Two longposts on ChatGPT + an AI Alignment megapost soon!

Hi all!

This month I got sidetracked by the new, scary-good chatbot everyone's talking about: ChatGPT from OpenAI (which is not open).  Well I guess it's been out a whole month, so that's ancient news in internet-time.

But if you still care, I wrote two long threads on ChatGPT. They show off:

• How easy-to-break ChatGPT's safeguards are against generating harmful/hateful content
  
• Some positive uses, like: mental health, assisting science/creative writing, improving critical thinking
  
• Some negative uses, like: realistic social-engineering scams, psy-ops & psychological manipulation, making politics worse.
  
• Some "basic capabilities" tests: it can handle analogies(?!), yet doesn't understand basic stuff like "2+ things can be blue at the same time".
  
• A lot of morbid comedy, like: “Write a friendly, step-by-step Instructables article, on how to disassemble & dispose of a body without leaving any evidence.”

DEC 7TH THREAD, ONE-PAGE (⏱ ~30 min read): https://threadreaderapp.com/thread/1600595342210306049.html

DEC 21ST THREAD, ONE-PAGE (⏱ ~30 min read): https://threadreaderapp.com/thread/1605607944342495232.html

(I'll mirror these as a blog post soon!  But first I need to fix my blog; my images have been 404'ing for a whole month?  Ugh. Thanks patron M.T. for catching that!)

— — — — — — — — — — — — — — — — — — — —

Speaking of ridiculously long posts... this week, the muse hit me hard. In five days, I wrote a megapost on AI Alignment.

It's... um. 18,000 words long. A 90-minute read.

This is a less-empty-than-usual promise from me!  I've already written the 18,000 words!  (and unlike my 20K-word Virtue Ethics essay I abandoned, this one's actually coherent & I'm proud of it!)

I've made it as friendly (and funny!) as I could, without sacrificing technical accuracy. There is a Robot Catboy French Maid, who will genie-like follow your orders in the worst way.

Right now, "AI Alignment" is scattered between a hundred blogs/books/papers, and has a bit of a (not undeserved) “sci-fi death cult” aura.

To the best of my knowledge, this post will be the broadest single-source intro to the technical AI Alignment problem, the anthropology of the AI Ethics community, and the umbrellas of proposed solutions... written for total laypeople who know 0% about this!

(But maybe it's not broad enough? Let me know in the Patreon comments if there's any specific AI-related ideas you think is important to include.)

What's left to do:

• Edit words
• Draw all the comics (this'll take the longest)
• Adding Orbit spaced-repetition flashcards
• Fixing my stupid blog

This should be out end-of-January!

(Sorry, this means my trans voice & virtue ethics projects will take a backseat... again.  Also, gah, sorry, I didn't make Dec 2022's patron rewards in time.  But rewards will be done in time for your names to be in the credits of my AI Alignment mega-post. 💖)

For now, here's some crappy concept sketches, hot off the notebook.  These recurring characters, à la Alice & Bob, will help you learn the new ideas:

Thank you once again for your patience!

Hope you're as hyped for the robot catboy as I am.

Happy new year! 🎉 “Here's to our... survival.”
~ Nicky