Can you imagine AI designing a beowulf cluster to ... ?

$ nohup paperclip_d &

The only solution is to prevent companies from collecting and maintaining this level of information.

Yes, this has been said by many people, in many countries over many decades. In some cases this sensative data is being kept years after a customer ceases business.

In this case I shake my head at the laziness of the security team, as credential management in Sec/Sys/Dev/Ops/CM is pretty standard, but does need the company culture that it matters, or one ends up with main admin password pairs being "admin/nimda" because so many scripts would break.

Putting in regulations, as you say, is the only way to add some swiss-cheese, but the beancounters and PHBs won't like the cost, so nothing will be done. And Marketing, they drool at the profits, so that push to "grab all the data, monetise the living jimmies out of them" in the board room is real too.

And not just the emdash, the bulleted list.

"* The prophecy becomes self-fulfilling through material concentration — as resources flow towards AGI development, alternative approaches to AI starve..."

For the love of cobol, if you are going to write a diatribe about the evils of AGI, don't use chatGPT to do your homework.

giggle. genuine thanks for the smile. out for six, home run, goal, whatever your sport might be. Irregardless is the best word. Can I add the "rediculous" though to your "could care less"?

I know the paperclip apocolypse was only a thought excercise, but with this loss of consumer RAM, and the shortage of other components like video chips, the growth in power and water consumption (and the associated emissions), and the way the tech is being force-rammed into everything, one does start to wonder if Nick Bostrom was actually a little too prophetic.

The website was only part of the cost, a small part and apparently less than a few million. Most of the cost came from upgrading the backend, and likely all the well paid consultants oinking at the trough over those. What we dio know is that all the servers were replaced, the networking, and a bunch of backend stuff because it really was time to get rid of the old servers from 15ish years ago. They had suffered a breach, and the server infrastructure was so much under stress that they couldn't even add in SSL certificates, access was unencrypted to reduce load.

However, that said, the new site is horrible and impossible to find the real needed information like mentioned synoptic charts, or even observations for the past three days. Important information like HF MUF's (Space weather) isn't even linked any more. The radar is inaccurate and a GenZ technicolour yawn.

Plus of course many people who use it are older, and change is harder.

I wish they would treat the Liquid Ass vista theme as bloat.

They'd be extra stupid if they didn't know.

Indeed. But I think they don't realise it's plonked onto Google/Bing/Ducky/Ecosia/Kagi searches for the whole world, and likely consider it's something nobody will ever see unless the (assumed incorrectly) private link is disclosed to them.

I absolute agree with your points there, about both it's toxic positivity, and the continuation prompting. The positivity comes from it's RLHF training and the continuation prompting is absolutely designed and implimented to addctively maintain interaction in the same way like-buttons on social media give the poster a little dopamine burst.

Secondly, ChatGPT always wants to get you into a conversation, it always wants you to continue interacting. After answering your question there's *always* a followup "would you like me to..."

When I did use the hallucination machine - before the uncanny valley effect really became unbearable - I tried with adding "Please do not suffix any engagement queries" and it sticks for a while, but like asking it to not use emdashes, it is an instruction that needs regular reinforcing as context grows. So you tend to need to paste your list of "Please do not (*) suffix any engagement queries (*) apologise nor say you will do better (*) use emdashes (*) use the word 'testament' (*) use purple prose (*) blow smoke up my arse... (etc)" stuff every few prompts so it's reinforced.

And finally, ChatGPT remembers everything, and I've recently come to discover that it remembers things even if you delete your projects and conversations *and* tell ChatGPT to forget everything.

It has a RAG-type memory that extracts 'facts' from your chats and it will search through those every prompt. Before I gave up on the fawn-machine I did find a section where you can edit those. Another one for reinforced preprompting "Please do not use memory unless I specifically ask" but of course that also keeps dropping off, so you do need to paste the entire preferences list every few prompts, or SmithersGPT trims it from it's context and falls back to corporate rules.

How many of those 47,000 chats had users which were actually aware that other people can read their shit?

Probably none. Although it is mentioned that chats can become public in some cases, it's not really made clear that they are also easily searchable by absolutely anyone. Add to this that most people have memories shorter than a goldfishes, so forget what personal info they put early into the thread when they do finally share their little personal echo chamber with the fawning SmithersGPT, and that all the preprompting and "memories" pulled from RAG-like storage will be infecting their shared thread as well.

But this may also be a sign of the current times. People are too used to sharing everything, from what they ate for breakfast to the size and texture of their bowel movements. So, it may well be that; they simply don't care nor understand the danger in it.

What features were they trying to even add?

Agentic clippy needs to be able to be able to use all features of your PC not just access all data.

Clippy and AI eh? Well that could lead to the "paperclip apocolypse" https://ancillary-proxy.atarimworker.io?url=https%3A%2F%2Fcepr.org%2Fvoxeu%2Fcolumns...

What does it matter what his side of the story is? Even if the IT contractor was unjustly terminated it's not carte blanche to access systems you no longer have rights to and perform a childish password-reset payback.

Generally the specialists needed to go over the system will cost many times more than $40 an hour. Performing a full company wide systems audit to make sure he didn't drop any viruses or exfiltrate data can cost many hundreds of thousands of dollars alone, especially if outside security specialists are used for the task as would often be the case.

On top of this, a lot of automated tasks can rely on stored passwords for backups, scripts, crons and the like. If he's reset passwords, then doing this could have a lot of broken system houskeeping out of the picture too, and depending on the amount of shell-script spaghetti that has grown over the years, this can be tough to find.

It's sort of sad that parents are not aware that they are not just installing something that by it's nature must record and send everything back to some faceless intermediate before it heads right into BigAIs training data, but more than this, a blackbox more than capable of subtle manipulation.

As long as it comes wrapped in a cuddly Teddy Bear shell, these parents seemingly lose all sense of safety.