Well I'm not quite sure either but let me tell you what I experienced on an older Intel iMac Pro - (2017).

I loaded up the largest model possible just to see what it would do... I entered some initial question, I forget what, and then got about 10-20 minutes of a "thinking" message.

Then, I got... an "H".

A few minutes later... an "I".

Yeah it too about 30 minutes to begin a message with "Hi", I gave up after a few hours.

So 20 tokens a second is sounding pretty good compared with that!

Assuming they are going to declare bankruptcy and re-organize, I assume the new name they continue business under will be "22 And Me" after they decrement.

It was always a counter!

The real question is if it's 0 or 1 based.

If an LLM doesn't understand something, the solution is to give it more context until it does...

So while "what month is it?" doesn't work with Siri now (I tested as well), what does work is "What month is it in Utah". It gives you the current date and time which you didn't ask for, but it does include the month and it does work.

But don't think I am excusing Apple here, I do think it's absurd they replaced a fairly functional Siri with something that can run into walls with simple questions... and as noted because context is important, probably simpler questions hi walls like this more often!

There are indeed weed "pulling" robots

Used the scare quotes because it's not technically pulling but if the weed is dead and gone who cares.

That kind of stuff trickles down over time so it will reach smaller operations. Until recently though it really is cheaper to hire illegal aliens at far under minimum wage to to some jobs, which also stunts the expansion of robots in farming... but over the next few years I think you'll see a rapid uptake of existing technology.

Robotics first use case should be companies, factories, farming

Farming is already highly automated, giant machines doing all kinds of work.

When a teacher assigns an essay to a student, it's not because the teacher needs that essay, it's because the effort of drafting that essay will cause that pint of bean soup the student is carrying in his skull to become more capable.

That's too bad, not gonna happen,

So the next best thing is what I said, have them use AI to generate output... but then they have to carefully screen what was output, maybe even edit it further to distill a point or correct something.

You are still building, just somewhat different muscles.

Your world is dead, get used to it and adapt or die.

How then do you explain Twitter ran better than before the Musk purchase (observed fact, still running with more features, no Fail Whales in ages) after 80% of the workforce (or more) were let go - (also widely known fact).

Given those absolute facts, why did it work for Twitter if not for the reasons I mentioned?

why look to examples of UK incompetence rather than examples of German success?

I thought Germany abandoned the Linux stuff as well, are they still using it? I figured it was all the same between the UK and other countries in the EU that had gone the same path, but if there was a good outcome I'm all ears...

I was pretty sure I had read about this approach before and sure enough I found a paper from 1996...

It seems like this new approach is maybe a lot further along and more flexible though.

Didn't the U.K. many years ago, try to ditch Microsoft Office for Open Office?

But then, some time ago they rolled that back if I remember right.

It's really, really hard to consider software not developed in the U.S. because software development keeps accelerating. Even if you were going to ditch Microsoft Office, the next viable alternative for an organization of any reasonable size is - Google Office!

They are claiming they can have 10 developers do the work of 100....

But what if that's not because of AI, but simply because those 10 coders are actually working at full capability?

After all, Twitter reduced headcount by over 80%, and not only kept functioning but started adding more features. They were not using AI tools to achieve this, they simply had tons of coders not doing much!

Maybe "vibe coding" is nothing more than finding a small number of developers that are efficient and actually work most of the time they are at work.

If everyone is going to be doing this anyway and you can't detect or stop it, why are you fighting? It's like trying to punch the ocean because the tide is coming in towards a sand castle you have built.

So what can you do instead? Help them learn to prompt AI more carefully, but also how to evaluate the results, and grade them down harshly for failures. It's in the evaluation they will have to learn enough to evaluate what is a good or bad (or even real) answer, if you are worried about them learning nothing.

So basically many courses become about common sense which on the whole is what society needs anyway! It's not like in real life at a job you don't tell someone to go do something, and have to evaluate what was done. We do that all the time in any workplace!

I hope they have the resources to try again, hopefully with an approach that offer much less chance of tipping.

One thought is that possibly the rumored cameras to be added to AirPods, could help with translation by lip reading along with audio...

Or maybe those are not spots for cameras at all, but larger external speakers that would let others hear you speak better!

If you invite one of these into city, say goodbye to what's left of your night sky.

I had a hotel room in Vegas overlooking the Sphere.

After around 10pm or so, the Sphere just turns into showing the moon slowly rotating - so no flashing colors, all gray tones, and also not as bright.

Plus if you think about it the mini-spheres would have a far smaller outside anyway, meaning effects on night sky are probably no worse than a Wal-Mart parking lot, and they would be in the middle of cities with a ton of light pollution anyway (very true of Vegas) so they would not be spoiling anything near pristine.