wyager

> Shareholders are pricing companies as if being a winner in the AI boom is the most important thing

Well yeah, that's the most obvious way this plays out

The stock market should reflect reality, and sometimes reality is volatile.

These days you can get a cheap-o chinese fiber splicer kit for like $500-800. They work pretty well and are in many ways simpler than terminating ethernet. Super convenient for wall pulls and so on.

Even if it's perfectly centered on a nice blackbody point, if the spectral distribution is spiky, it will still look bad in real-world usage

It would be interesting to have polytyped coroutines, where each poll(r:R1) returns a sum type over some subset of poll()able objects that each take their own R2, R3, etc

At some point maybe the ergonomics get so bad that you are just better off writing the state machine manually.

The ergonomics of this could be made reasonably friendly if your completeness checker can narrow down the set of admissible matches a la "GADTs meet their match". A single R sum type, but the compiler only makes you handle the cases that can occur at each poll() site

Isn't this the idea behind generalizing `async` to `Generator`? You can set the `R` type parameter to whatever type you want to resume the computation with.

Stablecoins only use the accoutrement of blockchain as smoke and mirrors to distract and confuse regulators, not because they actually benefit from it at a technical level. They aren't really decentralized at all and could just as easily be a database, except for the regulatory thing

Opus 5.5 doesn't write like this anymore

> because we're posting about humans fucking hating this

A small subset of humans who lost a bunch of comparative advantage hate it. Most people who care about math in the abstract probably think it's really cool!

> He described it at 'strip mining'

Strip mining what, platonic reality? Mathematicians don't have a monopoly on mathematical truth. It doesn't exist for their benefit.

A large fraction of people are open source contributors, hobby game devs, and various other types of people who fall under "devote years of your life with no formal remuneration"

Right, we are on the same page; it's hard to learn as long as you don't train on CoT, which is why training on CoT is "forbidden"

Very excited to hear that you've solved the fields of neuroscience and phenomenology!

> It could be encoding anything in its adjective choices etc.

This is very hard for models to learn

https://www.lesswrong.com/posts/MAww2kXP4cGWz4M5p/steganogra...

> monitor it to prevent users from using their models in certain ways

No, the core upside is that it lets you check what the model is thinking for safety/alignment reasons. Nothing to do with monitoring user activity. You can monitor user activity just as well without looking at CoT.

One of the few sci-fi storylines that doesn't now feel hopelessly dated or anachronistic!

An IMO under-appreciated aspect of attention is that it's an ingredient for building a Turing-complete evaluator, since (for example) it lets you implement mutable RAM by treating the attended data as a write log: https://yager.io/neural/turing/turing.html

It doesn't actually work for that

Is it a human right to live in Sweden?

Because they have a very strong track record of good performance, reliability, and defense of privacy. VPN businesses largely sell trust and Mullvad has earned a lot of it

> While Tor defends free speech, not all speech is equally compatible with our mission.

If Tor hems and haws about political speech, they 100% lose any rhetorical cover against (fallacious but politically compelling) arguments like "so you're OK with CP, but you draw the line at Swedes being against African immigration?" (which is the political issue in question re. Mullvad)

> a search engine which does not have human emotions

I mean TBD, not sure how capable the model they're using for search summary is

Food quality used to be better in Europe but that's flipped, except France which still has some things better (eg French dairy is still better than US dairy)

Average (and probably 10th percentile) retiree is much richer in US. Yes we have higher variance but the mean is much higher too

Europe also has racial ghettos, it's just different groups

> The people in Europe seem to have a higher quality of life

Absolutely not, most places in Europe outside of a few rich countries like Switzerland and Liechtenstein and maybe a few flagship cities seem very low-QoL compared to similarly sized population centers in the US.

I was there a couple months ago and all the cities I visited felt very poor. The buildings were depressing and drab, the trains were frequently delayed, the food was probably the worst I had anywhere in Europe.

The Pentagon is trying to bully contractors into also not using Anthropic's models

I have a probably ~$20k modular rack. It's a lot of fun to experiment with, but the fact that you don't have patch save/recall really slashes the utility.

Doing everything in a DAW is the most practical from a "business" standpoint, but it's way less amenable to creative exploration. You really have to know what you want already.

The practical compromise that works best for me is a well-designed hardware realtime synth. You have access to a small fraction of the design space of modular, but a higher fraction of the accessible design space sounds good, and you of course have patch save/recall.

Recently what I've been doing is creating modular patches, driving them with MIDI, and recording the output into my Elektron Tonverk using their on-device multisample recording. You can record samples at a bunch of different pitch/velocity pairs and save them into a multisample file.

Everyone is "doing that instead". Every successful company in every industry is heavily using AI at this point.

If that were illegal, the next best option would be AI labs acquiring the companies instead of just selling them tokens.

> how would they make money then?

By acquiring companies and letting the now-internal employees use AI, easily outcompeting any company they didn't acquire

The concept is stupid and the rule as stated is stupid.

"Any AI model a company offers to the public has to be released as open weights."

OK great, that means no models released to the public, complete concentration of power.