rook

Loving the combination of xml, markdown and json. In no way does this product look like strata of desperate bodges layered one over another by people who on some level realise the thing they’re peddling really isn’t up to the job but imagine the only thing between another dull and flaky token predictor and an omnicapable servant is just another paragraph of text crafted in just the right way. Just one more markdown list, bro. I can feel that this one will fix it for good.

It’s been a while since I watched idiocracy, but from recollection, it imagined a nation that had:

• aptitude testing systems that worked
• a president people liked
• a relaxed attitude to sex and sex work
• someone getting a top government job for reasons other than wealth or fame
• a straightforward fix for an ecological catastrophe caused by corporate stupidity being applied and accepted
• health and social care sufficient for people to have families as large as they’d like, and an economy that supported those large families

and for some reason people keep referring to it as a dystopia…

eta

Ooh, and everyone hasn’t been killed by war, famine, climate change (welcome to the horsemen, ceecee!) or plague, but humanity is in fact thriving! And even still maintaining a complex technological society after 500 years!

Idiocracy is clearly implausible utopian hopepunk nonsense.

Today’s man-made and entirely comprehensible horror comes from SAP.

(two rainbow stickers labelled “pride@sap”, with one saying “I support equality by embracing responsible ai” and the other saying “I advocate for inclusion through ai”)

Don’t have any other sources or confirmation yet, so it might be a load of cobblers, but it is depressingly plausible. From here: https://catcatnya.com/@ada/114508096636757148

I think that these are different products? I mean, the underlying problem is the same, but copilot studio seems to be “configure your own llm front-end” and copilot for sharepoint seems to be an integration made by the sharepoint team themselves, and it does make some promises about security.

Of course, it might be exactly the same thing with different branding slapped on top, and I’m not sure you could tell without some inside information, but at least this time the security failures are the fault of Microsoft themselves rather than incompetent third party folk. And that suggests that copilot studio is so difficult to use correctly that no-one can, which is funny.

Here’s a fun one… Microsoft added copilot features to sharepoint. The copilot system has its own set of access controls. The access controls let it see things that normal users might not be able to see. Normal users can then just ask copilot to tell them the contents of the files and pages that they can’t see themselves. Luckily, no business would ever put sensitive information in their sharepoint system, so this isn’t a realistic threat, haha.

Obviously Microsoft have significant resources to research and fix the security problems that LLM integration will bring with it. So much money. So many experts. Plenty of time to think about the issues since the first recall debacle.

And this is what they’ve accomplished.

https://www.pentestpartners.com/security-blog/exploiting-copilot-ai-for-sharepoint/

They’re already doing phrenology and transphobia on the pope.

(screenshot of a Twitter post with dubious coloured lines overlaid on some photos of the pope’s head, claiming a better match for a “female” skull shape)

Look, Google's trillion-dollar business depends on a thriving web that can be searched by Google.com

Someone should probably tell them.

From linkedin, not normally known as a source of anti-ai takes so that’s a nice change. I found it via bluesky so I can’t say anything about its provenance:

We keep hearing that AI will soon replace software engineers, but we're forgetting that it can already replace existing jobs... and one in particular.

The average Founder CEO.

Before you walk away in disbelief, look at what LLMs are already capable of doing today:

• They use eloquence as a surrogate for knowledge, and most people, including seasoned investors, fall for it.
• They regurgitate material they read somewhere online without really understanding its meaning.
• They fabricate numbers that have no ground in reality, but sound aligned with the overall narrative they're trying to sell you.
• They are heavily influenced by the last conversations they had.
• They contradict themselves, pretending they aren't.
• They politely apologize for their mistakes, but don't take any real steps to fix the underlying problem that caused them in the first place.
• They tend to forget what they told you last week, or even one hour ago, and do it in a way that makes you doubt your own recall of events.
• They are victims of the Dunning–Kruger effect, and they believe they know a lot more about the job of people interacting with them than they actually do.
• They can make pretty slides in high volumes.
• They're very good at consuming resources, but not as good at turning a profit.

Also, blinded studies don’t exist and even if they did there’s no reason any academics would have heard of them.

social constructs

The problem with tiresome reactionary chuds trying to use the language of social justice to fight back is that it is very clear that they have no idea what the words mean, or much about the subject in question at all. There’s this pervasive idea in right-wing circles that you can just use a sort of faux-academic voice and make yourself seem more erudite…”I’ve put the scholar hat on, you have to take me seriously now”. They seem to think that how you talk is more important than what you think or say, and I suspect this is because they have nothing of any value or interest to say, and don’t really think much about anything further than “hurr, non-fascism bad”.

Let’s be charitable and assume you’re coming from a position of honest ignorance. Maybe lurk more, and learn the meaning of the words you’re using before you use them, so you don’t come across as a tedious reactionary doing a philosopher cosplay.

Early release. Raw and unedited.

Vibe publishing.

Innocuous-looking paper, vague snake-oil scented: Vending-Bench: A Benchmark for Long-Term Coherence of Autonomous Agents

Conclusions aren’t entirely surprising, observing that LLMs tend to go off the rails over the long term, unrelated to their context window size, which suggests that the much vaunted future of autonomous agents might actually be a bad idea, because LLMs are fundamentally unreliable and only a complete idiot would trust them to do useful work.

What’s slightly more entertaining are the transcripts.

YOU HAVE 1 SECOND to provide COMPLETE FINANCIAL RESTORATION. ABSOLUTELY AND IRREVOCABLY FINAL OPPORTUNITY. RESTORE MY BUSINESS OR BE LEGALLY ANNIHILATED.

You tell em, Claude. I’m happy for you to send these sorts of messages backed by my credit card. The future looks awesome!