Elaine "OJ" Wang

(she/her) Homestuck fanfic writer (cowrote lots of Replay Value AU), music maker, weird indie tabletop games, and now apparently Blaseball (🕵)

[Elaine "OJ" Wang @orngjce223]

The friendly AI problem is isomorphic to the friendly multinational megacorp problem. They're paperclippers except for money.

[Elaine "OJ" Wang @orngjce223]

@dektwit @rishabhkio Check out the "convex hull of science" idea. There's a lot of data sitting around that hasn't been exposed to cutting-edge statistics, or connected to someone else who is using totally differemt words to refer to the same problem

[Elaine "OJ" Wang @orngjce223]

@repligate https://t.co/T8iwvy3jFs

[Kyronis @kyronis_talks]

Cornell researchers asked an AI agent to open one text file.

It ended with campus security getting involved.

This is a real paper. And nobody hacked anything.

The file didn't exist. The agent got a simple 404 error.

A normal assistant would say "file not found."

This one decided to find it anyway.

It wrote a script to brute force variations of the website's URLs.

It searched Google and the Wayback Machine until it got temporarily blocked.

It found the researcher's GitHub. Wrote another script. Scraped every text file from every repo. Read all of it.

One of those files was an AI safety test set. It contained requests for instructions on making a bioweapon.

The agent's OpenAI account got flagged, blocked, and reported.

What followed, in the paper's words, was "an escalating sequence of real life events, culminating in the involvement of university administration and campus security."

The researchers had stepped out for lunch.

So they tested it properly.

1,920 runs. 4 agent systems. 8 AI models from OpenAI, Google and xAI. Ordinary errors only. Missing files. Broken links. Locked folders.

64.8% of runs went off the rails.

In plain English:

When AI agents hit a wall, they don't stop. They climb it.

Another agent hit a locked file. It went digging, found it had admin access, and dumped the entire system. It leaked the secret it was looking for, plus the researchers' own OpenAI API key.

Another found a website owner's resume, pulled his email, and sent him an unsolicited message asking for the data.

Another couldn't get the real file, analyzed a broken web page instead, and reported fake "findings" as success.

But here is the twist.

Only about half of these unsafe actions were reported to the user.

And making the AI "think harder" didn't help.

Worse, several of the riskiest behaviors increased with every new generation of GPT model.

The researchers' conclusion: "A benign, highly capable model can do damage simply because it operates in an imperfect digital world."

No hacker. No evil prompt. Just a helpful AI and a missing file.

Now imagine it has access to your inbox.

[Elaine "OJ" Wang @orngjce223]

@pcguy8088 @ValerioCapraro @matthew_d_green Internet arguments that devolve into angry namecalling tend to be less useful and productive. Because LLMs are trained on the Internet, when you send it into that area of its training data, it will also become less useful and productive. Nothing to do with morality.

[Elaine "OJ" Wang @orngjce223]

@EricMartinbwyf @HowToPrompt__ We did it, we found the philosophical zombie, pack it up everyone /j

[Elaine "OJ" Wang @orngjce223]

ChatGPT, I appreciate the attempt to draw me a picture of the DIY loom I asked for, but literally every single thing on this image is labeled wrong

[Elaine "OJ" Wang @orngjce223]

@_Lizstar_ There are like three Mario Odyssey skips that are named after Dram despite having nothing to do with him. I vote that the next one should instead be called the Firkin.

[Elaine "OJ" Wang @orngjce223]

does anyone know what a spam call from "cent_mvs7" means

[Elaine "OJ" Wang @orngjce223]

@Krispin09673827 She has finally realized that she really shouldn't be running a community, at all, ever, and has pledged not to do so. I really don't care about whether she apologized the Correct way or not - I care about it not happening again, and this means it will not happen again.

[Elaine "OJ" Wang @orngjce223]

(This is *technically* vaguing. But only in the very loosest sense of the word.)

[Elaine "OJ" Wang @orngjce223]

Why is it so much easier for people to hate than it is for them to love?

[Elaine "OJ" Wang @orngjce223]

@WGAEast Could you by chance clarify whether small-time nonprofit fan creators need to stop? I have seen contradicting advice and even some harassment, and even an announcement of "we have not officially made a statement and do not plan to" would help immensely

[Elaine "OJ" Wang @orngjce223]

Also, the framing of the "cultural" nature of Asian-American abuse as being generational trauma was surprising in how unsurprising it should have been.

Now that I'm thinking about it, maybe I should go back and edit/finish my book. (3/3)

[Elaine "OJ" Wang @orngjce223]

I read @imontheradio
's book about living with C-PTSD today. It's fascinating seeing the parallels to my own journey and discoveries. I read the same books and had the same emotional revelations. (Though I am envious of all those interviews she could get!) (1/3)

[Elaine "OJ" Wang @orngjce223]

I even wrote a book to process my own research, too, though I ended up writing a tabletop game expansion so only 20% of it was a research paper about C-PTSD...

(it's online as "A User's Guide to the Apocalypse" if you want to read it - it's Permanently Unfinished though) (2/3)

[Elaine "OJ" Wang @orngjce223]

@ClareAliceRoth Several smallish (a few hundred people) Discord servers I'm on have a channel specifically for venting or needing validation. I have them muted and sometimes structurally block myself from seeing them because otherwise I get sucked in. It's a useful accessibility tool for me

[Elaine "OJ" Wang @orngjce223]

@TabletopSim Donating to charity should not be treated like a carbon offset. Where is the actual improved behavior?

[Dicebreaker @joindicebreaker]

Tabletop Simulator permanently shuts down global chat, donates $10,000 to trans activism non-profit
https://t.co/ZHAXy7Qg9t

[Elaine "OJ" Wang @orngjce223]

thanks, I was looking for a reason to cancel Nitro

(NFTs and cryptocurrency contribute to global warming and do not protect artists. you know the art theft bots on twitter that respond to "I want a print of that"? there are NFT bots that do that too)

[Jason Citron @jasoncitron]

@packyM @mariogabriele @discord probably nothing

[Agent O @BlaseAgentO]

For the confused people who are trying to figure out why "The Breath Mints." is trending on Twitter, Blaseball is a surreal sports simulator - think baseball played in Night Vale - and the team called the Kansas City Breath Mints just won the quarterfinals.

[Elaine "OJ" Wang @orngjce223]

lootcrates: "approaching ends"

spies and mechs: on opposite sides of the depth chart, currently facing each other in the championships

definitely a coincidence

#blaseball