you can't hedge a worldview

When we announced @OpenAI o1 some researchers from other labs told me we made a strategic mistake and should have kept it secret so we could accelerate ourselves and pull farther ahead of the competition. Studies like these make me confident we made the right choice.

Announcing a new division of Midjourney called "Midjourney Medical"

realizing karpathy was not the person we wanted him to be

machines of selectively loving grace

hell yeah

AFM Core Advanced on-device model running on A19 Pro is a sparse model. It's 20B parameters. It's fully Apple designed. It is an MoE but when it processes the prompt, it only loads the parameters needed and locks them in. If it's 20B parameters total, but on a specific request it's only 1-4B parameter total. It only loads in 1-4B for inference and decides them at prefill time. It is fully Apple designed architecture, Google had nothing here.

Anthropics way of calculating ARR would be a pain the ass in the second half of this year good luck with that IPO

how much money does labubu have

GOAT

1.1. Question 28: What happened 1973 and 1974 when your investment firm lost over half? Charlie: Oh, that’s very simple. That’s very easy. That’s a good lesson. That’s a good question. What happened is the value of my partnership where I was running, went down by 50% in one year. Now the market went down by 40% or something. It was a once in 30 year recession. I mean monopoly newspapers are selling at 3 or 4 times earnings. At the bottom tick, I was down from the peak, 50%. You’re right about that. That has happened to me 3 times in my Berkshire stock. so I regard it as part of manhood. If you’re going to be in this game for the long pull, which is the way to do it, you better be able to handle a 50% decline without fussing too much about it. And so my lesson to all of you is conduct your life so that you can handle the 50% decline with aplomb and grace. Don’t try to avoid it. (applause) It will come. In fact I would say if it doesn’t come, you’re not being aggressive enough. 1.2. “I regard it as a part of manhood. If you’re going to be in this game for the long haul which is the way to do it. You better be able to handle a 50% decline without fussing too much. Conduct your life so you can handle a 50% decline with aplomb and grace. Don’t try to avoid it. It will come. And if it doesn’t come I’d say your not being aggressive enough”.

Remember when everyone hated/sold $META off 80% because their capex was like 30bn & 50% of CFO...good times

just fix the siri

A video showing what appears to be a naval mine purported to be in Omani waters in the Strait of Hormuz has begun circulating on various social media platforms, following the announcement yesterday from Oman's Ministry of Defense warning mariners of a suspected naval mine in their waters. While the authenticity of this video can not be verified, the mine appears to be most similar to an Iranian Maham-1 moored contact mine.

sorry but this is fucking ridiculous what do you mean they are using firefighter hoses for a computer

AGI if this saturates

this

There's an arXiv paper from 2 weeks ago that the finops community hasn't absorbed yet. The authors ran identical agentic tasks. Same model. Same prompt. Same context window. Same tool stack. They measured end-to-end token consumption across many runs. The coefficient of variation was extreme. The same task could cost 8 dollars or 240 dollars, 30 times apart, with no change to any input. This is not a bug. It is a feature of agentic execution. The model decides how many tool calls to make. The retrieval layer decides how much context to pull. The verifier decides whether to loop again. Each of these is a stochastic decision conditioned on intermediate outputs the user never sees. Most enterprise AI procurement assumes the opposite. Cloud compute has tight variance around its mean. You can size a budget. You can quote a fixed price. You can sign an MSA on a per-million-token rate card. You can still do that with agents, but not with the old mental model. The piece of this that nobody is pricing yet is what happens when the procurement function at JP Morgan or BCG or Pfizer realizes the math. They will not just ask for cheaper tokens. They will ask for tighter controls on variance: better routing, better caching, better observability, better policy limits, and in some cases dedicated capacity. That is a very different infrastructure requirement from stateless inference. It does not mean cloud disappears. It means the winning cloud looks much more like an execution and control layer for stochastic workloads. The unit of AI cost is not the token, it is the distribution. That is also why I think platforms like @nebiustf and others matter.