Eyas

I'm not a consciousness proponent, but this strikes me as missing the point. We don't know how to measure that a fellow human is conscious (except that they tell us they are), and since LLMs trained on human text are also likely to output that they're conscious, the AI industry kind of put us in a place where we have no way of finding out?

Worth considering in the current climate. My impression a while ago is that the exact water and power usage of any large company's data center is seen as a trade secret of sorts; if a large company invested to become XX% efficient, it might not want competitors to know it hit a certain efficiency target.

Some things to correct here:

- While drugs laced with fentanyl are also a problem, lots of people _do_ knowingly consume fentanyl. In fact, it has replaced Heroin in the US 'market', for example. It is also a contaminant in other drugs (cocaine, meth), due to how easy it is for fentanyl to stay in crevices, etc. It is also 'snuck in' to counterfeit pills, which is the only aspect you mention.

- The traffickers physically moving fentanyl into the country are overwhelmingly Americans. Pre 2019 traffickers used to get this straight from China, but stricter controls in China have shifted things to be staged in Mexico instead.

- The demand and traffickers are still largely American, and it is a "symbiotic" problem. US drug demand is empowering Mexican cartels and making everyday Mexicans less safe (the cartels also get most of their guns from the US). At the same time, powerful Mexican cartels mean it is much easier to "stage" fentanyl in Mexico then import it into the US than the older model of importing it directly from China.

The best pattern I've seen is hash the request and validate that it is the same one. If not, error out as it is an invalid argument.

"Best guess" can be bad if it is not well-defined, but you can still make error detection obvious rather than hidden.

Was I the only one that found `docs.fastht.ml/llms.txt` more useful than both fastht.ml and docs.fastht.ml?

Zooming out, it's interesting how many (especially dev-focused) tools & frameworks have landing sites that are so incomprehensible to me. They look like marketing sites but don't even explain what the thing they're offering does. llms.txt almost sounds like a forcing function for someone to write something that is not just more suitable for LLMs, but humans.

This ties in to what others are saying: a good enough LLM should understand a resource that a human can understand, ideally. But also, maybe we should make the main resources more understandable to humans?

It probably can in a .cc file but if you're importing another library and just have access to the header, it wouldn't know how to.

Honestly I kind of just enjoy writing. My blog has gotten tons of views and still gets a fair amount of traffic, made on HN's front page a bunch, and have some post translated to a few languages.

I haven't actively tried to monetize it (though I'd like to shift to income based on my own work rather than be tied to a company). I tried putting ads once in a while and it was ~decent but negligible, and kind of ruined the vibe. I have done a few one-off consulting things out of it and got some nice side-cash, but nothing meaningful.

What I do get is feeling engaged with the wider tech community. Seeing common questions and comments. A feeling that what I'm saying might resonate with some people. Interesting discussions on Twitter and HN. A few podcast invites, etc.

"What did the Arab"

1. say the moment he left the White House? "No wonder they hate us."

2. man say to the Arab woman at the bus stop? 'Your dad's the bomb!'

3. guy say when he saw a woman driving a car? "Finally, a woman who knows how to use a clutch."

hmmm

Right, but their claim is that there were proper separations that successfully did limit the blast radius.