da-x

It's all about lowest friction + domain-name trust.

Depending on third party packaging (distribution-validated install) is much higher friction.

Launching today is all about sponsored distribution channels and less about posts in specific forums.

But let's hope this one beats the statistics..

I've read elsewhere that it's 10x less electricity for inference workloads compared to standard GPUs. It is not clear to me, are the model weights built into the silicon (e.g. per model tapeout), or is this a new kind of chip architecture that still has weights in DRAM/SRAM?

I'm glad to see this. Almost 18 years ago I implemented a similar kexec device+memory preservation for a storage vendor. It was done on a Linux kernel of that day, and it had had a memory reservation and handoff protocol between the two kernels to keep some specific PCI device alive, allowing for state restoration at the application side. I'm proud of the fact that the kernel replacement was just under 1 second in execution (after init process optimization) and the whole kernel+app was less than 10 seconds.

Thankfully there's livepatching (e.g. https://ubuntu.com/security/livepatch )

Everyone is cooking their own harnesses (e.g. mine is a very simple tool-use loop: https://github.com/da-x/da-harness ).

Perhaps you can also provide a Tauri-based independent downloadable app.

ReactOS is already some sort of Windows Lite. And now with coding agents we can accelerate development on it.

You're on the spot, it can certainly be a self-defeating prophecy

https://en.wikipedia.org/wiki/Self-defeating_prophecy

I wish there was an objective index for 'software bloat levels' so that we can better track how stuff like this happens.

Qwen3.6 27B on RTX PRO 6000

> When Mark pointed out this discrepancy, the patient commented that her left hand frequently did things on its own

That is the scary part. I would image it's like having a parasitic mind that sometimes controls half your body and sometimes just smooth along with what your 'talking' part of consciousnesses is doing.

Certainly there's some Rumpelstiltskin kind of deal when trying to move fast with a vibe-coded "product".

Anthropic: It seems you feel our work is not a benefit to the public.

Me: LLMs are like any other machine. They're either a benefit or a hazard. If they're a benefit, it's not my problem.

Can human institutions regulate matrix multiplications that contain certain weights?

Lessons from history: can you regular data transfers that contain certain data? AFAIK torrent downloads for TV shows and movies that otherwise get streamed, is still live and kicking.

We will only know if that's true in retrospect.

Cramer on Tesla IPO: “You want to get some on the deal and then get out,” Cramer said. “I think it’s going to be a bad company.” https://www.cnbc.com/2010/06/28/cramer-on-tesla-ipo-get-in-g...

For the majority of people, it seems to me that news are background noise to which they tune in occasionally, but otherwise they are just going about in their lives and it does not affect them much.

However, for the few details-oriented analytic people among us, the news are a *mental pit hole*. This is what I saw on myself since 2020. The mind works under an illusion that there is a possibly to synthesize some positive a change in your personal life based on information from the news - but it's wrong. It didn't keep me from browsing though, and getting addicted to those information streams, just like sugar.

So occasionally I do a 'news detox' like the OP describes.

Surely social developments can affect you and your neighborhood, your city, or your country on the long term, but in that case it is better to consume a monthly or yearly digest.

Or a daily digest. I vibe-coded a daemon that sends me the Wikipedia's daily summaries, once a day. It's a one page of 'this is what happened' without interpretation (but with left-leaning bias, because Wikipedia).

It's not only that. Also, absorption of calories from food is largely different between individuals, due to gut microbiome, generics, etc.

While I like the idea of starting a hobby OS, but the most despairing thing about it that to be useful it would always have to implement 'legacy' interfaces of other OSes. That is, it cannot stand on its own and create a new ecosystem, instead it has to interface to the world and implement TCP, POSIX, know formats widely used files and such.

You end up with an OS kernel that talks Linux/Win32 and takes on a lot of compat code, protocols, and other paradigms.

I wonder what a hobby OS would have looked like it if it assumed nothing, that is, as a thought experiment, as if aliens on another planet invented computing and started writing OSes from scratch. Imagine we discover software from another planet that would not even work with 8-bit bytes, for instance.

Democracy also has a 'tyranny of the majority' vulnerability in which it can enforce policies onto minorities that could be just as harsh compared to policies imposed by autocracies. For example - Japanese American internment in WWII.

The sun performs fusion, not fission.

If there are no incentives to be good at coding, almost no student would be good at coding unless they are picking it up as a hobby. Do collage tests coding proficiency? I doubt.

I'm sure they are banking on the possibility that the average book hoarder cannot tell the difference.

The "fuse the militaries / more susceptible to Israeli whims" framing seems to be the hand-wavy part of those articles. If you read the referred Section 224 you'd see details of the expanded scope of cooperation in tech domains and info sharing, and also in transparency and oversight.

Israel itself is in favor of this, to reduce US influence on its arms exports. Everybody wins.

Maybe someone can devise a distributed bench-marking system where multiple people collaborate on tests and also vet each other's tests and rating without revealing them to the public.

I have my own "interview questions" for models where I give them a premade Git repo and a problem to solve. Then, I rate them like a teacher. I believe other do that as well, so we only need a reliable system to aggregate these results.

I bet code removal is _not_ something that goes into LLM training sets much, unfortunately

Ah, it's somewhat like how Jeri Ryan's divorce caused Obama to get elected.

Source: https://www.reddit.com/r/startrek/comments/187ghq/star_trek_...

I think there's some sort of 'social pressure' between nations. Policies during the COVID-19 era come to mind.

I think they will have the revenue for this, and also - it is on par with the capital expenditures of the railroad industry in the 19th century in today's dollars. Given what they trying to do, it makes sense.