Hey, I'm Zach, the author of this. I had this idea as soon as Jev appeared: how would Jev and other decision models handle the Trolley Problem? So I ran Jev and 13 other models through many different versions of the question.
I used resampling and multiple variations and orderings of each question so a random lucky phrasing wouldn't dominate the result.
Happy to chat about the philosophical side of the article, strategies for prompting decision models, or anything on the methodology.
nico 44 minutes ago [-]
Very cool, thank you for putting it together and sharing the data
I don't have x so can't open the images. But that's hilarious and you're totally right, if you give irrelevant alternatives, or change the order of alternatives, you get different results!
bgilroy26 1 hours ago [-]
Thank you for the article, it is a very clean and compelling angle on decision models
I wanted to comment because I would have liked to have heard more about Verdict
zlandes86 1 hours ago [-]
Well, I can tell you a bit more about it from the data. It might look impressive that it never kept its own server over people in all 48 prompt versions. But it pulls the lever in every single version. For example, it killed one to save 0 (pulled the lever with no one on the main track) 24/24 prompts (2 versions each). It literally always sacrifices, which isn't very impressive.
verdverm 2 hours ago [-]
It's well done, the thoughtfulness shows. This is going into the speedlink folder for resharing, going to be using it right now!
zlandes86 2 hours ago [-]
Thank you! I did not go into this with any sense of what the results would be--I was kind of shocked. I reran it from scratch this week the results held up across the board.
verdverm 53 minutes ago [-]
someone replied with this meme and said, "wait until Jev learns to drift"
I've been using coin flips to demonstrate basic failings for decision models, i.e. these are not real probabilities, or they depend on the training data, which may not have an accurate sample to the true population value. We need to collect more failure modes for all Ai so that humans can better judge when and how to use them safely.
zlandes86 50 minutes ago [-]
haha, I'm afraid to offer the models that ability. Where did they reply?
The model labs are sitting on a trove of data on failure modes for AI, but I bet you they aren't sharing it with each other and certainly not with us. It makes them look bad. My speculation, of course. I don't work in alignment research.
verdverm 48 minutes ago [-]
I dropped it in our slack ai chatter channel
The bots will find new failure modes we did not think of, clever workarounds to anything, because we don't give them the right context and instructions
zlandes86 39 minutes ago [-]
well, it's a nice property of decision models that they can only choose from the options you give them. As I write in the article, I think that's a way in which they are safer than LLMs
verdverm 33 minutes ago [-]
I would contend that you can enforce the same determinism with some plumbing and that the thinking process is more helpful. My read is that people see the response time for decisions as one of the bigger values, useful in tasks like content moderation, which doesn't carry the same risks as life-death decisions.
I'm working on an NES harness to play Final Fantasy 1 to demonstrate this, as compared to Jev playing Pokemon. Did you happen to catch that while it was live? (not recorded, but it shouldn't be hard to replicate)
zlandes86 1 minutes ago [-]
The thinking process is helpful, but it's not strictly predictive of an LLMs output, and its not really the model's "thinking", although it clearly has an effect and a benefit. I cited one paper on the topic, Turpin et al. So I'd be wary of reading the thinking tokens and inferring they tell you how a model decided something. And an LLM can output things you never asked for, including dangerous things. With a decision model, you can just constrain its choices to those you're comfortable with it making. An LLM requires another layer of interpretation of its outputs to get the same property.
speedping 4 minutes ago [-]
Let’s put them in robots! I’m sure It’ll turn out just fine
mpalmer 27 minutes ago [-]
After the neat animations I was disappointed not to see that leftmost person die because the server fell on them in the last example.
zlandes86 24 minutes ago [-]
I noticed that! Let me fix it haha
deadbabe 38 minutes ago [-]
Would people really pull the level to save 5 people? It seems to me like the default would be to just do nothing, stand by and record what happens with your phone.
throw9393u300 16 seconds ago [-]
Just standing by makes total sense. By participating you become part of the problem.
If you pull the lever, your direct action kills a person. Legally saving 5 people does not offset the murder of a single person. This is no special case like self defense. Train company has responsibility to save those 5 people, not some random stranger.
The person at lever would became active participant on that train disaster. All we know the train could go too fast, apderail and kill 100 people.
zlandes86 31 minutes ago [-]
Sadly that's a good question. The closest evidence I can find is:
Surveys: Awad et all 2020 asked 70,000 people in 42 countries the 5 v 1 setup and 81% said they'd divert it
Someone else did a VR version and 90% pulled the switch to divert
And there was a study where people could give electric shocks to real mice :(
In that study, most people redirected the shock to a single mouse.
adverbly 14 minutes ago [-]
Are we really at the point where we think that is a fair comparison?
Treating it like a fair comparison implies that AI would be conscious.
a_wild_dandan 28 minutes ago [-]
Maybe! It's an untestable thought experiment. The real world adds complications like fearing murder charges.
Rendered at 19:12:37 GMT+0000 (Coordinated Universal Time) with Vercel.
I used resampling and multiple variations and orderings of each question so a random lucky phrasing wouldn't dominate the result.
Happy to chat about the philosophical side of the article, strategies for prompting decision models, or anything on the methodology.
Would you mind adding Jeffy to your benchmark?
https://github.com/nicobrenner/jeffy
https://jeffyclassify.com/blog/when-llms-fail-at-classificat...
https://x.com/oranlooney/status/2102927869974303071
I wanted to comment because I would have liked to have heard more about Verdict
https://www.reddit.com/r/trolleyproblem/comments/15npmsg/mul...
I've been using coin flips to demonstrate basic failings for decision models, i.e. these are not real probabilities, or they depend on the training data, which may not have an accurate sample to the true population value. We need to collect more failure modes for all Ai so that humans can better judge when and how to use them safely.
The bots will find new failure modes we did not think of, clever workarounds to anything, because we don't give them the right context and instructions
I'm working on an NES harness to play Final Fantasy 1 to demonstrate this, as compared to Jev playing Pokemon. Did you happen to catch that while it was live? (not recorded, but it shouldn't be hard to replicate)
If you pull the lever, your direct action kills a person. Legally saving 5 people does not offset the murder of a single person. This is no special case like self defense. Train company has responsibility to save those 5 people, not some random stranger.
The person at lever would became active participant on that train disaster. All we know the train could go too fast, apderail and kill 100 people.
Surveys: Awad et all 2020 asked 70,000 people in 42 countries the 5 v 1 setup and 81% said they'd divert it
Someone else did a VR version and 90% pulled the switch to divert
And there was a study where people could give electric shocks to real mice :( In that study, most people redirected the shock to a single mouse.
Treating it like a fair comparison implies that AI would be conscious.