Ben L

But they're growing so in only a few years they could double those quarterly losses!

With more investments those losses could be exponential,

I can't see how counts of parameters will relate to any AGI.

Pretrained models cannot in my view ever be AGI because general intelligence learns as it goes.

Finished rustlings.

Rust is such an opinionated language.

It's like a type checker and the most opinionated developer you know implemented as a compiler.

We require human sacrifice before anything can change.

Ok but can I eat gingerbread already?

It's been a few years, where's all the new innovation other than slop generation and having a 20 line summary generated from a 2 line email?

We got spicy autocomplete and a copyright infringement machine!

Having a conversation with a chatbot isn't right for anyone. They are software, it's like having a conversation with ms word, it's not healthy.

The fact they can fool and gaslight people with no sentience or meaning giving weight to the words makes them intrinsically hostile.

I've had a physical reaction to hearing the teams beep on holidays when the hotel was using it.

I can't think using ERP products is healthy.

I have noticed some people seem to have their ideas of pain/illness/injury/death from movies only where it's not real?

After going through some very serious injuries it is very real to me, but even before that I knew. The hunger games wouldn't be cool and you wouldn't win.

Some people watched squid game and really thought "wouldn't it be cool if this was real"

It also needs to be able to draw an inference from the text with and for a lot of scenarios that won't happen.

I would guess that permissions problems requires interpretation of internal state that is not in the text .

If it's not in the training data it can't tell you anything.

I asked a few different models about a Perl module I wrote which exists on the CPAN and GitHub but isn't popular, and they all just told me lies.

As they can't produce facts they will just confabulate nonsense.

If that's the future it sucks and I hate it.

Of all the things that can justify the incredible cost of building and operating LLMs, making it a little easier to make documents in LaTeX is not in my top 10.

There's non-LLM interfaces to LaTeX that work great and are decades old and didn't cost trillions of dollars to make.

Pain is not your fault. Lots of things can happen to us for no good reason.

I recently installed Linux and used the fairly ordinary Gnome photo organiser and realised that being online using google photos basically had split my photo library.

And using an offline photo library is so much faster and better.

The monopolist tech companies have ruined personal computing.

Ha I just thought of and posted the same line! Couldn't be less punk.

Hey let's follow rules and boss people around. Not punk! Not even a little bit.

You ain't hardcore if you spike your hair if a jock still lives inside your head.

Fascists reckon they're punk for 50 years now and they never have been.

Shoplifting can be stopped by paying for more staff. They accept the extra "shrinkage" in exchange for lower payroll cost.

Again this is an example of people who don't understand how prices are set. Not a single person has gone 'there's less inventory shrinkage this year, reduce prices"

It takes ages to get the upgrade the just does it once it's done.

Absolutely, and I even get it. I'm in the same boat. And no-one wants to say "we did this because I thought it was a good idea and no other reason".

But also that's why management isn't science. 🙂

Yikes

I can't think of any way you slice those numbers and get to an incremental cost of $125 or so but I don't think that's by a smart person who does math.

So if a barista is being paid $30 an hour and I'm an idiot who thinks prices are set as cost plus, are we saying it takes them four hours to serve one coffee?

Or are they saying busses are paid for out of consumer pricing?

I know that it doesn't make sense but what story are we even telling?

In Australia it has to be allowed specifically by sign (left turn on red permitted) which means anywhere that it causes incidents they can just take the sign down.

Not as good as a blanket ban but still, taking down a sign is easier than changing the law,

I tried and failed this year to explain that "the data cannot show any conclusion for the target metric" is different from "no change occurred from this experiment".

We love to think it's about the data but it's often more about telling a story, because that's how people understand things.

I think you have much smarter ways to approach this than I do!

Someone reading this in management will have a very thin interpretation and may make decisions that the evidence doesn't support. That always makes me nervous.

So you could easily get a zero score from someone who's doing good work but feeling bad about it. And things like how well goals and outcomes are communicated will bear on a supposed performance measure.

It's interesting but it means something other than what it sounds like I think.

Love it. I think if someone could sneak that into the right presentation deck it'd get a laugh. 🙂

That one made me wince!

I know how I would do it but I would surely have some arguments over it.

If someone tells you it's valuable and you do a great job but 6 months later nothing worked, is it valuable? It's valuable to you right, you did the work and got paid!