You don’t say
We vibe coded the Epic Games Launcher to be faster than the other guy.
Don't believe us? Test it out for yourself in Preview:
launcher.epicgames.com
The discourse around consumer AI is so boring. Close your eyes for ten seconds and then open them and look at the world around you.
The greatest technology of our lifetime and all anyone can imagine is booking flights and buying things.
People don’t want to be more productive.
2
2
341
Don’t trust anyone with your data, and don’t ask your users to trust you.
Start from a solution that requires zero-trust and build from there. That’s what we’re doing, anyway.
9
447
OpenAI will steal years of your work and you’ll be happy about it
We’re sharing a solution to the Navier-Stokes Millennium Prize Problem, one of the deepest problems at the frontier of mathematics.
The proof was produced by a group of agents, using an OpenAI next-generation model significantly more capable than GPT-6 Astra.
The problem concerns whether the description of smooth three-dimensional fluid motion modeled by the Navier-Stokes equations can break down. It has remained unresolved for roughly 90 years.
286
The real world is messy. Clean up your input.
Having visualizations for debugging your tech really hits different
1
360
FLML OCR is now 2x faster and 35% smaller now that we can selectively skip JIT compiled execution and opt into AOT compiled execution plans
1
3
340
“Taste” is the accumulation of a lifetime of individual and extraordinarily niche perspectives and preferences on a wide variety of things that shape who a person is and the things and spaces they create.
Today’s LLMs, even post-trained on a Rick Ruben book with the temperature dialed up, are an amalgamation of human creativity jammed into a statistical model and incapable of producing anything novel or uniquely aesthetic from a design perspective. Any positive output is the direct result of you feeding in a solid design system and coaxing it over the finish line.
Hire a designer, read a book, save some money, and create something noteworthy. AI cannot have “taste” because these systems do not have general intelligence or preferences - and anyone saying otherwise is selling you a bridge to nowhere.
We’re excited to introduce Taste Labs.
Our mission is to end AI slop. We’re building the data and infrastructure layer to give AI models and agents taste.
And today we’re coming out of stealth, announcing our $18.5M seed funding, co-led by @CRV and @AmplifyPartners
AI has nailed objective domains and made it easy to generate anything. But it still feels off. Now, the challenge is judgement. What fits, what feels like you, what’s GREAT. This requires turning a fuzzy, subjective domain into something we can measure and codify. We’re starting with design.
There are two sides to cracking this, the foundation model layer and the agent layer:
- We’ve already been working with the top frontier labs to evaluate and improve their models, crafting the right post-training data and RL environments.
- We’ve also been working with app-layer companies to build the context and verification tools for their agents to produce better, more on-brand, more creative outputs.
We want a future where AI feels right.
If you’re passionate about this mission, join us!
Their mission and product are incompatible because their solution to the challenge of explaining and codifying “complex”concepts is to impose a rigid set of parameters on design, inadvertently introducing a bias that skews the output towards a single, uniform direction where everything is the same.
There are far too many untraceable decisions that go into the creation of design and art that it’s an intractable problem to try to index or label them in any meaningful way.
18
873
People keep pointing to MLX as the answer for local AI on Mac.
But nobody ever mentions the real bottleneck: you can only run one thing at a time. All your memory or nothing. That’s not how useful AI works.
We’ve been optimizing end-to-end for something different — maximum occupancy, minimal latency, concurrent inference.
848M parameters. 80ms compute. 200 MB memory. 4K HDR at 90-100ms, real time. Still room for improvement - and the requirements for larger models can be even lower.
Local AI needs to drive real apps, not just toys. More soon.
At @6over3 we build our infrastructure to squeeze every drop of performance out of modern hardware. That means going against the grain and building from first principles.
Our latest paper is on the data interchange format we built to do exactly that — up to 213× faster decode, 1675× faster on embeddings, and 86% of peak memory bandwidth on large records. 👇
the secret to fast AI inference is just to hire anyone who has worked on a game engine
Introducing TurboQuant: Our new compression algorithm that reduces LLM key-value cache memory by at least 6x and delivers up to 8x speedup, all with zero accuracy loss, redefining AI efficiency. Read the blog to learn how it achieves these results: goo.gle/4bsq2qI
3
555
