So glad I got out of IT when I did (and that was after 30 years) and there's not a snowball's chance in hell I would ever get back into it professionally now.

Today's IT sucks so much.

Of course Co-Pilot adoption is lagging. It's a moral nightmare. If you want to be entirely pragmatic about it: It sucks.

More code isn't _better_. More features isn't _better_. Nobody wants AI.

It's been a bad bet from the get go and private equity and the VCs just _refuse_ to take the hit.

I'm getting this too, but it seems like desperation. I'm supposed to "learn to use AI" but they're not sure for what, have some "tools and resources," just boost productivity people!

I'm also wishing I could bail from tech, though the silver lining is I'm doing security and compliance stuff, so more slop probably doesn't hurt me. Earlier in my career I did a lot of contract work on shitty PHP sites that somehow made money, so panic and hurl money to maintain them!

There are stil AI-less versions of Windows available for enterprise customers.

Thee are downloadable and work. 😁

Here’s something for people who are still actively working in tech to consider: could it be possible to get municipalities to provide residential access to the Internet? I mean by this like a city run ISP in web hosting service?

In many states, I think large ISPs have made this difficult if not possible, but it would be interesting to see people in tech apply themselves to making this possible somehow

note it’s the same exact energy as the RTO thing. Same exact reason too.

(Employees are a cost)

I really like all kinds of AI. You know, the actually useful kind that has been part of many systems for years now, in some cases even decades. And the fun kind used in games. The video filters that swap faces or turn people into cartoons. I also like generative AI, but I prefer the early versions that are all weird and trippy and out if this world, the ones that don't make much sense but produce funny results. I like to put dada poems, silly jokes, song lyrics, or passages from science fiction and fantasy novels into image generators to see what happens.
I had a lot of fun with Stable Diffusion 1.5 based models, much less fun with SD2, then with SDXL, the images suddenly became a lot "better", but it became increasingly difficult to get really interesting, unexpected results. The same goes for audio generators and chatbots. Does anybody remember the endless stream of automatically generated death metal music on the Dadabots YT channel? That was bloody awesome. It sounded weird and strange, yet it still somehow resembled something not entirely unlike real death metal.

But putting highly polished yet almost entirety useless AI agents into everything is just stupid. It's like back in the day when they discovered radioactivity and didn't know if it was good for anything or not, but they put radium into everything because it was exciting and new.

The main problem with using LLMs as user interfaces for natural language processing is that those systems cannot distinguish between instructions and data, anything inside a prompt is both, that's why you can always use prompt injections, and unlike with other systems where instructions and data can be found in the same context (like SQL, for example), it is impossible to catch those injections before they happen. With a formal language, an artificial language adhering to strict mathematical logic like a computer language, whether it's a database query language like SQL, a markup language like HTML, or a programming language like C or Python or Smalltalk, you can always find a way to divide instructions from data, even if such a division is unfortunate optional like in SQL, leading to all kinds of code injection vulnerabilities when some programmer forgets to add some quotes. Natural languages, however, don't give us such an option since they aren't artificially constructed to do so. Any system that processes natural language and takes commands in natural language cannot be secured against doing anything it is capable of doing. It is simply not possible. Whatever global master prompt you give your LLM in order to keep it in line, a user can always find a way of convincing the chatbot that it must ignore that prompt. Whatever data the LLM eats it can be made to spew out again, once the global restrictions have been lifted by talking the chatbot into acting like they don't exist (yes, it's basically that easy). Once an LLM has learned something it wasn't supposed to learn, like personal data or national secrets, nobody has found a way to make it forget again. Oh, we have problems with artificial neural networks accidentally forgetting old stuff if they learn too much new stuff, but we can't make them forget anything on purpose.

There are a lot of ways of making AI more secure, more useful, and more efficient (so that they can run on the local machine even if it's just a cheap Android smartphone), but most of them require much less automated data mining and much more human curated, carefully selected training data, and reinforcement learning through interactions with humans. Which of course means that using AI to make a lot of money isn't going to work because building the AI in the first place requires an awful lot of human labour if you want to do it right. And since the humans working on it need to actually understand the knowledge they're trying to transfer into the model, that labour won't come cheap.

A pivotal moment for me was using ChatGPT to write my performance review a couple years ago. Other than a few minor hallucinations about the company mission (let's be honest, they're all the same), I never had an easier time legitimately selling my work.

Point being: If China can tell its bureaucrats to make more decisions with #AI, like it or not, it's here to stay.

Coercive Capitalism was born from Surveillance Capitalism.

When you can't opt out of products, you're a captive consumer on the path to perdition of enslaved consumer.
theguardian.com/technology/201…

hbs.edu/faculty/Pages/item.asp…

en.m.wikipedia.org/wiki/The_Ag…

en.m.wikipedia.org/wiki/Survei…

news.harvard.edu/gazette/story…

theguardian.com/books/2019/oct…

We no longer have a choice to switch to renewable energy sources like wind & solar.
open.substack.com/pub/bleeding…

researchgate.net/publication/3…

1/

'The goal is to automate us': welcome to the age of surveillance capitalism

Shoshana Zuboff’s new book is a chilling exposé of the business model that underpins the digital world. Observer tech columnist John Naughton explains the importance of Zuboff’s work and asks the author 10 key questions

^{John Naughton (The Guardian)}

The fools.

Just like my incompetent managers at a former job.

They think programming is about typing in the code, and having a synthetic text extruder that types faster than me is therefore going to make me more productive.

Never crossed two brain cells themselves, I thought even then.

Now I think I was right.

Must. Burn. Up. Planet.

#BurnItUp

INSANE.

This is a sign of weakness, an acknowledgment that these tools are not universally useful in any way and therefore must mandated by corporate fiat. Employees should stage a public revolt.

tools so "good" no one wants to use them.

That's a thing that happens in tech.

I've aged out but it's still foisted on me. Just disabled Gemini on my phone. Don't want any US cloud-hosted, data collecting & monetising stuff whatever. I use Linux, LibreOffice etc, want out of Android and to use RISC-V or similar where possible. Willing to pay for privacy, open source hardware, self-hosted / community-hosted things.

Resistance is better than quitting if you can make it work.

@Phil

Oh, absolutely! 😂