vish @vshdeep

Everything you own was manufactured. The knowledge of how (which parts, which materials, which machines) is concentrated in a small number of companies and countries. More people who can read a BOM means more people who can make things. And making things is where most of the economic value is created. BOMwiki tries to lower the first step: seeing what products are actually made of. bomwiki.com

A friend just revealed to me that the middle name of Benoit B. Mandelbrot is Benoit B. Mandelbrot.

@kenanhsaleh @a16z @speedrun Would love to chat about my work on AI for clinical trial infra :)

Unpopular take: the bottleneck at 950 tok/s isn’t the model anymore. It’s you. I’ve been thinking about what breaks when agents get this fast and it’s human review. We’re not wired to supervise something that outruns our reading speed. So I’m building behavioral monitoring for agentic loops. Think a system that notices when your agent starts going off-script before it merges something catastrophic. SWE-1.6’s own story proves the point bash-over-tools went from 97% to 11% through post-training. That’s exactly the kind of behavioral signal I’m training my system to catch in real time, on models that haven’t been cleaned up yet. At 200 tok/s, you have time to catch mistakes. At 950, you need a system that catches them for you. Would love a shot at the Windsurf Max plan building this fast requires moving at the speed of the thing you’re watching

This is a type of chart crime: when you plot value tuples from a timeseries on a 2D scatter plot like this as if they were independent samples, you are leveraging temporal autocorrelation to magnify any existing x/y correlation and hide the actual variance. You're also hiding temporal distribution drift (the typical value range for x and y, and the relation between them, change over long periods)

Patient Investor @patientinvestor

3 months ago

Howard Marks: "When you buy the S&P 500 at a 23x P/E, your 10-yr annualized return has always fallen between +2% and –2%, IN EVERY CASE, EVERY CASE!"

125 371 2K 1.3M 2K

Huge congratulations to Bodhana Sivanandan on becoming England’s top female chess player at just 11 years old. We once played each other in the Downing Street garden. Let’s just say her success has not come as a shock!

If you sit at the center of the largest supply chain ever built, you set costs, define standards, and capture the highest-value layers. Regional hegemony can’t get you there. Scale can. The primary goal is for China to make itself indispensable to the world.

People assume China wants “global hegemony” in the traditional sense. That’s a wrong frame to look at it. It’s less about domination and more about centrality. The real playbook is that China wants to embed itself into the core of global production: - resources flow in - manufacturing spreads outward (Vietnam, Indonesia, Mexico, etc, for cheaper labour than China) - coordination, standards and scale stay centralised in China Rather than controlling just territory. It controls the entire system.

Jeet Heer @HeerJeet

3 months ago

Why are people assuming Chinese elite want global hegemony? Wouldn't regional hegemony suit them just as well?

120 34 686 186K 41

Bonsai proves if you’re willing to invest in the training methodology (QAT, distillations, etc.), you can represent 8B params in 1-bit and remain competitive on benchmarks. The implication is that we’ve been optimising for “bigger model in standard precision” when “same size model in extreme precision + training cost” might be better for most deployment scenarios. The bet here isn’t “small is better”. It’s “the right precision matters more than we’ve priced it.”

PrismML @PrismML

3 months ago

Today, we are emerging from stealth and launching PrismML, an AI lab with Caltech origins that is centered on building the most concentrated form of intelligence. At PrismML, we believe that the next major leaps in AI will be driven by order-of-magnitude improvements in

176 587 4K 1.3M 3K

i think we should all experience something like this atleast once in our lives

Wholesome Side of 𝕏 @itsme_urstruly

4 months ago

Bro pulled off the best video of his life

999 29K 196K 23.7M 29K

Today marks #WomensDay. Let us renew our commitment to recognize the equal dignity of man and woman. Unfortunately, many women, from childhood onward, are still discriminated against and suffer various forms of violence. My solidarity and prayer go to them in a special way.

The more I play around with GPUs & renting them, the more I agree with Don Wilson @drwconvexity. The only sensible way for this to converge is for the market to resemble power markets. Abstract the market to compute, trade futures & have merit order stacks of spot availability

What an iconic shot. Finding pleasure in the *process* of high pressure competition, rather than the win itself, is the key to lasting success. It’s really hard to experience both relaxed joy AND elite performance at the same time, but she nailed it.

Have a child going through the selective college admissions process for the first time and the system seems designed specifically to maximize stress for student and parents alike. The process has gotten worse in recent years and will will keep getting worse until it breaks. 1/n

@bennpeifert Ah right, with two weeks left, the put retains extrinsic value with no intrinsic value at spot = strike. Shorting it is short vol + long theta

A guy wanted to adopt a cat, but he wouldn't leave his brother's side, so he had to take both of them.

All parallel operations can be performed sequentially, but not all sequential operations can be performed in parallel. This failure to parallelize is what we call "causality". (Holds true for general relativity, quantum information, Petri nets, monoidal category theory, etc.)