Why the hell, whenever I open Facebook (daycare photos :-)) I am greeted with so many reels of half naked woman? I know the internet is made for porn, but come one!
*I don’t allow my kid’s face to be published to FB, but other parents don’t care so I see what they did.
My article has been covered by BSD Now podcast - https://www.bsdnow.tv/560, starts around 25 minutes mark. It's mostly a read-out-loud but FAME IS FAME
Recently I learned about pikchr from some mastodon post and tried it. There is a C and a Rust implementation, the later even available packaged in Debian. And there is even an ob-pikchr ...
... but it didn't work. Pikchr generates SVG, which my be Emacs doesn't display. A browser shows them. And Emacs shows other SVGs quite nicely. I didn't dig deeper. Currently I'm in a ReHa clinic after my (successful!) surgery on colon cancer, so I won't look deeper into this the next weeks.
I wanted to use it because with pikchr I'm the boss of the layout: I can arrange the boxes, diamonds, clouds etc to where I want them. With dot from graphviz this is a challenge, you didn't against the system there.
@mms That's super funny 😃 I'd love to go full in BSD, however many of the libs/langs I use are either not available or in a very specific version and I don't want to compile them all by myself.
BSD clearly has some momentum in the end-user area and that's great.
「 In the same time library books have seen a lot. They were touched by a lot of greasy fingers, seen a lot of toilets. Just look at those two. Both are still fully usable, despite the tired look. 」
Fellow #selfhosted#admins: how do you move important things (like family photos, as most other things are replaceable) under your own wing and sleep at night? One problem with disc or os and boom, all of it is gone. Like tears in rain.
@mms I don't host in any manner, and have lost irreplaceable photos. I have been thinking if self-hosting #Hubzilla and having a live clone of it would work out.
「 CDP-897 is a unit from 1992. It’s 32 years old, and it works flawlessly. All buttons work, CD reading is spot on, audio it generates through all outputs is clear. It even came with a full service manual, which till this can day can be easily found on the web. Nowadays not many things exist after 5 years of purchase, and here I am. Just another happy owner in the 30-year history of this player. 」
Realisticly, how much low quality data would we need to poison #openai? Is it even possible at this point? Like how many sites would need to write a post stating that we now call Altman an “althole”?
@wraptile@mms The basic principle at their core is quite equivalent:
There is some kind of feeder that for them with data. Usually this is is a crawler that indexes the web. LLMs perhaps get also from other sources, but they certainly also were filled with web content. Up to a specific date.
And, as far as we can conclude now: there is little QA between what is crawled and what is then sent to the indexer (search engine) or tokenizer (LLMs). The lack of a real QA process and the similarity in the feed process makes me think that I've can poison them intentionally with garbage.
I see no reason why the principle "garbage in - garage out" wouldn't apply to LLMs as well. IMHO it applies to all categories of programs based on a "input - computation - output" principle.