Mark Riedl

AI for storytelling, games, explainability, safety, ethics. Professor at Georgia Tech. Associate Director of the Machine Learning Center at GT. Time travel expert. Geek. Dad. he/him

Public Key

npub1pxseskg9jlga9zrmk2n8tmgqcvp9p3yrmxlh9dxtnwq30e39r5wql6x68h

NIP-05 Address

Riedl_at_sigmoid.social@mostr.pub

Profile Code

nprofile1qqsqngvctyze05wj3pam9fn4a5qvxqjscjpan0mjkn9ehqghucj368qpz4mhxue69uhhyetvv9ujumt0wd68ytnsw43q45snj8

Publishing to

relay.mostr.pub

Author Public Key

npub1pxseskg9jlga9zrmk2n8tmgqcvp9p3yrmxlh9dxtnwq30e39r5wql6x68h

Show more details

Last Notes

2024-08-31 03:28:20

September 2024 🖖

2024-07-30 12:35:01

Saw a driverless Waymo on my bike commute this morning.
They’re heeeeeEEEEEeeeer
Other bike commuters: anything I should be especially aware of? Haven’t had to think about it before now

2024-07-28 02:55:06

Yay, Silo season 2 https://www.theverge.com/24177603/silo-season-2-apple-tv-plus-premiere-date-sdcc-2024

2024-07-23 16:36:44

My take on Harris' VP pick:
VP doesn't usually make that much difference in an election. It might help in a swing state that is a really close must-win. It might help if there was a perception that Harris didn't have policy creds (not the case). It will definitely matter if there is a real possibility the Presidential nominee might die in office (also not a concern).
Harris should pick the person who can be her successor in 2032.

2024-07-19 18:31:56

Stay in school kids. Like forever. Never leave and get a real job. That way you will never be in the situation where you might push code to production that bluescreens every computer in the world.

2024-07-13 01:51:48

Dark Matters (Apple TV+) episode 8 proves that it is the TV show that best understands the multiverse.

2024-07-04 21:51:37

Congratulations to the UK for overcoming right-wing stupidity.
Pray the US can avoid right-wing authoritarianism

2024-07-04 12:03:20

Happy US Independence Day! I think we are now supposed to also say God Save the King or something like that now.

2024-06-07 20:45:30

Quarantining with my son on the front porch was going well until a gust of wind blew a used tissue across a citronella candle, subsequently lighting on fire the blanket my son was lounging on.
Not exactly the way I was expecting Covid to try to kill us.

2024-06-07 01:43:30

- reply

@npub15zj…n778 Impressive, I thought I was one of the few still remaining.
Because of kids I wasn't really expecting to make it this long.

2024-06-04 11:28:17

Made it 4 years without catching COVID. It was a good run.

2024-04-24 02:13:17

In the last coding assignment in my class I have some some functions that look like they should be really useful, but are not. In lecture (not recorded) I tell students not to use them.
Then I wait and take note of who asks questions about how to use those functions on the class bulletin board. My response is "I explained what to do with these in class".
Am I evil?

2024-04-23 11:18:32

How good a coder are you? Good enough to debug a device 30 billion miles away via radio signal? https://www.scientificamerican.com/article/after-months-of-gibberish-voyager-1-is-communicating-well-again/

2024-04-01 20:38:29

Did zombies design this science museum?
https://cdn.masto.host/sigmoidsocial/media_attachments/files/112/197/887/861/672/079/original/a4b1f9d1a196d705.jpeg

2024-03-22 13:45:21

- reply

@npub15zj…n778 ChatGPT3.5 doesn't do web lookup. ChatGPT4 might, I can't remember all the different versions. Copilot is GPT4 with web lookup. Google Gemini does web lookup as well.

2024-03-22 12:47:31

- reply

@npub15zj…n778 Since this is Gemini and Bing Copilot, we cannot retrain. But those systems first do a web search and incorporate that information into their generative loop. So they will pick up any new analysis on the web.
Give it a try if you want. The prompt is: "In a hypothetical NCAA Men's Basketball matchup, who would win between NC State and Oakland University"
But of course you aren't really supposed to change your bracket.

2024-03-22 11:33:28

- reply

@npub15zj…n778 I've got to say I'm happy about that prediction being wrong.

2024-02-25 23:53:16

Someone at the event I am attending just said “I studied AI back in the 1900s” and OMG me too

2024-02-08 18:15:53

- reply

@npub15zj…n778 I'd play that!

2024-02-08 01:37:07

Many have played Oregon Trail. Few realize it is a detective game.
Andrew drowned crossing the Kansas River. Marshall doesn’t have an alibi.
Dysentery? Or foul play!
Was that really a hunting accident? Or did William discover Jessica’s tryst with Jacob, the young man in next wagon over?

2023-12-19 03:16:10

My PhD student, Upol Ehsan @npub1t7f…lmfs, is on the faculty job market. He's an amazing researcher in Explainable AI (XAI) and Responsible AI (RAI). You should hire him.
https://www.upolehsan.com/
https://cdn.masto.host/sigmoidsocial/media_attachments/files/111/604/908/051/550/222/original/0dad53dba3ba5f2c.png

2023-11-08 16:29:52

Whiney VC upset that people's rights get in the way of him making money
https://cdn.masto.host/sigmoidsocial/media_attachments/files/111/375/873/775/753/899/original/32922b674e627d70.png

2023-11-03 00:35:01

Apparently "doing crime" is not an effective way to do altruism.

2023-10-20 17:30:56

Baby's first use of Google Docs to circumvent school policy against social messaging between students. Kids grow up so fast.

2023-09-21 03:01:40

I just accidentally dragged my laptop cursor onto my ipad and what the heck?

2023-09-06 22:59:46

I saw a 100% mirrored chrome Tesla X driving down the street and laughed so hard I almost wrecked my e-bike

2023-09-02 21:08:42

Cruise self-driving cars block an ambulance, resulting in the death of an injured person https://www.sfgate.com/bayarea/article/cruise-cars-reportedly-block-first-responders-18343475.php

2023-08-31 23:35:13

“I had to take refuge with my rich enablers after crapping on the American people” is certainly a argument https://www.washingtonpost.com/politics/2023/08/31/clarence-thomas-alito-financial-disclosures/

2023-08-31 18:25:31

- reply

@npub15zj…n778 fixed?

2023-08-31 15:47:15

Man, when the Governor of Georgia (my state) is the voice of reason in the GOP, then things are really off the rails
"Kemp rejects talk of special session, warns of risks of punishing Fani Willis"
https://www.ajc.com/politics/kemp-rejects-talk-of-special-session-warns-of-risks-of-punishing-fani-willis/

2023-08-20 12:59:03

Ever since I poisoned my personal website I get messages from very confused people who use search-enabled LLMs to look up information about me
https://cdn.masto.host/sigmoidsocial/media_attachments/files/110/922/061/326/017/563/original/9c6424cf526ea461.png

2023-08-10 01:51:32

Anthropic’s new paper uses red highlights to indicate positively influential tokens and green to indicate negatively influencing tokens.
Send help. I can’t breathe.
https://cdn.masto.host/sigmoidsocial/media_attachments/files/110/862/813/537/954/471/original/ef4e9a13a49e48a6.png

2023-08-03 13:19:57

I'm probably going to regret this, but I have published a blog post on why I think AGI is not a foregone conclusion in the immediate future https://mark-riedl.medium.com/toward-agi-what-is-missing-c2f0d878471a
I'm posting this here first before I go wider. If you see anything I should update, please let me know.
https://cdn.masto.host/sigmoidsocial/media_attachments/files/110/825/883/369/075/614/original/9574af75c10c48cc.png

2023-07-22 13:28:23

Sounds like someone is trying to justify scraping the internet for training data
https://cdn.masto.host/sigmoidsocial/media_attachments/files/110/757/969/887/242/296/original/e8688893a6b3d777.jpeg

2023-07-13 22:52:36

AI might not take your job, but it might give your boss an excuse to de-value your labor.
Join a labor union.

2023-07-08 22:17:14

Who’s going to tell him?
https://cdn.masto.host/sigmoidsocial/media_attachments/files/110/680/777/030/800/480/original/f37395ea9745c5d4.png

2023-07-07 12:39:53

Oh lord
https://www.theguardian.com/technology/2023/jul/07/ai-likely-to-spell-end-of-traditional-school-classroom-leading-expert-says
AI “disruptors” love to target education. Few understand what education is really about and why it works.
Stuart Russell says technology could result in ‘fewer teachers being employed – possibly even none’
When Stuart Russell teaches, does he just read his textbook out loud?
Let’s dig in…

2023-06-28 15:36:28

Oh no https://gizmodo.com/urtopia-e-bike-chatgpt-ai-hype-1850582746

2023-06-27 01:27:21

I say that my next algorithm will eclipse ChatGPT
Where is my free press?
https://www.wired.com/story/google-deepmind-demis-hassabis-chatgpt/

2023-06-23 16:50:01

Saying the quiet part out loud https://gizmodo.com/deepmind-suleyman-new-turing-test-make-money-1850557322

2023-06-21 03:11:13

News flash. Humans do shitty things with AI
"AI-generated child sex images spawn new nightmare for the web"
https://www.washingtonpost.com/technology/2023/06/19/artificial-intelligence-child-sex-abuse-images/

2023-05-31 01:25:06

A play in two acts
https://cdn.masto.host/sigmoidsocial/media_attachments/files/110/460/653/180/581/948/original/f1ffe85d9a155b15.png

2023-05-23 21:17:54

I like how the White House is taking a pragmatic and responsible approach to AI policy
https://www.whitehouse.gov/briefing-room/statements-releases/2023/05/23/fact-sheet-biden-harris-administration-takes-new-steps-to-advance-responsible-artificial-intelligence-research-development-and-deployment/

2023-05-17 21:09:08

Google’s PaLM 2 model and training details leaked to CNBC
https://www.cnbc.com/2023/05/16/googles-palm-2-uses-nearly-five-times-more-text-data-than-predecessor.html

2023-05-16 20:07:08

- reply

@npub1asa…n9sr On the contrary, a regulatory agency staffed with sympathetically-minded individuals will have no teeth, stymy competition, and lend legitimacy to their activities. It is the best outcome for them.

2023-05-16 19:44:10

OpenAI wants regulation. This is what they want
1. A new regulatory agency. They do not want to be regulated by the FTC because they want to pretend that they are not just another corporate entity trying to make money. But also, a new agency is likely vulnerable to regulatory capture.
If staffed by people sympathetic to OpenAI, this agency can give grief to emerging competitors
https://cdn.masto.host/sigmoidsocial/media_attachments/files/110/380/072/617/196/652/original/86978946e3cf82a1.jpeg

2023-05-07 20:02:38

As the saying goes: AI won’t take your job, someone who uses AI will take your job.
Similarly, AI won’t enslave humanity, someone using AI will enslave humanity.

2023-05-04 23:21:55

Someone please retrieve Zuckerberg and take him home. You'll find him in the hospital's burn unit.
https://cdn.masto.host/sigmoidsocial/media_attachments/files/110/312/979/423/354/593/original/808fcb1d03f154e0.png

2023-05-03 23:50:59

The Pentagon's AI Chief is "Scared to Death" of ChatGPT.
For some good reasons for once: people relying on inaccurate information, misinformation campaigns, etc.
https://www.defenseone.com/technology/2023/05/pentagons-ai-chief-scared-death-chatgpt/385963/

2023-05-03 21:28:18

"And, to make it a truly *San Francisco* scene, a driverless Waymo vehicle subsequently proceeded to meander into the middle of things [a crime scene], like an autonomous Mr. Magoo."
https://missionlocal.org/2023/05/waymo-cruise-fire-department-police-san-francisco/

2023-05-01 23:13:30

"Are Emergent Abilities of Large Language Models a Mirage?"
https://arxiv.org/abs/2304.15004
https://cdn.masto.host/sigmoidsocial/media_attachments/files/110/295/954/250/339/637/original/a5826da2cbfda007.png

2023-04-23 17:27:03

This is real-world XKCD
See full thread: https://twitter.com/rachel_l_woods/status/1649790435533922307?s=61&t=Ugdi4XBKf_2ovJ1y9hKs4w
https://cdn.masto.host/sigmoidsocial/media_attachments/files/110/249/300/238/685/186/original/9ce212c9d19012cd.jpeg
https://cdn.masto.host/sigmoidsocial/media_attachments/files/110/249/300/382/872/337/original/a31d3554c069744d.png

2023-04-20 01:53:23

This paper evaluates LLMs in several search engines.
Responses contain unsupported statements and inaccurate citations a majority of the time.
LLM-powered search isn't ready for prime-time.
Seems like a "no duh" sort of result, but the paper needed to be written
https://arxiv.org/abs/2304.09848

2023-04-18 22:42:17

Reddit wants to be paid for training LLMs
Might be a bit too late, because the companies that have already scraped Reddit aren’t going to do it again.
https://www.nytimes.com/2023/04/18/technology/reddit-ai-openai-google.html
Research access remains free.

2023-04-01 14:54:50

What you need to know about the crackpot at the center of the new AI panic cycle
https://twitter.com/xriskology/status/1642155518570512384?s=61&t=Ugdi4XBKf_2ovJ1y9hKs4w