We have a firm grasp on actual inference costs from the various open weights model providers on OpenRouter. They don't have the money to subsidize inference and it's quite a competitive market, so the prices are representative of the costs.

The problem of citizens having anonymous internet connectivity.

That's an illusion. Two days of location data and you can pin down the owner pretty well.

I thought about getting a SIM when Germany was about to introduce ID requirements. I quickly realized this being a moot point.

There a significant difference between "the user can be identified fairly well if you can get access to sensitive stuff" and "the owner is always explicitly recorded in a searchable database".

My burner would only be powered on for the duration of the call and the time to spin up gsm connectivity.

The free anonymous internet was only ever a ruse to get people to use it so the CIA could spy on them. DARPA, folks, created a “free as in beer” global surveillance network and we all bought it.

Not that we didn’t get anything in return but the idea that the worlds foremost military industrial complex just gave this to the world because they loved us is laughable.

Huh? DARPA created something to help retain communication and coordination capabilities for our government in event of nuclear attack.

Indeed they did, at great expense and gave it away for free!

He said "as soon as", so it's a pretty testable prediction.

"implode" is not testable. A good prediction is specific, time bound, measurable, etc. Otherwise you can flatter yourself however you like with confirmation bias. This is basic stuff.

>It's the other way around (but the result would be the same): Alphabet has no need to make a 100x exit for the investors, and so can offer the service at cost + %markup, while Anthropic and OpenAI are VC funded, meaning that they need to show 10x - 100x exit for the investors.

If this was true, Alphabet wouldn't currently be charging more for a worse product than OpenAI and Anthropic.

>I fear that OpenAi and Anthropic would not be able to compete against an adveserial Alphabet which owns it's own models, hardware, large corpus of data, talent and network effects.

You might as well say the same about GCP vs AWS. At the end of the day, in spite of how much superior engineering prowess it has, Google still treats its customers like it views them as a steaming, fly-covered pile of crap. This reflects just as much in Gemini as in their other products; after their initial competitive Gemini 2.5 Pro release, they just kept dumbing it down and reducing quality of service while charging the the same amount, trying to pull a bait-and-switch, and with their latest Gemini Flash release they're charging customers even more for a worse product. No amount of engineering or hardware can overcome such a customer-hostile corporate culture.

I don't think the medical establishment is covering it up; if they could sell Ozempic as preventing cancer they'd jump at the idea.

The medical establishment isn't covering it up; they're just embarrassed to say it out loud, directly. Instead they talk around it. GLP-1 drugs like Ozempic may well get to a point where they can be marketed as cancer preventing.

>instead of enjoying the craft digging deeper into problems in the span of 2 days, now you are rushing into some slot machine with the hope of it giving you the right answer with the right prompt.

If you're treating it like a slot machine you're doing it wrong. It will give you exactly what you ask for if you ask clearly, i.e. write a clear, detailed specification, not just "do X!". The nondeterminism comes from vagueness in specification.

>we're only a decade away from the US or Israel etc. airdropping 1000 armed humanoid robots into a civilian space to hunt for "terrorists"

Currently they just bomb the buildings into the ground, killing everyone indiscriminately, so robots can't be any worse than that.

A robot with enough arm strength to properly clean a stubborn toilet wouldn't be safe around kids.

I let my current cleaner be around my kids.

Now the question is is it riskier to have basically a stranger with strong arms in my house near my kids, or a robot with strong arms in my house near my kids?

I feel like a robot has the technical capacity to see behind it and stop (I have many times for example been using the vacuum and moving my arm forwards and backwards and whacked a kid in the face with my elbow on the backswing because they've walked up behind me and I've not known, but a robot with literal eyes and radar in the back of its head would spot that situation and freeze). Similar to self-driving cars: they have lots more eyes than a human has, and can be looking everywhere at once etc.

But do we trust the programming? Do we trust the human cleaning my toilet's "programming" (thoughts, emotions, motives etc)?

Your cleaner is a human and you trust them to behave like a human including having a quite low probability of suddenly having a seizure. Robots do not think the same way. Any software glitch may cause it to move in an unexpected way. I had one freeze for half a second (unknown reason) while motors were full forward, and it rammed itself into a literal brick wall.

It just needs the patience to apply and gently scrub with mild chemicals for several hours.

You know what does this better than a humanoid robot? A toilet with a built-in dispenser for mild chemicals! It could do it periodically, or with every flush so that the mild cleaning chemicals are always sitting there. An unsophisticated one made of cheap plastic could be bought for $5 a the grocery store and clip onto the side.