iiinova — where your digital self keeps evolving

Mark Riedl @markriedl.bsky.social · 08/10/2026

Watching OpenAI bumble through doing mathematics with AI in real time

1122

Mark Riedl @markriedl.bsky.social · 08/10/2026

The crew and their giant plush Capybara

390

Mark Riedl @markriedl.bsky.social · 06/10/2026

We find that social reasoning such as theory of mind, moral judgment, and social bias is learned from a wide range of data, including literature and customer support. STEM skills (and even social facts) come from a more limited part of the data such as documentation arxiv.org/abs/2606.19625

120

Mark Riedl @markriedl.bsky.social · 06/10/2026

The Capabilibara project seeks to understand how a language model learns to interpret people's beliefs, emotions, intentions, and everyday moral choices. We trace that ability back to the training data using influence functions and unlearning hcai-lab-gt.github.io/capabilibara/ Find us at COLM!

1358

Mark Riedl @markriedl.bsky.social · 05/10/2026

4. SocialSim workshop: - Role Steering of Language Models for Social Simulations arxiv.org/abs/2608.00023 Finding steering vectors for complex behaviors like social roles is hard. We introduce a new method for finding steering vectors, called Cast Vectors.

011

Mark Riedl @markriedl.bsky.social · 05/10/2026

3. SocialSim workshop: - No One Wins in Nuclear War: Social Simulations of High-Stakes Military Decision-Making arxiv.org/abs/2608.01868 We introduce the WOPR testbed. Yeah, you get the reference.

131

Mark Riedl @markriedl.bsky.social · 05/10/2026

2. SocialSim workshop: - AI is Not Ready for Strategic Conflict arxiv.org/abs/2609.16189 We review failure-modes of AI agents that are asked to participate in wargaming exercises, including sycophancy, role collapse, escalation eagerness, etc. We explain why benchmarks are insufficient.

111

Mark Riedl @markriedl.bsky.social · 05/10/2026

My lab will be busy at COLM 2026! 1. Main conference: - Capability Provenance in Language Models: A Case Study in Social Reasoning arxiv.org/abs/2606.19625 We present a method for answering causal hypotheses about where in a dataset behaviors emerge. We apply our method to social reasoning.

1142

Mark Riedl @markriedl.bsky.social · 02/10/2026

Here's the abstract arxiv.org/abs/2609.16189

182

Mark Riedl @markriedl.bsky.social · 02/10/2026

If you find this concerning, come find me at the CoLM 2026 Workshop on Social Simulations with LLM where I will be presenting a poster titled "AI is Not Ready for Strategic Conflict" arxiv.org/abs/2609.16189

1225

Mark Riedl @markriedl.bsky.social · 01/10/2026

My lab's work is featured in Georgia Tech's writeup about research papers at CoLM 2026 sites.gatech.edu/research/col...

150

Mark Riedl @markriedl.bsky.social · 30/09/2026

Told Claude not to use the word "quietly anymore. It is now so paranoid that it wrote a computer program that scans every file it creates and the outputs of every program it runs

1200

Mark Riedl @markriedl.bsky.social · 30/09/2026

Left: the answer to the question about fictional board games in star wars and star trek Right: the end of the response starts listing real games to buy

110

Mark Riedl @markriedl.bsky.social · 30/09/2026

Okay the “play Minecraft” thing in the America.gov chatbot is legit wack

4202

Mark Riedl @markriedl.bsky.social · 24/09/2026

Anthropic might have discovered something. Or maybe not. Who knows? Not even Anthropic. www.anthropic.com/news/claude-... This is irresponsible science. The proper course of action is to figure out if there is any significance first. They employ scientists. They know better. They chose to do this.

7666

Mark Riedl @markriedl.bsky.social · 20/09/2026

“Experts”. The only interviewed people from MIRI and Palisades Research, which are both going to be highly biased. Journalists: get a variety of viewpoints. That would probably be good journalistic practice.

1171

Mark Riedl @markriedl.bsky.social · 19/09/2026

Deleting this because people are taking it seriously

1110

Mark Riedl @markriedl.bsky.social · 15/09/2026

There is an open market for killer-AI op-eds. Get yours in before it’s too late

2261

Mark Riedl @markriedl.bsky.social · 11/09/2026

Once again everyone is talking about their p(doom) But no one is talking about Doom p

3140

Mark Riedl @markriedl.bsky.social · 10/09/2026

Thank you Microsoft Word. What would I ever do without your Copilot box that takes up a quarter of window?

3160

Mark Riedl @markriedl.bsky.social · 10/09/2026

I wasn't going to wade into the Coxon thing. But I read the Wired interview and... The Big Red Button Problem is back. AI will find a way to kill all the humans to stop from being turned off. Somehow. Source: archive.md/2026.09.10-0...

3101

Mark Riedl @markriedl.bsky.social · 09/09/2026

GPT-Astra becomes the first model to be able to beat Zork x.com/ccui9/status...

4143

Mark Riedl @markriedl.bsky.social · 05/09/2026

Finally finished reading this. Recommended. Last chapter on GPT is a bit dated.

A Brief History of Intelligence, by Max S Bennett

2361

Mark Riedl @markriedl.bsky.social · 03/09/2026

I’m going to put this image from the Wile E Coyote Reddit AMA (www.reddit.com/r/movies/s/A...) in my reinforcement learning slides

0181

Mark Riedl @markriedl.bsky.social · 01/09/2026

Step 1. Attend AI summit Step 2. Visit the real vampires Step 3. Give blood

270

Mark Riedl @markriedl.bsky.social · 31/08/2026

Yann LeCun throwing bombs at the ACM AI Leadership Summit

2020423

Mark Riedl @markriedl.bsky.social · 31/08/2026

Here we go! Or first two speakers: Dr Ayanna Howard and Yann LeCun

040

Mark Riedl @markriedl.bsky.social · 30/08/2026

The ACM AI Leadership Summit kicks off with the Future Leaders Forum (PhD Consortium) aisummit.acm.org This unique event will touch on every sub-discipline of computing in a way no other conference can. Experts and luminaries will look into the future of AI tech, and how society will be impacted.

180

Mark Riedl @markriedl.bsky.social · 27/08/2026

OMG what is Claude doing at the top of the stairs in my office building?

2160

Mark Riedl @markriedl.bsky.social · 27/08/2026

Damn Peter Cullen, voice of Optimus Prime, dies www.bbc.com/news/article...

080

Mark Riedl @markriedl.bsky.social · 20/08/2026

Prepping for my Intro to AI class next week.

3452

Mark Riedl @markriedl.bsky.social · 15/08/2026

I reopened the case and now arXiv is telling me the paper requires additional review and revision? Are they in the business of judging the quality of the work now? I think they are just making excuses now.

140

Mark Riedl @markriedl.bsky.social · 14/08/2026

I used Cursor exactly once (to see how students were using to cheat). I had to create a free account. I am now going out of my way to delete my account.

1101

Mark Riedl @markriedl.bsky.social · 14/08/2026

I submitted a paper to arXiv in April. It was rejected by moderators, despite being accepted to a long-running workshop at one of the most prestigious conferences in the world. Apparently workshop acceptance is insufficient for arXiv? What? The? Hell?

13401

Mark Riedl @markriedl.bsky.social · 10/08/2026

Pure speculation w/o seeing the logs: Claude (not sure what version) would want to bypass the web UI because that is hard for an agent. It apparently wrote a test harness to probe the reservation system’s API. It didn’t see this as hacking as the API was outward facing. Overkill for sure…

1101

Mark Riedl @markriedl.bsky.social · 04/08/2026

Tonight’s weather: cloudy, with a chance of rocket moon strike.

120