Or you could, you know, let people have access to the base model and engineer their own system prompts? Instead of us hoping you tweak the only allowed prompt to something everyone likes?
Also I should add, being a Factorio veteran with 2-3k hours in this game, I think the goal of making the "largest possible factory" is too vague and not the right metric. When Factorio players make large megabases, they don't go for "size" per se, but rather science research per minute. The metric you should be telling the agents is SPM, not "largest" base!
There was a HN post here not too long ago about a team that used reinforcement learning to train an agent to beat pokemon red. They mentioned how they had to tweak the cost function to give small rewards for exploring and big rewards for completing "essential tasks" like beating gyms.
I wonder if this same approach could be used here in factorio? Using the pokemon red analogy the main "essential tasks" in Factorio are setting up automation for new items and new science packs. I think a good reward function could involve small rewards functions for production rates of each item/sec, medium rewards for setting up automation for new items, and big rewards for automating each new science pack.
Telling a Factorio agent to just "make a big factory" is like telling a pokemon red agent to just "beat the game", it has to be broken down into smaller steps with a very carefully tuned reward function.
Thinking about this is really making me want to jump into this project!
> if you have a path in the upper bunk that gets broken you are screwed
Counter-factuals like this don't apply when talking about average probabilities. If you cross over, it's an identical graph with identical probabilities. idk, to me it seems really counter-intuitive that the opposite bunk's node would be easier to get to than the current bunk's node.
Suppose all the groceries raised the price of item X by exactly the same percent Y on the same day. At first glance this would look like perfect collusion in the data. But how do you know it wasn't just the supplier price going up by Y percent?
Suppose we took this line of thinking to its natural conclusion. If we wanted to be completely neutral, in theory we could have no male/female division at all, and just compete for the "best human" in each sport. But because of the vast biological differences between men and women, men would win every single time, hands down. This goes back to why do we have men and women divisions in the first place? Because we want a space for people without the vast advantages from male physiology to compete fairly with each other. Allowing for people who are technically "women" based on their reproductive anatomy, but have all the male physiological advantages otherwise, feels like it defeats the entire purpose from having separate women divisions.
Except sinusoids are special in that they are natural solutions to the Helmholtz wave equation. There's other problems too like square waves having infinite energy. This article might make sense to a mathematician or computer scientist but neglects the underlying physics of sound and waves.
For anyone who's interested, there's a group of us dedicated to natural myopia reduction at reducedlens.org, which is a free and open-source fork of endmyopia. I've even started measuring my axial length to try and get better data on if this stuff actually works (only one measurement so far though, so nothing interesting yet).
The nature paper was pretty crazy. Basically because blue bends more than red (think of a prism), it also focuses a bit sooner. This phenomenon is known as longitudinal chromatic aberration, or LCA for short. This means if you're myopic, blue might be more blurry than red, and vice-versa if you're hyperopic. The researchers in the nature paper had participants watch a movie where they straight-up blurred the blue or red with software, in order to produce fake LCA signals. They found the participants axial lengths still shortened or lengthened anyway in response.
The two-state system was the first example I learned in undergrad stat mech, and really helped me understand the first-principles definitions of entropy, temperature, etc. My parent comment was more for the casual HN reader, but if you really dig into the two state system, negative temperatures aren't all that weird.
The first thing to understand is microstates, which is just the number of ways a system can have a certain energy. Eg in a two-state system with ten particles and energies +/-(E/2), there's one microstate where the energy is -5E (all negative), ten states with -4E (one elevated), etc. Then entropy is just the log of the # of microstates, which is much easier to deal with, since microstates tend to behave exponentially. Eg entropy(E=-10) is log(1)=0, entropy(E=-9) is log(10), etc.
Then temperature like you said is d(entropy) / d(energy). Two systems with different temps brought into contact will exchange energy until the temps are equal, since this configuration maximizes entropy.
The two-state system can have negative temperatures since entropy starts decreasing with energy once more than half of them are in the higher-energy state. This can't happen in more familiar scenarios (eg ideal gas, blackbody, etc) since usually entropy always goes up with energy.
Everyone here seems totally lost on the physics connection. Suppose you have a box of atoms, each atom can be in one of two states, a low energy E1 and a high energy E2. If the box has a temperature T, then the probability that any atom is in state E1 is e^(-E1/kT) / [ e^(-E1/kT) + e^(-E2/kT) ], and similar for E2. As you lower the temperature most of the atoms gravitate towards the lower energy state E1, and as you raise the temperature they gravitate towards a 50/50 mix of E1 and E2.
Also the novel rulesets make it so much more interesting than vanilla sudoku. Oftentimes the puzzles have really abstract/mathematical/beautiful break-ins. Definitely adds a lot of variety and replayability.