not designed to be invoked by the agent itself and it doesn't know about it by default, but you can put the following in your AGENTS file: **Media attachments:** To send media back through chat, you can call `echo MEDIA:` and OpenClaw will auto-attach the file. You can only send workspace-local files. If a file is outside your workspace and you need to send it, copy it there first. Skill scripts already print their own `MEDIA:` line after creating an image or other artifact, do not print a duplicate yourself unless you specifically need to retry or resend.

Taught OpenClaw to use Blender

@vladyslavmoroz I thought you meant it metaphorically but then I turned on the sound 🤦‍♂️

Can't shake off the feeling that I've heard this tune before, almost like another design tool used it somewhere Can anyone find it? 😆

@zeeg Personally for me, Codex confidently solved problems even on medium reasoning where Opus would get stuck in infinite thinking loops, but that might be harness specific.

I think a lot of it is due to the stark contrast between human intuition for the first pass of a feature vs all the hairs and outgrowths that code would naturally accumulate over time accounting for all the edge cases, which codex tries to spit out in one go. Another reason might be that the current research focus is trying to optimize for long running tasks on complex codebases, and the more defensively the model codes the different modules, the less likely it is to shoot itself in the foot later, and if it does, it’ll more specific information on what went wrong where so it can correct it.

Has anyone done any kind of research on the complexity of outputs from different models? Anecdotal, but I'm convinced that Codex is significantly wrose than Claude. I dont know how you'd reliably measure it, but even a subjective "yeah most humans would agree this is better"..

@haydenbleasel Awesome job, curious to see what’s next!

Today is my last day at Vercel. I've had an absolute blast over the last year. I got to work on a lot of my own projects that helped form the basis for our AI Cloud, helped other teams ship their great ideas, and worked alongside some of the most talented people I've ever met. I'll have news on my next adventure for you on Tuesday but for now, wanted to reflect on my time here and the projects I worked on. Read on if you're interested ↓

Yeah it’s the off center. I think a dev must’ve eyeballed the centering while anchoring to the right, which means it is slightly off center when it is on, and completely off center when it is off and the spacing changes with the brightness icon disappearing. What’s even weirder is that the center point of the arc is also not horizontally centered between the arc’s start and end points either, leading to the slightly lopsided arc when it is very short.

@AndrewPrifer Is it the fact that the arc is not centered, the fact it doesn’t go away when the flashlight turns off, or both?

This one annoys me the most. Because why would you obsess over every detail just to stop short of perfect.

“In IBM there's a religion in software that says you have to count K-LOCs, and a K-LOC is a thousand line of code. How big a project is it? Oh, it's sort of a 10K-LOC project. This is a 20K-LOCer. And this is 5OK-LOCs. And IBM wanted to sort of make it the religion about how we got paid. How much money we made off OS 2, how much they did. How many K-LOCs did you do? And we kept trying to convince them - hey, if we have - a developer's got a good idea and he can get something done in 4K-LOCs instead of 20K-LOCs, should we make less money? Because he's made something smaller and faster, less KLOC. K-LOCs, K-LOCs, that's the methodology. Ugh anyway, that always makes my back just crinkle up at the thought of the whole thing.”

I genuinely get where all of you are coming from as someone who wanted to be a 3d artist and then got into programming 3d engines. But I feel that the discussion requires more nuance and cordiality, because outside a very specific Twitter bubble, genai is anything but losing. This kind of mindless dogpiling is not the kind of “agreeing” we should embrace and will hurt your chances of convincing folks who aren’t already on your side.

@AndrewPrifer @andreintg Everyone already agreed with us lol supporting GenAI is the losing position and we weren't even getting into that. we just said he was wrong about ControlNet and image to image taking 3D data as input because that is literally just factually incorrect

ask me how my life's going since we announced putting AI on top of our SDF tech 😆

Hey with the new DLSS announcement you’ll have infinite material and everyone will agree with you. Arguing with an indie dev who is very well respected in the community is a bit of a losing position and the wrong audience (highly technical people who already thought in detail about all this).

@AndrewPrifer @andreintg We've been very patient and kind and it doesn't change the fact that he's wrong.

I am generally sympathetic to what you’re saying, but this kind of zeal and bullying tone is a very bad look. In my experience the folks most effective at shaping opinion about ai approach conversations with kindness and an open mind, which should be easy in this case, because you know that Andrea of all people put in the work and knows what he’s talking about.

@andreintg yeah you're getting corrected because what you're saying is wrong.

@RhysSullivan Finally I can replace jquery mobile

This might exist already but I made a mobile app feeling template for NextJS Swipe to navigate, haptics on button clicks, uses native UI I’ve been building more local apps for just myself, running them on my local network and it’s helpful github.com/RhysSullivan/n…

@tmuxvim @paularambles ChatGPT has been doing follow up hooks for at least a year iirc, but only recently has it started sounding consistently cryptic and clickbaity. “It’s not the one people usually pick” lmao the audacity

has anyone else noticed that GPT-5.4 often ends its responses with like, clickbait? it often promise to reveal "the one surprising X that will do Y" or something like that

@ctatedev @DamiDina How do LLMs decide between the 3 patch formats? Are they good at generating the one that’s shortest for the task or is there some special prompting required? Or some LLMs just prefer one over the other?

json-render now supports YAML as a wire format JSONL needs a full element before rendering YAML is valid at every prefix, going from element-level to property-level 💨 YAML looks like source code to LLMs And we use 3 standards they know: JSON Patch, Merge Patch, Unified diff

New Coke