laverya

We all want our AI models to be trustworthy, right? But it would also be nice if they achieve the tasks we request of them. What happens when the task requires planning and responding to deception and betrayal?

For those who are unfamiliar, Diplomacy is a very simple 7 player board game where every outcome is deterministic - combat is purely a "more units beat fewer units" affair, with the defender winning ties. This means that real opportunities to advance generally come from convincing another player to help you out, thus Diplomacy. And of course while some players find great success from always being honest and sticking to agreements... That is not the most common form of human play.

I built this tournament to see what today's models would do when given reason and opportunity to form alliances, gang up on each other, backstab partners, and generally behave like terrible people. Spoiler: the best models are willing to do so, and will plan opportunities over the course of multiple turns.

Location: Michigan, USA

Remote: Preferred

Willing to relocate: yes

Technologies: Go/Golang, Kubernetes, Helm, Docker, Operators

Resume: https://laverya.com/resume/

Email: laverya at umich dot edu works

I spent the last few years shipping enterprise software with Replicated to 3rd party clusters - installation, licensing, support, reporting, and more.

Do those people actually _repair_ hardware, or do they swap bad hardware for good chips?

Can SpaceX not just say "OK, GPU #7 on satellite #15872 is broken, don't use it" and just accept that they're now overbuilt on power/cooling for that sat?

Australia and the UK are commonly used in the US gun control debate as places where gun confiscation worked, and CA/HI are the states with the most restrictive gun control policies.

My response was meant to illustrate that this was essentially not a "preventable gun death", or at least not preventable by any level of gun control ever implemented in a Western country. Similarly, the assassination of Shinzo Abe using a homemade pistol/blunderbuss was not a preventable gun death.

Agreed, but the difference in the use of rifles in assassination attempts between the US and UK/EU/AUS/etc can't purely be because of a lack of gun control in the US if the same rifles are available in those other countries too. (semiautomatic military style rifles like used in the first attempt on Trump are almost always more restricted overseas, but again this was only a single shot and could easily have been from a bolt-action rifle)

Even California and Hawaii don't ban hunting rifles. Same with Australia and the UK. Can you name a single country that totally bans the ownership of all firearms and enforces it? Would you like to live there?

(The rifle used isn't known yet, but only one shot was fired)

Yeah except for arpanet, GPS, satellites in general, jet engines, composites, computers, and everything that came from there... What has military r&d ever done for us?

Not to mention "just fly east" will get you to land just fine from Hawaii. It might not be the land you were looking for, but it will be land - so even if the direction finding failed the error would be survivable.

Compare that to computers, which had been around for the best part of 40 years... in 1980.

> The best national security is for it to be in the interest of the counter party to trade with you rather than attack you.

Sure, but if your biggest export is your country's particular flavor of monopoly money, is it actually in your counterparty's best interests to trade with you?

I had one at University of Michigan, but that was almost a decade ago now.

Third world countries unironically have bigger militaries than Canada.

Certainly Iraq pre-invasion made them look like a pushover!

That's not to say that invading would be painless, but let's not pretend that Canada would have a chance.

Has everyone forgotten the lessons of open-source cryptography all of a sudden? Just because the systems are public does not mean that they are vulnerable, and "security through obscurity" is no security at all.

> You don't get to "put back" security for SSN data when it's already been accessed by foreign governments.

"Good" news then, this is already the case! Or should very much be assumed to be the case, following the various GSA hacks.

The tariffs on Canada have now been paused after a promise to enhance border security (which was reportedly already in the works)

So who knows at this point

The official statement from the White House calls out increased fentanyl production in Canada as the reason - but doesn't say that it's the majority of imports, you're right. The twitter summary says "There is growing production of fentanyl in Canada, and enough fentanyl was seized at the northern border last fiscal year to kill 9.8 million Americans.", though, and that seems like a lot.

https://www.whitehouse.gov/fact-sheets/2025/02/fact-sheet-pr...

https://x.com/WhiteHouse/status/1885812342526460184

What I've been seeing is people saying that Canada is the source for most fentanyl coming into the US, not Mexico.

97. Well, I guess the US is one of the countries best able to implement Autarky today, so let's give it a try!

Trade makes things more efficient. Trade gives access to resources not available otherwise. Trade allows greater economies of scale, and production efficiencies via comparative advantage.

Trade is also not fundamentally necessary for the US in the way it is for essentially every other country on the planet, because the US is large enough, diverse enough, and resource rich enough to (with a fair bit of pain!) survive on our own.

I do not think it will come to that, but it is true.

Most rockets flew test flights before sticking people inside the same model, but most rockets are also single use and so each stack is fundamentally new.

A future starship could plausibly be the first rocket to fly to space unmanned, return, and then fly humans to space!

Imagine you're an ISP. If your customer has slow bandwidth to some random website, they will blame the website. If they have a slow connection to YouTube, they will blame you.

So YouTube gets more favorable terms on transit bandwidth than the random site does.

General aviation stats might not be entirely applicable to commercial aviation, of course. The planes are less redundant, less well maintained, and flown by less experienced crew - it's hard to tell which of those factors will tend to dominate, or whether they cancel out.

You only need to be moving ~7.5-8km/s, but getting to orbit tends to take 9.5-10km/s of total delta-v due to needing to ascend, aerodynamic drag, and other losses.

Don't forget the Cornfield Bomber! The pilot ejected from this F-106 (a single-seat, single-engine all-weather interceptor) after passing through 15,000 feet in a flat spin - and the plane promptly exited the flat spin and proceeded to a soft landing in a farmer's field.

And it was truly a soft landing - the plane's engine was still running after coming to a stop, and the aircraft was returned to service!

https://en.wikipedia.org/wiki/Cornfield_Bomber

It's not even that. You can host third-party speech without Section 230, you just also gain liability from moderating that content.

> Please explain to me how lower launch costs will help weather prediction.

Cheaper launch means more weather satellites covering more spectrum from more angles than otherwise.

> What is this special low cost rocket sauce that enables it?

Everything is dependent on cost. If we had a medicine that gave an extra 10 years of healthy life to everyone but cost $100,000,000 per person, it would be utterly infeasible to give to the masses. If it cost $100,000 - now that's an easy decision.

If something is cheap you can do more of it.

> I am trying to imagine how building reusable rockets leads to improving GPS

GPS satellites are incredibly expensive because they need to be light enough to fit in existing heavy lift launchers and reliable enough to last for 20+ years. Cheaper, heavier, more frequent launch means you can dramatically reduce the cost per satellite in a constellation, and thus send up more. Having more GPS satellites reduces time to first fix, improves coverage in adverse environments (cities in particular) and improves accuracy.

I think they can probably do both at once - heatshield testing and pez dispenser testing shouldn't interfere with each other, so might as well use whatever spare upmass you have!

Yeah you might lose the payload, but SpaceX has the cheapest satellites in the business from what I understand.

There have been quite a few demos over the years where voting machines are hacked. Now, this does not mean that they were hacked, for real, in a real election.

It does mean that it is possible to do, and in ways that paper ballots are not.

Perhaps we need to try the Tokyo model, with multiple competing subway/rail companies.

But unfortunately most cities are nowhere near the size and prosperity of Tokyo.

I think you're missing that the suggested path is not "let's seize the factories", but instead a return to "distributed charity" like we had before the Great Depression.