Louis

Hey! I am in the midst of hosting a YCombinator Hack in Paris. If you are in Paris, show up! We have people building the next gen of AI Identity and more! It's quite cool. luma.com/8ucy347o

We just hit 200K game dev users at @forgegui. 100K was only ~15 days ago. The next era of games won’t be defined by who can learn the hardest tools or who has the most money. It’ll be defined by who has the best ideas, taste, and communities. ForgeGUI is making game creation accessible across every platform. Build on any platform, build any game, build with ForgeGUI. We’re hiring founding engs who want to work full time, in person, in sf, in a intense environment. DM me your background and the coolest things you’ve made.📩 150k-250K plus generous equity.

it took 1 intern 3 months of continuous work, but eventually, a quantization method that beat every other algo in the market, including @nvidia's official modelopt to explain why this matters, i ask for exactly 69 seconds of your attention (275 words @ avg reading speed of 238 wpm): frontier models (like glm52) are huge (~0.8T params). as released, each parameter takes 2 bytes (bf16), so overall size is about 1.6 tb a b200 has 180gb of memory. a node of 8 gives you 1.44 tb, barely fits weights, much less activations / kv cache must quantize the model (reduce the size of each individual parameters) to serve. fp8 quantization means each parameter takes 1 byte (fits in 0.8 tb), fp4 takes 1/2 a byte (fits in 0.4 tb) cutting the model to a quarter its original size is necessary for it to run a) cheap b) fast, and every lab serving models does this. but, quantization lobotomizes the model if not done correctly (this is why you see people complain about @AnthropicAI nerfing claude or @OpenAI nerfing codex) there are currently several algorithms (like Nvidia's official model-opt) that attempt to figure how to quantize a model with the least amount of damage. they find the redundant layers that can be slashed, and sensitive/important layers that need to stay in full-precision. these algo's have two drawbacks: 1) they take a long time to run 2) they quite often result in a sub-optimal configuration for the past 3 months, a research (and, as always, waterloo) intern on our model perf team (@the_joshua_hill) came up with a new quant algorithm. it consistently finds the optimal configuration: a) in less time than SOTA b) with more aggressive quant than SOTA c) scoring higher on benchmarks than SOTA achieving just one of the above is a feat on its own. all three...excited for the paper to come out this week

I built a vector-weighted mind-map of @thecollectiv3e. A living graph that reads members' skills, interests, and goals, then computes exactly who to meet and why. the math inside: ● aura → a member's skills, industries, role, and hobbies mapped to hues on the color wheel, drawn from ~30 signal types and weighted into a vector that blends into one gradient, unique to every member. ● matching → 13,000 pairwise comparisons across 165 members, surfacing each person's strongest fits. ● trace connections → find how any two members are linked or explore who your connections know (second degree). ● the layout → a force-directed graph - connected members attract + related people cluster on their own. ● center on you → see your strongest matches, ranked by aura fit. lmk what features I should add next 👀

Use @𝚌𝚘𝚗𝚝𝚊𝚒𝚗𝚎𝚛 queries so your layout responds to the space it's actually given, not the full viewport. This is especially valuable when something like a chat panel takes a slice of the screen.

@parisbayarea Then Meet me again bro, in London this time

Maybe I'm not depressed I am just in Europe

@drapzdesigns Yeah I just experimented with it and honestly just throw a wordmark on there after you get the bg

@lnv007 show me a midjourney created logo

Fable 5 is impressive, but it still can’t create logos like these.

Introducing Amicro, A curated library of premium micro-transitions and interaction components for React. Now you can add beautiful micro-interactions and seamless transitions to your website with just one CLI command ✨ Open source. Try it: amicro.vercel.app

we just wrapped our Spring quarter at @diabrowser. before i jump to the next big thing (!), a short behind-the-scenes on one of my favorite projects to date: artifacts! the design challenge: how do you make AI-generated docs that don’t *feel* AI-generated? (🧵 1/10)

Our designer @chrstnerode wrote about the thinking behind them here: x.com/chrstnerode/st…

AI isn't just changing the tools designers use. It's changing the way they think, prototype, and build. In this episode of Design Review, YC Head of Design @eve_bouff joins @aaron_epstein to share the AI-first workflow she uses to design products, websites, and events. Using projects like Paxel, @sotazine, and YC Startup School as examples, she explains how coding agents are transforming everything from rapid prototyping and branding to design systems, and why the biggest bottleneck is no longer software. It's imagination. 00:00 - Eve's AI-First Design Workflow 01:27 - Why She Designs With Her Voice 03:44 - Paxel: Spotify Wrapped for Coding Sessions 07:13 - Building Tools for Yourself 11:01 - Turning Coding Transcripts Into Feedback 12:53 - The Story Behind SOTA Zine 16:57 - Designing With Context, Not Prompts 21:07 - How to Get Better AI Design Outputs 23:20 - Building an Interactive Map of San Francisco 25:17 - Behind Startup School's Visual Identity 29:59 - The Future of Design

It's out. The sounds your UI has been waiting for. 10 cues · synthesised live · 0 deps · ~2kb cuelume-site.pages.dev

introducing dither-kit a library of gorgeous dithered charts built from the ground up, no dependencies, they're built on top of a tiny <canvas> engine install it today at tripwire.sh/dither-kit

I often want my @evedev_ agent to join a slack thread mid conversation and know what the discussion is about you can just enable threadContext, and the agent can read the full thread

@shardara @evedev_ Wow entire thread context is cool. Just built something cool with eve today. x.com/Vishal_anton16…

We've built a coding agent factory inside LangChain, and fully open sourced every component of it OpenSWE has: - coding agent that runs in the cloud - can tag it from Slack, GitHub or Linear - fully sandboxed, supporting @LangChain LangSmith Sandboxes, @daytonaio, @e2b, @modal and @RunloopAI - a frontier code review agent that automatically reviews GitHub PRs - GitHub OAuth support - A full UI web app for coding sessions and reviews - much much more We've been iterating on OpenSWE for over a year now, so you could say we know what we're doing here

>i got banned from OpenAI for "Cyber Abuse" >no idea what I did >paste the ban notice into Codex >ask it to figure out what triggered the ban >Codex found that I asked it for an API key to my own server >Codex writes appeal >Codex submits appeal >a few minutes later appeal auto-approved by some AI at OpenAI banned by AI, convicted by AI, defended by AI, and pardoned by AI in about 10 minutes

get this thing at gradient-spin.vercel.app