ok123456

Closer to 95% if you count Teams.

https://dl.acm.org/doi/pdf/10.1145/237170.237260

Related: The authors wrote a paper on their design of the layout engine.

7-OH is the name for derivatives of kratom that contain the active ingredients.

It has opioid-like addiction tendencies.

Lots of people who used kratom to wean themselves off opioids are now addicted to 7-OH. This includes many people over the age of 30.

Strong appears to be a fitness tracker. It's in competition with a spreadsheet.

OnX appears to be heatmaps as a service for outdoor recreation. They have to cover their opex costs.

These are not the subscriptions people have in mind when they say they hate the subscription model. People dislike obvious rent-seeking behavior; they really dislike it when there is no technical reason something can't be sold to them in a one-time transaction; and they hate it when a product that was offered as a permanent one-time purchase is then only available as a subscription service.

I like subscriptions because, in the worst kind of corporate management speak, subscriptions align our interests. You pay for the app for the duration that you see fit.

We see the same pattern over and over again: (1) subscription service starts and operates at a loss, (2) people recognise this and sign-up, (3) the service gradually enshittifies to the point where there is no real value propisition, (4) company banks on having enough customers from 1 and 2 that enough people will put up with 3 that they can remain profitable.

How does this align with my interests?

Literally the "new old thing":

https://www.andysarcade.net/store2/all-other-stuff/vintage-c...

aka FIPS.

The OS handles all of that transparently, without requiring any additional code. I think that is the draw.

Won't somebody please think of the Chinese cyber children!

Peiter Zatko profits, personally and professionally, from conflating the two. Making such bold pronouncements is a way for him to stay relevant.

Also, I've known some thoroughly unimpressive people who have affiliated themselves with DARPA. I wouldn't use it as an appeal to authority.

Pretty unimpressive as security vulnerabilities. It would be better to just say these are simple bugs for the most part.

Or maybe they have different incentives and don't see providing an inference as a service for the entire world as a viable business model.

Open source isn't a zero-sum game.

I think China "advancing" faster is better. Hoarding some magic numbers on a hard drive isn't a sound business plan, and shouldn't be what we're betting the farm (literally) on.

Yes. They're turning on the consent manufacturing machine to make it an issue of "national security" to download some gguf file from Hugging Face. Absolutely disgusting.

Use gemini-cli for mcp support.

So I define the data type to be "asdklfjaslkdfjiolsadfjoiusadfoiasfoikasjfdoisadf" and give you a decoder for it.

For images, it makes sense: people dealing with 16k x 16k PNGs are uncommon. Give them an error message that tells them the setting to bump. But what should be the threshold for "big data"? I'm sure it will follow Zipf's Law, but the tail will be fatter.

Stranger danger was always rare. It was a media hype-cycle. Child abductions have always predominantly been by family or other close persons.

How do you prevent compression bomb attacks when files can define their own compression functions?

You could have some kind of OOM killer, but that will be a "footgun" that people who are actually doing "big data" will constantly shoot.

This pretty much kills any ingestion pipeline where the source is untrusted.

No. I'm stating where the paper's assumptions are clearly violated.

AI companies are intentionally trying to monopolize the supply of inputs needed for R&D. This violates homogeneity of degree 1.

This completely breaks down under the current reality of AI investment, as players large and small are no longer price-takers. The marginal costs of investment are not constant because we have finite supplies of GPUs, TPUs, memory, hard drives, and power. The Hamiltonian in equations 5 and 6 needs to account for this.

Exposing services that use io_uring is a hard pass. It's only been a handful of weeks since the last security advisory.

stringstreams and fmt are a godsend compared to printf/scanf, which have historically led to most memory bugs in the first place!

Printf/scanf are implemented as variadic functions without type checking and rely on the compiler to perform its own internal metaprogramming to inspect and warn about format mismatches.

Anyone advocating the use of the old cstdio as a primary design decision about which C++ language features to use is not serious.

No exceptions or RTTI make sense in an embedded system that needs to ensure determinism, but are arbitrary and unnecessarily hobbling for high-level systems and application programming. How do you do runtime method dispatch without creating vtables and RTTI from first principles? How do you propagate a runtime error deep from the bowels of a component all the way to some top-level event loop? The "orthodox" approach would be a mess of integer return codes with associated enums (and none of that enum class nonsense!). No Thanks. It's clear the author has no idea what he's talking about.

Google's TPU and Groq's LPU are the only real, commercially viable ways to provide all the compute required for the inference people want.

Just having an ungodly amount of capex by blanketing the Midwest with datacenters full of GPUs is a disaster in slow motion.

And now you can do all of that locally with qwen3.6:35b.

We hear about a few big famous ones in the news here, but most of it goes completely unenforced.

So much for "Hacker" "News".

Without Musk, they wouldn't have squandered their market advantage with high-profile failure after high-profile failure. They would have released a good enough pickup truck in 2017. They wouldn't have spent a whole decade gaslighting everyone about self-driving cars.

The real reason Tesla was remarkable, and why Musk bought into it, was that they released a mass-market electric car that didn't look and perform like a vacuum cleaner and that got an okay range for commuting. That was the value proposition. China can now exceed those design parameters at a wholesale price of about $8k USD.

They're motivated not by the actual loss, but the checkmark of having attestation for a compliance framework.

So the fact that Microsoft let remote hands-on-keyboards in the PRC fix problems on GCC-High Azure nodes used by DoD contractors doesn't matter, since they're too big to censure in any meaningful way without impacting tens of thousands of businesses that rely on them to get a letter that satisfies a compliance assessor.

Actually knowing what you're doing, or being able to critically assess the risks of using a specific provider, doesn't matter.

If it fails, you fix the clone and push it back.

I use local remotes all the time for testing as a form of "local CI".

Check it out from '/tmp' and make sure it still builds.

For a single-dev or small team, it beats having to do github runner epicycles to accomplish the same basic goal. Add in Firejail if you want environment isolation.