OutsourcedMath Social Network

Joe Uchill 2 years ago • mastodon (AP)

Me too, AI. Me too.

Adam Shostack :donor: :rebelverified: 2 years ago • mastodon (AP)

As someone else said, all it needs to give a TED talk is adding inspirational background graphics

ares 2 years ago • mastodon (AP)

as with self-driving, ai chats don't have to be perfect, they just need to be better than the average human

Cassandrich 2 years ago • mastodon (AP)

So they're competing with the orange site?

Pwnallthethings 2 years ago • mastodon (AP)

A zillion ways this revolution is going to be great.

One way that's going to suck tho is in places where human interactions are a *useful* friction / proof-of-work. When it's not, those systems will get really overloaded

Pwnallthethings 2 years ago • mastodon (AP)

Few examples:

1. Online troll bots & fake personalities
It's going to get a *lot* harder to distinguish bots from people and much easier to create entirely fictitious credible online personalities to troll/harass/do crime

2. Persuasive letters by e.g. constituents to regulators.
Volume of (~sensible, unique) letters was a valid indication of sentiment. Soon won't be.

3. Ransomware victim communications & negotiations
Used to be one of the few costly areas in scale. Not for long

ares 2 years ago • mastodon (AP)

so you're saying we could automate discussions with reinstated haters on the other place

even more diabolical than shadowbans

Pwnallthethings 2 years ago • mastodon (AP)

4. Places where writing is assessed for cash or credit

Going to be much easier for e.g. students to fake/plagiarize essays or other homework for-credit

Combatting that will probably eventually force schools and universities towards towards more heavy focus on proctored work (exams, supervised essay writing) (ugh)

Going to be much easier to create fake-news clickbait farms too for ~ the same reason

This entry was edited (2 years ago)

Steve 2 years ago • mastodon (AP)

That latter one is going to suck. While systems like turn-it-in served the last 2 decades well for catching plagiarism, they really can't work with "an AI generated this to be genuinely unique, but not demonstrate student knowledge on the subject"

Pwnallthethings 2 years ago • mastodon (AP)

What worries me a bit here is that in trying to preserve some of these systems, there'll probably be a lot of folks pushing tech designed to *prove* humanness to combat it. And politicos and regulators are less likely to hear that and think "guess they should use creative CAPTCHAs" and more likely to think "oh, people should upload passports", and that's gonna be a whole new shitshow

Pwnallthethings 2 years ago • mastodon (AP)

Also excited for the new era of "who is liable when the AI tells you to cook the frozen Turkey in the deep-fryer" / "can you maliciously interact as user A with the AI, such that the AI interacts with user B and advises them to do X" wars

Pwnallthethings 2 years ago • mastodon (AP)

Anyway. New world. New cool stuff. New unforeseen challenges. Gonna be weird.

Jordy :chaos_emerald_blue: 2 years ago • mastodon (AP)

gonna be super weird. 🫣

Pwnallthethings 2 years ago • mastodon (AP)

Gonna add one point to this, since it's easy to focus on the downsides. There's some great stuff too.

One way AI is going to be *fantastic* is for accessibility. Imagine, for example, tools that summarize pictures as the default text for alt-text fields, making it faster and easier to make apps and sites accessible by default.

Another is making cool learning technologies, especially for beginners. Turning text into example code and vice-versa is a great learning tool!

Pwnallthethings 2 years ago • mastodon (AP)

And enabling technical creativity, especially by young folks / non-tech folks.

Imagine kids can *describe* a level they want to play as the starting point for making games to share with each other.

So, yes, big change comes with new risks, but also no shortage of new opportunities.

Alas, like most change, will be hard to see exactly how those shoes drop and the balance shifts across society in response until they do though.

That Anonymous Coward 2 years ago • mastodon (AP)

I think we shoudl rethink the Turing Test.

I think a much more valid test would be any AI that can understand & explain copyright law without exploding.

Allan Chow 2 years ago • mastodon (AP)

the scope of such a thing needs to be contained otherwise you can quickly get into a djinn giving you a million bucks or a 10 inch pianist territory quickly.

Allan Chow 2 years ago • mastodon (AP)

this is something pixel press should look into!
http://www.projectpixelpress.com/#left-and-right

S★m V★rm★ 2 years ago • mastodon (AP)

I was at an event this weekend for a biotech company that uses AI to create novel cancer treatments. Everyone at the table was convinced this is a net good for society, and that the field of medicine is more equitable than other tech-adjacent spaces. I argued that, just like everything else, the benefits will split 80/20, with the lower 80% of the population getting only 20% of the benefit.

I believe that #AI will disrupt faster than any rebalancing across society.

#ai

Roy Wiggins 2 years ago • mastodon (AP)

it's shockingly good at just taking descriptions of a text adventure and generating an entire game. I asked it for a Thomas Hardy-based text adventure here:

Rachel 2 years ago • mastodon (AP)

I just uploaded some screenshots with text, and clicked the add alttext automatically, and it did a pretty great job!

Would be neat for people viewing the post to have the same button!

Daniel Howard 🌻 2 years ago • mastodon (AP)

In the day, folks were concerned that writing would destroy our ability to think because we wouldn't have to keep everything in our heads. That's a fair concern, too. And the world keeps turning.

🦄 Vlt (Cédric Hebert) 🇫🇷 2 years ago • mastodon (AP)

The training aspect is great, let's see where this will go. The AI created a CTF challenge and it's write-up upon my request, but the answer came with a "this request might violate our code of conduct".
I hope infosec learning won't be sacrificed in the name of security!

Dante Mercurio 2 years ago • mastodon (AP)

Text to example code is a very cool use case. I’d look forward to that.

IsWacky 2 years ago • mastodon (AP)

I asked of openai playground bot the following request:

Ralph Nader said, "Republicans stab you in the front." The implication is that Democrats stab you, the voter, in the back. Please show examples of republicans stabbing you in the front.

It puzzled out the Q and gave a good answer!
(Cont'd)

Brian Grinter 2 years ago • mastodon (AP)

it would make it easy to add alt text here

But you just know some will abuse it for all the wrong reasons

bigiain 2 years ago • mastodon (AP)

This could make adversarial image attacks fun again!

https://www.theverge.com/2017/11/2/16597276/google-ai-image-attacks-adversarial-turtle-rifle-3d-printed

Fish Id Wardrobe 2 years ago • mastodon (AP)

tools that summarise pictures in a way that "sounds convincing but can be really really wrong"…?

WndlB 2 years ago • mastodon (AP)

Go though a Good Housekeeping/Underwood Labs 'Seal of Approval' portal?

Adam Shostack :donor: :rebelverified: 2 years ago • mastodon (AP)

"It is not recommended to cook a frozen turkey in a deep fryer. Turkeys should be thawed completely before being cooked in a deep fryer to ensure even cooking and prevent the risk of fire...."

Mark Gardner 2 years ago • mastodon (AP)

“AIs don’t kill people, #AI #programmers and trainers do”?

#ai #programmers

@krisnelson 2 years ago • mastodon (AP)

I will say that as of now, of all the worries people express about AI, who is liable is one of the lower on my list. It won’t take very long for lawyers to sort this out—and we’ll just start by suing absolutely everyone in sight to see what it sticks to, probably beginning with the theory they (programmers, host, referrer, everyone) are all jointly and severally liable for anything and everything.

Andrei Kucharavy 2 years ago • mastodon (AP)

you don’t want to know how easy are those models to hijack, especially on niche topics. At some point on Twitter I had a thread of Google search results that had artifacts due to errors in parsing or short-term fads that affected search results in a durable manner by creating an association between terms.
Given the linguistic specificity of different population sub-groups, you can also target the hijack to a specific sub-group you might need to influence/hurt/…

ScatalogicAI 2 years ago • mastodon (AP)

I don't understand why the AI keeps telling me to put a Sharpie in my butt.

Nour Agha :popos: 2 years ago • mastodon (AP)

At this point, some digital passport standard in the future feels inevitable.

Pwnallthethings 2 years ago • mastodon (AP)

@Nour Probably. Also I hate it.

@Nour Agha :popos:

Nour Agha :popos: 2 years ago • mastodon (AP)

It already sort of exists with stuff like Keybase and Keyoxide where you cryptographically verify your social identities and online presence, so I imagine governments would use a standard that builds on top of that. But
yeah.

Enrique Barcelli 2 years ago • mastodon (AP)

Have been thinking about this issue too, and I don't see how we can effectively prove we are humans through a software endpoint which is also available to a bot, unless we have a global/national complete record of humans identified with private/public keypairs.

AI is already capable of defeating our most creative CAPTCHAs.

These efforts are already existing in some countries. #Singapore #SingPass is actually this.

1/2

#singapore #singpass

Wry Shupe 🌲 2 years ago • mastodon (AP)

The educators in all areas I speak to are trying to get a handle on this.

Flipped curriculum (lectures recorded, classes proving comprehension in lab form) is actually how I'd like to learn if I was going to school again, and addresses this.

Unfortunately it also likely pressures people with learning styles that don't do well in that environment.

Mark Gardner 2 years ago • mastodon (AP)

It’s OK, the textbooks and lectures will be #AI - generated too

#ai

Michael S. Kirkpatrick ♾☸️⛷ 2 years ago • mastodon (AP)

Even proctoring is going to be hard. Tools like Copilot make it easy for students to “take” a coding exam with little effort if they have Internet access. It’ll be the same for other types of written work.

Nathan Arthur 2 years ago • mastodon (AP)

if the AI can credibly write the essay, why do students need to learn to write the essay? Why would students be expected to write that sort of thing in the “real world” if computers can already reliably do it well? (The only people who will need that skill are people who want that skill - so if they cheat, who cares?)

RRB 2 years ago • mastodon (AP)

But you know that the universities will on!y install automated graders to save money, so you end up with a GAN

Paul Coen 2 years ago • mastodon (AP)

I do feel like demonstrations of mastery via discussions, in tutorials, and apprenticeship (collaborating on research, etc) could actually be better anyway, but they’re labor intensive.

Probably will be more focus on showing all stages of work during longer term projects. Which would have been tough for me - I was a “structure the paper in my head and write it in one go” kid. I’d edit and revise, obviously.

Ed Sanders 2 years ago • mastodon (AP)

My first thought on seeing this was “so long to one of the easiest ways to spot phishing attempts.

Bad grammar was always the first tell.

Henry Farrell 2 years ago • mastodon (AP)

Bruce Schneier and I had a piece on this a while back - https://www.washingtonpost.com/outlook/2021/05/20/ai-bots-grassroots-astroturf/. Not that it was exactly hard to predict.

MJS :trans: 2 years ago • mastodon (AP)

Especially as machine learning gets better at emulating expected norms. https://www.technologyreview.com/2022/11/23/1063648/metas-game-playing-ai-can-make-and-break-alliances-like-a-human/

Meta’s game-playing AI can make and break alliances like a human

It could be a step toward building AIs that can handle complex problems requiring compromise.

^{Will Douglas Heaven (MIT Technology Review)}

Georgeann 2 years ago • mastodon (AP)

You probably already know this but when I went to search mastodon for you and exact replica of you came up only the icon was in black and white and it was labeled bot.

Enrique Barcelli 2 years ago • mastodon (AP)

this whole can of worms is pretty dangerous.

Imagine attacks with bots impossible to tell apart from humans to services like 911 or the suicide hotline? or how effective scams can be if a bot can effectively impersonate your mum, your partner or your child?

Yes, there is a very problematic world ahead of us, and I think the breakthroughs are not going to be in the AI vs AI realm, but probably in the introduction of 'human-to-human stoppers' in critical points of the workflows.

Mark Atwood 2 years ago • mastodon (AP)

if you follow and listen to randos you dont know, is it actually a harm if its an AI?

whetstone 2 years ago • mastodon (AP)

We’re going to have to go back to writing it in cursive on paper and mailing it in with a stamp

Cassandrich 2 years ago • mastodon (AP)

🤔 DoS courts with giant volume of realistic lawsuit filings with real scraped names and plausible circumstances. 😈😱

GenKnoxx 2 years ago • mastodon (AP)

I asked Marv what’s going to
Happen…

Screenshot of GenKnoxx asking Marv the AI chatbot what’ll happen when AI generates everything. Marv says we will just complain that it isn’t good enough.

Stever Robbins 2 years ago • mastodon (AP)

Technologists operate under the assumption that reducing friction is always good. I submit that many of our social problems today come from too little friction. From personalized on-demand media eroding formerly community experiences to one-click impulse purchasing.

dr2chase 2 years ago • mastodon (AP)

Let me guess who explained all the things to these chatbots.

Avi 2 years ago • mastodon (AP)

@Pwnallthethings

William Gunn 2 years ago • mastodon (AP)

I asked for a recipe for leftover turkey from bonappetit. I got a recipe that looked just like one from NYT cooking, but it gave me a totally made up link to bonappetit because that's what I asked for.

I mean, literally a link to Bonappetit.com that returned 404.

really? why? 2 years ago • mastodon (AP)

the last thing the humans invented was an automated way to confuse others

Pwnallthethings 2 years ago • mastodon (AP)

@MootQuestion Doesn't scale for marking tho, (unless you can get an AI to do that too)

@MootQuestion

Peter Principle 2 years ago • mastodon (AP)

We're already there.

https://www.technologyreview.com/2022/11/18/1063487/meta-large-language-model-ai-only-survived-three-days-gpt-3-science/

Why Meta’s latest large language model survived only three days online

Galactica was supposed to help scientists. Instead, it mindlessly spat out biased and incorrect nonsense.

^{Will Douglas Heaven (MIT Technology Review)}

trimyak 2 years ago • mastodon (AP)

I’ve been messing around with it getting it to write code for a bunch of stuff. One of the more fun things is coaxing it into writing simple malware examples- and then having it explain it, line by line. It’s really cool 😎

Ag Terrane 2 years ago • mastodon (AP)

Lawyers are having a field day with these over at Popehat on twitter.

Clara Listensprechen 2 years ago friendica

Marv the sarcastic AI? I thought Marv was a depressed paranoid.

Humbird0 Fandom 2 years ago • mastodon (AP)

A variation of this has already been a big problem for years with AI output often turning up in search results for fake SEO websites to generate sentences that look grammatically correct as a kind of advanced form of keyword-stuffing.

Nevermind that there's bound to be gullible people out there.

I swear, advertising ruins everything it touches.

JamesTDG 2 years ago • mastodon (AP)

they make me want to punch something a lot of the time

Tess :knuckles: 2 years ago • mastodon (AP)

hahah just like half the men in my life #amiright

#amiright

BostonPete 2 years ago • mastodon (AP)

Sounds like Trump

antipode77 2 years ago • mastodon (AP)

Stay skeptic.
Validate 'facts' by using multiple sources.

Tim Burnham 🇺🇸🇺🇦 2 years ago • mastodon (AP)

So these HUMAN chats are pretty cool.

One thing that's gonna cause some problems tho is that they sound really really convincing while sometimes being extremely extremely wrong

Cameron Purdy 2 years ago • mastodon (AP)

They have a future in politics?

Matt J. 2 years ago • mastodon (AP)

To be honest that’s exactly what I sound like when I talk about #Pilates.

Emma (@sailhpy ) says I sound very convincing to people who don’t know anything. But I am a bluffer (and she is not).

#pilates @SailyHPY