Olives @olives@qoto.org

If people spoke about cars like with "AI".

In ten years, your car is going to turn into Optimus Prime.

Imagine if people spoke about bikes like with "AI".

Normal: The pedals aren't working.

AI: The pedals are disobeying his instructions.

Even if it was used in some sort of security research, the terms "scan" or "circumvent" would be preferable to "deceive" with "AI".

It's not "the AI". It's not a person. It's a tool. It's an unreliable tool.

Bad Language: The AI is being deceitful.

Passable Language: This program can be used as a tool to assist in producing misleading messages. The chatbot might generate misleading messages.

Bluesky doesn't appear to thread well when the subsequent post is made a while after the first such that there are other posts in between.

I haven't had the time to look into this that deeply but these are my immediate thoughts on the matter.

I haven't had the time to look into this that deeply but these are my immediate thoughts on the matter.

I haven't had the time to look into this that deeply but these are my immediate thoughts on the matter.

https://techcrunch.com/2024/11/14/eu-ai-act-draft-guidance-for-general-purpose-ais-shows-first-steps-for-big-ai-to-comply/
1) What if there are computers ten, twenty, thirty years in the future with far more processing power than those now?

2) I haven't looked into this deeply but "risks" sounds awfully vague. It seems likely to drive censorship.

3) Some of these "risks" are hypothetical and appear to be inspired by Hollywood disaster films.

4) "having a tendency to deceive"

If you have an issue with hallucinations (or inappropriate uses like customer support), can you just say that straight-up instead of beating around the bush like this?

Also, once again, this leans into a fantastical framing of the "AI" being an intelligent agent, that plots against people, rather than being an over-hyped tool.

"woke" originally meant being socially conscious or something but that was back in 2015 and I think that meaning was forgotten.

I see a concern about how "mitigating AI risks" could lead to censorship. I haven't had enough time to look into these.

https://mullvad.net/en/blog/2024/11/13/remaining-paypal-subscriptions-are-being-canceled Mullvad transitioning from using recurring payments for #privacy reasons.

More things for me to read.

For something like analytics, maybe there could be a DNTA header which clients / browsers could use.

I don't understand why sites don't just use the DNT header (which has been around for ages) or something to say "okay, this user doesn't want these extra cookies".

I'm sure that was Amnesty's mistake with the cookie prompt, but basically if you hit the option to customize what cookies are set, it won't let you read the page, it just won't work.

Honestly, a prompt on every site is probably not a good way of conveying consent.

If only E.U. sites were designed so that they could be archived.

I commented on the issue before but then I figured I didn't know enough about the issue and deleted them. Turns out I was right to do so.