People will tell you excellence can only be achieved through discipline, which is not a teachable skill. But there's one other path that I've observed, and that path is obsession. Obsession can be contagious, if it's powerful enough. You can gauge a person's discipline by how

$400,000 USD in GPUs and RAM, 52 Pentillion Gigawatts of electricity. Just to come up with the prompt.

Obsession drives excellence. That's what I always say. Well, it's what I started saying just now. So consider it retroactively "always said." Thank you for your attention in this matter.

Don't forget, you can still read the docs, if you want. If they were good enough for Candalf, they're good enough for me.

The new Grok feels like a old dude who dresses up like a younger guy trying to be “hip” and “cool” without realizing how abjectly disingenuous he sounds, often doubling down on his blatant and insidious insincerity when pressed further. Missing the mark pretty badly.

we would have AGI already if the training data didn't include Reddit

Straight facts. Just do the thing. You are your own ideal first customer. When you shop for venture capital you are shopping for bosses — and chances are they will be an inferior customer compared to you. Product led growth and venture capital are fundamentally at odds.

Good (or benevolent) and bad (or malicious) things can both come through portals. As Rilke said, “every angel is terror” We might not know the difference between angel and demon, proud as we might be in our convictions.

> Qwen 3.5 397B NVFP4 > Clocking in at ~80 tok/s on my machine > Actually seems smarter than Opus in some ways, especially in terms of UX and design > It feels verbose, but thorough instead of other models which feel noisy > I am ranking this as #1 right now for agentic work

This has already been a trend. Companies get fat and decadent. A lot of people contribute very little once you have *thousands* of employees. The stocks went up because the market is more confident that the company won’t die. And people are like “HOW DARE YOU?” How dare they

"1000 flies on shit can't be wrong" One of my favorite sayings, because they definitely can and usually are. OpenAI's 120B param model just got smoked by some tiny 9B from China. They are laughing at us. For eating the surveillance slop ChatGPT sandwiches. Own your computer.

Prediction: Within 6 months the US Government will be running on Qwen 3.5 397B NVFP4 Abliterated

This is my life now. I pace around my yard, holding a MacBook, talking out loud to Claude Code on my inference box over SSH, while monitoring the GPUs in a separate window. Haven't sat at my desk or used a mechanical keyboard in a while, and can't say I miss it all that much.

You will eventually need a license to own a computer. This is where all this "Age Verification for OS" stuff is coming from. It's not about protecting kids. If it were, the CA law would not demand that you share this age data with any dev that asks for it. It's about

This is the way. Try it with any desktop you may have laying around. Your get the portability and ergonomics of your laptop, with the power and strength of the desktop.

The way we teach math to students is insidious by design, to create a world of people who feel anxiety and demoralization when it comes to math. Think about the average person. They are afraid of math! This seems cool, if I had to learn it over again I would try this:

Ever hear someone say something like, "Oh yeah? Show me the peer reviewed study!"? This is a sort of red herring leveraged by midwit sciencists* to justify their existential beliefs about reality. Effectively a mind virus to disempower others from recognizing the grandiosity

i am no longer sure dario is conscious

This is at least a year behind open source. Something insidious is going on with the folks at xAI. Usually when things like this happen in tech, it's because the folks up top aren't making good hiring decisions because they lack the discernment to evaluate good candidates.

Regarding the "Buy a cluster of Mac Minis to run Local AI" hype train: One of the notable downstream effects of this is that it discredits the local AI movement at scale. If enough random people had a bad experience, wasted money, etc, it slows down the movement, and those