The past tense phrasing of the title is perfect. And don't miss this gem of a sentence:

From its earliest days, surveillance has helped minimize capitalist dysfunction by regimenting labor, stimulating consumer demand, satiating Wall Street’s hunger for reliable returns, and indulging the security state’s demand for total information awareness.

Quantum physics has a similar parallel. Living things work with a macro interface and not the reality of quantum fields which cause those macro events to be observable.

It really is turtles all the way down.

Exactly. Interfaces all the way down — and we're always at one of them, never at the bottom.

I've never heard of any model maker announcing anything until it is actually available. Anybody who had any different info would be violating their NDA to share it.

Fair point! I was hoping maybe there'd be some rumors or speculation floating around, but you're right. Thanks anyway!

Because your code in isolation has almost zero value. Only working, integrated, shippable code is of monetary value, and that's not something you'll get from an AI prompt, nor from Stack Overflow.

Programming isn't the mystery we devs like to pretend it is, and that wall is coming down quickly with AI coding agents. Everyone has access to the same languages and tools as you do, so your work and mine aren't special because of the words we typed into the files. It's special because it meets the needs of a user, as a finished product they're willing to pay you for.

All the programmers getting paid vast sums of money out there are providing no value? I get what you are trying to say, but it is a misguided notion. As a basic example, if I go and fix the code on someones WordPress site and they pay me for it, that's because I provided a valuable service for them.

And that service is the working product, not any extract of code from it. Thus humans will always be valuable and needed in the loop. We have nothing to fear from sharing our code and everything to gain as AI agents get better at that grunt work.

Its quite a stretch to say this guy had anything to do with it. We've tried to acquire Greenland twice before.

I tried the demo and asked about top selling science books in the US of the last 10 years and was told it can't find sales data so it presented a made-up list. As text. Not a chart.

Hard pass.

There is no such thing as continuous context. There is only context that you start and stop, which is the same as typing those words in the prompt. To make anything carry over to a second thread, it must be included in the second thread's context.

Rules are just context, too, and all elaborate AI control systems boil down to these contexts and tool calls.

In other words, you can rig it up anyway you like. Only the context in the actual thread (or "continuation," as it used to be called) is sent to the model, which has no memory or context outside that prompt.

Furthermore, all of the major LLM APIs reward you for re-sending the same context with only appended data in the form of lower token costs (caching).

There may be a day when we retroactively edit context, but the system in it's current state is not very supportive of that.

> Furthermore, all of the major LLM APIs reward you for re-sending the same context with only appended data in the form of lower token costs (caching).

There's a little more flexibility than that. You can strip of some trailing context before appending some new context. This allows you to keep the 'long-term context' minimal, while still making good use of the cache.

A non announcement from a non company about their non product. I'll go on record saying nothing from these people will ever ship.

MD won because it's HTML in a short form and HTML already won because of the web. The adoption of MD as the primary format that LLMs use to read and write doc just sealed the deal. It's a classic case of "worse is better."[0]

[0] https://en.wikipedia.org/wiki/Worse_is_better