Sounds like RAG to me.

What you label "government bureaucracy" is very often environmental protections working as intended.

One AI company's "bullshit" is another citizen's "clean air and water". Since the AI companies don't care about those negative externalities, the government can act help to mitigate the damage.

Yes, this is actually a reason why an AI executive could be a formidable alternative in theory.

The CEO is ultimately an employee of the board, and managing the CEO with respect to the board's oversight and interest is a somewhat adversarial challenge.

However, the AI CEO's intentions are fully transparent to the board and entirely pliable to the board's demands. The AI CEO also offers the board the opportunity to retain significant equity shares that materially impact their wealth - in this way AI executives stand to be an even more tantalizing target than fixed labor costs.

Of course, this presumes LLMs will become reliable enough for this, which doesn't seem likely.

One person isn't a company. Legally maybe, but there's no point in a company if you can profitably do all the labor yourself.

I may be able to do all the labor myself, but need investors because I can’t afford the capital.

I may be able to do all the labor myself, but want to ensure continuity beyond my ability or interest to do the labor myself. (Sometimes this is by my wish; sometimes by the wish of consumers of my service.)

I may be able to do all the labor myself, but not want to expose my entire net worth to the financial obligations of the company.

There are plenty of cases where a single person company makes good sense.

At what point does a company become a company if one person doesn’t make a company? What’s the threshold for non-labor where a company becomes a company?

I'd start with division of labor as a minimum.

There are holding companies with zero employees. They typically pay a management fee to someone.

Attorneys commonly do this.

True, but better nvidia than any other big tech contender. Nvidia has a very strong business incentive to make huggingface thrive, whereas pretty much everyone else has the opposite incentive.

Especially when all major labs are building their custom inference chips.

Yeah I think this is probably one of the best outcomes you could hope for if you want to use open models.

They are not connected - this is a pause on immigrant visas, not work visas.

What do you think EB visas are: immigrant visas or work visas? Answer: they are both.

You're correct. I spoke a little too loosely, there are many different kinds of visas, but something like 95% of all work visas issued each year fall into the category of non-immigrant.

I'll concede that this statistic is cold comfort if you fall into that 5% category, but the point is that work visa almost always means non-immigrant.

Immigrants don't work?

This is the most missed point in all of the flame throwing in this thread.

Any particular issue you have with the content of the article?

There is no "future proof" for >1T param models, there is no present or future where you can run a model that size on consumer hardware.

6 Mac Studios for 100k USD. Consider it a rule of thumb now - 100k to run 1T params, scales linearly.

If you have 100k you can just buy a few b200s.

One does not simply "buy a few b200". They come in 8 packs minimally, at >5x that budget.

Out of curiosity, what do you do to distribute the weights and inference across the multiple machines?

Thunderbolt 5, and software

at a 100k$ you're not buying apple products to be limited by the 120gb/s thunderbolt port.

I don't know that "consumer hardware" is a useful distinction anymore, it's just "what's your budget and what's your speed requirement".

Consumer hardware means:

- 120v input plug

- not rack-mounted

- has a video out port

To me, "consumer item" is when i don't have to contact sale people for price.

Ah, a MacBook Professional is Consumer /s

In computing "Pro" is not the opposite of "Consumer".

Putting aside the fact it is a marketing label, "Pro" usually means "designed for work" while "Consumer" (in this context) means "doesn't need a special environment".

In computing the distinction is primarily noise, power and cooling requirements.

If a computer is designed to use home power and is quiet enough to use without annoying people and doesn't require specialist cooling then it is a consumer device, even if it is used for work.

Apple's use of "Professional" is just a fancy alias for high end or premium, it in no way indicates anything about being "for professionals" (most extreme example: "professional" iPhone models)

Hence why they had to make up the "Studio" brand for the workstation market, because they'd already fully removed any meaning from "Professional"

Yes?

Pro?

Want to bet that consumers with enough money buy it?

Sonnet is estimated around 1T, so that is far beyond what's practical as well.

For local LLMs with a Mac, rule of thumb is you always want an Ultra (due to memory bandwidth). Even an M1 Ultra is superior to an M6 Pro in this regard.

There are no configurations even close to running something comparable to frontier model variants, they're simply far too large, but something like full precision Qwen 35b or DeepSeek 70b at 50+ t/s is well within available configuration, and potential for plenty of room for large context sizes.

256GB is enough for DSv4 Flash, expect maybe ~30tg/s, and a lot better profile.

I'm using Flash heavily, and I would describe it as nearly as intelligent as Sonnet-class in agentic coding, but more usable. Less world knowledge of course, and definitely a bit less intelligent; but not _that_ much.

On usability: Takes less handholding, less likely to make unsolicited refactors or whatever, and the writing style is readable.

It's not great at super-long-horizon goals as the Claude 5 models are; but if you have a good harness, you can get around that.