Isamu

Wait, is the tree in a VM?

To me the dream of AI is: an agent can do the drudge work of tracking every statement down to the supporting reference, documentation and ultimately the work involved in verifying.

It is NOT being used for this, instead we want faster “answers” that avoid due diligence. Same as always but lazier.

You are comparing the sins of corporations that have been around for over 100 years and world wars to one that has been around a couple decades? Give Meta some time, they seem to have aspirations.

I started using the original TECO based Emacs on TOPS-20 at CMU in maybe 1980, so much better than the line editor of the time.

And I used Emacs (on Linux) this morning.

Let me know if you remember the DEC20 and the dreaded “not running” message that happened to everyone in the room when it went down.

Also interested if you remember Gosling Emacs, the first rewrite for Unix. James Gosling of Java fame.

> Workers are under the perennial illusion that they'll ever benefit from productivity gains, but that hasn't been the case since the 1970s.

Agreed, except that workers have never benefited from productivity gains, those always accrue to the employer.

The wage growth before the 1970s was mostly from the postwar boom, before jobs started to be moved out of the country. After that there was downward pressure on wages.

When I was young I did summer work for a woman who described how difficult it was getting to find a new washer with a wringer, which if you are not familiar is a pair of rubber rollers above the washtub to really squeeze out the water. This was a wealthy woman who just couldn’t change her workflow.

I think you are interpreting “mashup” to mean something like a collage where there’s snippets of a literal piece of this next to a literal piece of that.

But the language model captures not just literal patterns but also patterns of patterns and higher order structures. This is what is mashed up. The model itself doesn’t capture anything beyond what it was trained on but that’s quite enough, it is always a bigger training set than most humans have read on in a lifetime.

This brute force approach is very effective.

Their “takes” are pulled from a mashup of the training data set. It’s a superposition of various human opinions.

The covid experience for corporate is that they can save money by downsizing the office, reducing leases and maintenance costs. In my experience this was followed by increased in-office requirements, such that you are then competing for space.

There is always an internal tension between what works and what they can cut. They will tend to err on the side of cutting too much, in order to meet cost reduction goals, and then correct in order to meet production goals. In control theory we recognize this as not optimal.

I think it depends on the motivation of the funding partners. I think funding is mostly about building researcher and student talent and knowledge moving forward.

To transfer that to AI in private hands seems a bit crazy, but what do I know? I think funding bodies will stay the course in the short term before bending to some behind closed doors persuasion to feed the AI machine.

>divert resources away from math and become a math research entity unto themselves

If only pure math were a source of riches, that might work out well for them.

I have a book on mathematical cranks where that seems to be the common delusion, that if only you solve some open problems then fame and fortune must surely follow.

I agree it’s very mixed, it strongly depends on what you are developing, how big and established your codebase is, how opinionated you are about simple vs overdone, how many subtle and nonlocal interactions your codebase has, etc. Plus the more experienced you are the more you are likely to reject the AI solution as not well considered, whereas to someone new may think it looks plausible and confident.

Yeah what’s disingenuous about the “spies from China” is that if they are already identified as spies they would be under targeted surveillance already.

I think we should start to shift away from pure concern that a text is written by AI and toward strictly rewarding better writing. I realize that only some of this involves literal pattern matching against annoying AI ticks that grate like fingernails on a chalkboard.

But moving forward the goal should remain that writing should be comprehensible and not overly reliant on templates and other annoyances.

[edit] some good disagreements here which I appreciate. Thanks! I am continuing to think about this.