@gray17@mastodon.social avatar

gray17

@gray17@mastodon.social

another wolf in the crowd (he/him)
systems, design, infosec, toolchains, furry, bdsm
searchable, CC-BY-4.0
Portland, Oregon, USA

This profile is from a federated server and may be incomplete. Browse more on the original instance.

gray17, to random
@gray17@mastodon.social avatar

#tldr - jailbreak LLMs by steering the conversation with an LLM

LLM is "plausible improv" that's seeded with text like "You are an ethical AI". Since the LLM has been trained on fanfiction (and reinforcement learning), its continued improv then sounds like an ethical AI.

Recent text affects the improv more strongly than the initial seed, especially if you refer to what the LLM itself says. This reliably jailbreaks existing LLMs, and this can be automated with an LLM.

https://arvr.social/@mpesce/112285299868317440

zarfeblong, to random
@zarfeblong@mastodon.gamedev.place avatar

Latest round of puzzle games. By pure coincidence, these are all easy puzzle games, and three of them are bite-sized. (Beyond Horizons is longer, but not because it’s more difficult.)

  • Skaramazuzu
  • Boxes: Lost Fragments
  • Botany Manor
  • Between Horizons

https://blog.zarfhome.com/2024/04/spring-puzzle-games

gray17,
@gray17@mastodon.social avatar

@zarfeblong re: police procedural + procedural generation - there's Shadows of Doubt (which I haven't played yet)

yingtai, to random
@yingtai@zirk.us avatar

I remember someone compiled a list of platforms that don't mind explicit content, but I can't seem to find it again. Does anyone have the link? Boosts welcome.

gray17,
@gray17@mastodon.social avatar

@yingtai https://thicc.horse/@QuanZillan/112106917445308915

JordiGH, to random
@JordiGH@mathstodon.xyz avatar

You know that trope where...

There is some character who is presented with a challenge, a difficult situation, and someone or something tells this character, you will fail/die/get trapped/go insane/suffer a horrible fate...

"... like all the others"

?

I swear this trope is common in fiction. I can think of it in Coraline (the kids with the Other Mother), Lizard (the other lizards who came before you), the unfortunate souls Ursula has in The Little Mermaid...

I just can't find it in TVT.

gray17,
@gray17@mastodon.social avatar

@JordiGH hm, I can find several things adjacent, but nothing exactly what you're describing. this is kind of a weird hole in tvtropes. it's very.. character focused, not so much about situations where character isn't the focus

gray17, to random
@gray17@mastodon.social avatar

Today in #twine porn, I finally moved most of the js and css out of .tw files and into .js and .css files, with type annotations that typescript can check. type checking found a few bugs, of course. I wish types/twine-sugarcube didn't use any so much, but oh well, I'm not going to fix that today.

gray17,
@gray17@mastodon.social avatar

today's #twine porn problem. I have a function "replay this path", which renders each page and click()s on the right links. This flashes each page and is kinda slow.

In principle, I can traverse the path without displaying the page or calling click(), but SugarCube has a couple different ways that you can attach code to a link, and there's no simple way to determine what the code is or run it, without calling click().

solution: replace SugarCube with a pack of wolves running across the tundra

pupether, to random
@pupether@woof.group avatar

Having completely failed to get anywhere with the last lot, I have done another refresh of my Library stash so I am trying mk. 2 of this reading thread.

gray17,
@gray17@mastodon.social avatar

@pupether @malleche @lolo Scalzi is great, I've liked all his books that I've read. I recently finished Starter Villain -ordinary guy finds out his estranged uncle died and left him the keys to his supervillain enterprise. Chaos ensues.

Heinlein, I liked a lot of his work when I was younger, but haven't read any recently. Not sure what I'd think today

JorgeStolfi, to random
@JorgeStolfi@mas.to avatar

What is the proper general term for someone who is bent on imposing their religious beliefs on those who do not share them? Like the Taliban, Iran's Revolutionary Guard, the Spanish Conquistadores, ...

gray17,
@gray17@mastodon.social avatar

@JorgeStolfi crusader?

gray17, to random
@gray17@mastodon.social avatar

Night of the Living Furries
Pride and Prejudice and Furries
World War F
28 Days After the Con
Furryland: Double Bap

foone, to random
@foone@digipres.club avatar

I think my next project along the lines of VGAPride might be writing programs to display a trans pride flag for as many platforms as possible.
Like, a simple c64 program, qbasic, win32, js, etc.

gray17,
@gray17@mastodon.social avatar

@elithebearded @richardloxley @brouhaha @BRicker @foone and some versions of Unix added a copyright statement to the otherwise empty file of /bin/true
https://trillian.mit.edu/~jc/humor/ATT_Copyright_true.html

timbray, to random
@timbray@cosocial.ca avatar

Here's the actual study: https://bfi.uchicago.edu/insight/research-summary/a-discrimination-report-card/

It's well presented and defies summarization, but also considers gender bias. Big-tech names do not feature prominently on either the good or bad side of the report. A few companies (especially in things like auto service) got some explaining to do.

https://flipboard.social/@CultureDesk/112253573398394283

gray17,
@gray17@mastodon.social avatar

@timbray big-tech is largely not in this study (eg, Microsoft is not included), because they looked at large companies that have entry-level positions in many locations, which ends up being mostly retail, regional service, healthcare, supply chain, etc. (Most big-tech is too few locations for their methodology. Not sure why Amazon didn't qualify.)

gray17, to random
@gray17@mastodon.social avatar

... I can use these wireless earbuds continuously if I just wear one while charging the other. so... ideally a set of earbuds will be three: two in use and one being charged, swap out at 50%

dkub, to random
@dkub@woof.group avatar

As I'm currently wrangling with airline checkin I'm currently taking bets on Real ID requirements actually kicking in May 7, 2025. ^_^

gray17,
@gray17@mastodon.social avatar

@dkub one time I lost my wallet on a business trip. no id, TSA let me through anyway. I'm not really sure why they let me through, and I don't know what I would have done if they had said no.

zyd, to scifi
@zyd@emacs.ch avatar

Are there any novels/short stories/film/tv/etc. which feature a "hive mind" that doesn't involve "a complete loss (or lack) of individuality, identity, and personhood"? Maybe it'd be better to simply call it a collective mind.

#scifi

gray17,
@gray17@mastodon.social avatar

@zyd Ann Leckie's Translation State has an alien-constructed humanoid that matures by merging their mind with another one. The story is partly from the point of view of one resisting the process of merging

gray17,
@gray17@mastodon.social avatar

@zyd Also, Leckie's Ancillary Justice trilogy is told from the point of view of a humanoid that used to be part of a starship hive-mind, but it doesn't really explore much of what it means to be individual within a hive mind

gray17,
@gray17@mastodon.social avatar

@zyd Alastair Reynolds's Conjoiners are close to what you're asking for, but it's been a while since I read any of the books, so I don't remember exactly what they were like (but I did enjoy them)
https://revelationspace.fandom.com/wiki/Conjoiners

gray17, to random
@gray17@mastodon.social avatar

How AI will transform your business (excalidraw edition)

troed, to llm
@troed@ioc.exchange avatar

I see lots of posts here on Mastodon where people state that today's "AI" (LLMs) have no use, waste energy and are just doing copyright infringement on a vast scale.

I don't get it.

I just put together "summarize.sh" - a bit of glue between some open source and self-hosted LLMs. It takes a Youtube URL as its only parameter, and outputs a summary in text of the important parts of the spoken words in the video.

That is, I run yt-dlp, Whisper and finally Mixtral 8x7b. And I no longer need to sit through someone yapping about for a few minutes to tell me what should've been a short blog post.

Example output from a 4 minute video:

"The text describes a video tutorial on how to reset a Corsair keyboard when it's not working properly. The keyboard in question has three white flashing lights at the top and is experiencing issues with its RGB lighting and key input. To reset the keyboard, the user should unplug the USB cables from the computer, hold down the escape key, and then plug the USB cables back into the computer while still holding down the escape key. After releasing the escape key, the keyboard's lights should flash, indicating that it has been reset. The tutorial notes that this method has worked for other Corsair keyboards as well."

How is this not a great thing to have?

#LLM #AI #Whisper #Mixtral

gray17,
@gray17@mastodon.social avatar

@troed @buherator every time I try an LLM for summarization, it's about 90% right and hallucinates 10% that sounds plausible but is very wrong. I never feel like it's worth it. Maybe if it were closer to 99.9%, but it's very far from that

gray17,
@gray17@mastodon.social avatar

@troed
"In 2021" - it was 2023
"In 2018 ... was retracted" - it was published in 2020 and retracted in 2022
"noise subtraction that may have removed too much data" - "removed too much data" is a weird statement
"has been accused of fabricating data" - Sabine doesn't mention accusations of fabrication until later, the APS investigation
"investigation by Nature" - was not reason for UoR investigation
doesn't mention it's the 4th UoR investigation
"stripped of lab" - missing "briefly"

gray17,
@gray17@mastodon.social avatar

@troed
and I think it misses a couple important points about the Schoen story.

these are kind of nits, but some of them might be important in some contexts.

also, I suspect that LLM summarization works better for scripted info-dumps like Sabine's videos. I've seen much more hallucination when LLMs try to summarize unscripted conversations

gray17,
@gray17@mastodon.social avatar

@troed the wrong years, I'm guessing the LLM interpreted "last year" based on a wrong idea of the current year, but this is bad. there's no way to know if a year in the summary is based on Sabine saying an absolute year or a relative year.

(also, the summary elides the distinction between publication date and retraction date)

gray17, to random
@gray17@mastodon.social avatar

How AI will transform your business:

Slide 1: complicated diagram of existing processes

Slide 2: simple diagram with "Inputs" on the left, "Results" on the right, connected by a single box in the middle, labeled "Magic"

green, to random
@green@transfem.social avatar

okay this is a persistent problem. whenever i try to type something in a gtk menu, and i press one of the underlined keys, it focuses that element, and doesn't let me type it.
here i was trying to set i3-sensible-terminal in pcmanfm, and couldn't get past the first e

gray17,
@gray17@mastodon.social avatar

@green my first guess is something weird about the kb modifier map? Those access keys are only supposed to trigger on alt+letter

timbray, to random
@timbray@cosocial.ca avatar

Many won’t have noticed, but the Fediverse “Trending” (called “# Explore” in the default Mastodon client) is pretty good these days. I use the phanpy.social client, and Trending is something that I now visit every day. Here’s a screenshot.

Something like this is an attachment point for alternative discovery algorithms, one of Bluesky’s more interesting features.

gray17,
@gray17@mastodon.social avatar

@michael @timbray you can often see trending from other instances, and phanpy lets you put that in a shortcut (or column)

gray17,
@gray17@mastodon.social avatar

@timbray @michael yeah, I wondered about that too. There are some differences, especially niche instances, but trending has a lot of overlap across instances.

Trending is: post must be public, not a reply, no cw or sensitive media. Posts are given a score that's reblogs+faves, decayed by post age, with half life of.. um, I forget, 4h? Reblog+fave count is local knowledge, so it's probably lower than actual. (Reblog count seems to get propagated, fave count not so much)

  • All
  • Subscribed
  • Moderated
  • Favorites
  • •
  • megavids
  • thenastyranch
  • rosin
  • GTA5RPClips
  • osvaldo12
  • love
  • Youngstown
  • slotface
  • khanakhh
  • everett
  • kavyap
  • mdbf
  • DreamBathrooms
  • ngwrru68w68
  • provamag3
  • magazineikmin
  • InstantRegret
  • normalnudes
  • tacticalgear
  • cubers
  • ethstaker
  • modclub
  • cisconetworking
  • Durango
  • anitta
  • Leos
  • tester
  • JUstTest
  • All magazines
    •