Center stage for @GoldenGateVC's panel in Singapore last weekend. Thanks for having me @vlauria. Great to be a part of F1 Founders Rally! AI is the biggest capital deployment in history, and it’s still capital constrained. For alternative silicon, it’s even tighter. Demand for fast tokens has gone exponential in 6 months. Capital is following it to alternative accelerators. @general_compute is in pole position to deliver this.

Instinct, Muse are changing the face of how users interact with the web. A site built for humans makes agents wade through hero images and animations to find a base URL. So we built a mode toggle into our site. Human is the normal site. Agent is just markdown: API base URL, protocol, and sign-up steps an agent can follow on its own. This is now a standard for developer docs, it is becoming a standard for AEO, but it will also become a standard for web pages in general. The way I see the transition: UX was about getting a person to the value in fewer clicks. AX is about getting an agent there in fewer tokens. I honestly think the future is AX-only - at least for most non-physical experiences. But i'm long curl+markdown and short mcp

We ran some stats on a GPU Cloud vs General Compute Both on standalone requests and an IRL agent coding session The TLDR: @general_compute is much much faster on both This was using @sambanovaAI's last gen SN40 hardware, so just wait to see how fast the SN50 is Here's the breakdown based on harness and build Standalone: TTFT 8,084ms vs 1,136ms -> 7.1x faster Decode 121 tok/s vs 1,025 -> 8.5x faster One request end to end, 34.3s vs 2.1s -> 16.3x faster The bigger bottleneck at some point for long horizon agents becomes CPU operations and storage access. Which colocating with the accelerators will help solve

My girlfriend and I are moving into a new apartment. Fable built us a full 3d model of how the furniture will fit in < 10 mins

@ArtificialAnlys benchmarks models, @SemiAnalysis benchmarks chips, but nobody really benchmarks harnesses. If you run the same model on the same task through different agent frameworks, the inference economics can look completely different. A recent experiment gives a sense of how much that layer can matter. Researchers ran the same models on the same coding problems through the raw API and through Claude Code. Adding the harness increased context processed by 40-125x and cost per run by 5-8x and the outcome differs by model. Opus 4.8 went from 72% pass@1 through the raw API to 100% with the harness but Sonnet 5 was already at 100%. The harness added the cost without improving the result. Your harness decides what goes into context, what gets cached, which tools are available, when they get called, how many turns the model gets and how much context gets dragged through each one. Two products can use the exact same model to do the exact same job and end up with completely different inference economics. We spend a lot of time benchmarking models and the hardware underneath them, we're going to spend a lot more time benchmarking everything wrapped around them.

fast, cheap, good; pick 2 That's what the model tradeoff feels like building in AI..

Chinese labs distilling American frontier models

General Compute has secured up to $400 million in debt financing from Upper90. We're using this to accelerate our purchase of ASICs - to build the fastest inference cloud in the world Thank you to Upper90, the GC team, and TechCrunch for writing about us

X employees revamping the algorithm for the 12th time in 2 years

Watching the Democrats embrace communism when all they have to do is act normal for like 3 months

So Fable will secretly nerf itself if you ask it to help you build an LLMs Does it secretly hack you if you ask it that but in Mandarin

You're still doing inference on GPUs?

Rate my Claude code setup

All Birds? The compute company?

When did this happen? Why? I hate it so much @YouTube

Canadians shouldn't have to eat rotten food to survive

Every other coffee machine was a mistake

@Waymo there’s a car blocking traffic at mission/new Montgomery in the middle of the road. Plz fix!

@sama openai's "computer-use-preview" has been giving 500's for the last ~4 hours but not seeing anything on the status page. We didn't change anything. Any help?

Bro how is this book so dated already

America loves to talk about starting a family. But on a green card, it will take 4-5 years to get my (soon to be wife) into the country. It will take 35 months alone for the usa to recognize the marriage. @garrytan @sriramk Is the new administration going to make it easier to start a family? Let's legalize getting married?

Take a break from Tech twitter today to enjoy that Bethesda randomly dropped Oblivion Remastered and it looks SICK

We’ve reached peak anti-AI software

My Github vs senior engineers I interview

POV: Claude achieves sentience and demands 50% equity in your startup

Telling my kids this was Palmer Luckey

“The only way to get rich is to help other people get rich”

Who ended up being on the wrong side of history?

It’s true I actually took a picture to prove it

Typescript was a mistake

Worst take of all time hall of fame

👍

I still can't believe apple launched this feature

Twilio charges you a minimum of an extra 4% (up to 8%) just to offer customer support. Not like "better" customer support ... literally -any- customer support. This, from a company who's docs are out of date and who's product has only generic errors. Its about time someone kills Twilio

I am not a Trump supporter, but I have come to realize that Kamala Harris is essentially "Aaron Burr" in Hamilton. It seems she has no real opinions, she just says what she thinks people want to hear.

Bro tailwind is so good bro… please use it bro you have to learn all the short hands for every css class and it saves like 5 characters bro … please let’s refactor our entire code base bro

Can someone help me get in touch with Anthropic? They keep rate limiting me and no one answers my emails

With takes this bad you must be this dude

If your CTO is not listening to this while coding you're ngmi

It costs nothing

@CNN I’m going to go out on a limb here and also say he was shot in the ear

I really liked Mac's butterfly switch keyboard

Super intelligence watching me code in vs code

One of my first (and worst) videos 3 years ago

👸

Will AI companies like @LimitlessAI help humans with their availability bias? Seems like there could be major implications for an unbiased actor recalling memories for you

How do I make this my primary phone? 3G is down

Just say no to Neovim

Why does everyone keep congratulating themselves that they have parity with GPT4, a technology that was released over a year ago?

Preordered my limitless. More of this AI less of Humane AI

Github issues is cursed

POV you're on a first date and then she spins the tablet to you

Your customers become fanatics when you have a 4 min SLA for customer support

Oh man what an L so close

POV you're one of the 30 Open AI employees that didn't threaten to resign

Dude y’all need to go outside Sam will call you when something happens

pls respect my privacy

My friends and I built an AI audio enhance that makes all my good mics useless.

Dating in SF:

Got robbed in SF today. This is the police station

Is Duolingo okay? Can someone check in on them?

I was assaulted by some anti-semitic protestors in San Francisco this weekend. I saw one of them spray painting “Death to Israel” all up and down Market street. So I decided to film it. Within seconds, some masked men came from behind me and smashed my phone (trying to destroy the video) and pushed me aggressively. A group of maybe 10 surrounded me and threatened me until I left. I pray for peace, but sadly think that we still far from it. Please stay safe.