Ah, since Atlassian has been increasingly messing with Trello over the past couple years, it has really gone to shit. I currently have a Firefox profile dedicated solely to it, using >2 gigs of memory and about 1/3 of an M1 core. It has cumulatively used about a day's worth of CPU time in since I booted in 6 days ago. In contrast, the profile dedicated to Slack is using 750 MB and has burned about 27 minutes of CPU time.
One reason might be that one owns more than one set...for example, living in an area that has both hot summers and cold winters, we have summer sheets (sateen) and winter sheets (fleece).
> One is that with some frontends you can't actually get the raw context window so the LLM is actually more capable of seeing what happened than you are. The other is that these context windows are often giant and making the LLM read it for you and guess at what happened is a lot faster than reading it yourself to guess what happened.
I feel like this is some bizzaro-world variant of the halting problem. Like...it seems bonkers to me that having the AI re-read the context window would produce a meaningful answer about what went wrong...because it itself is the thing that produced the bad result given all of the context.
It seems like a totally different task to me, which should have totally different failure conditions. Not being able to work out the right thing to do doesn't mean it shouldn't be able to guess why it did what it did do. It's also notable here that these are probabilistic approximators, just because it did the wrong thing (with some probability) doesn't mean its not also capable of doing the right thing (with some probability)... but that's not even necessary here...
You also see behaviour when using them where they understand that previous "AI-turns" weren't perfect, so they aren't entirely over indexing on "I did the right thing for sure". Here's an actual snippet of a transcript where, without my intervention, claude realized it did the wrong thing and attempted to undo it
> Let me also remove the unused function to clean up the warning:
> * Search files for regex `run_query_with_visibility_and_fields`
> * Delete `<redacted>/src/main.rs`
> Oops! I made a mistake. Let me restore the file:
It more or less succeeded too, `jj undo` is objectively the wrong command to run here, but it was running with a prompt asking it to commit after every terminal command, which meant it had just committed prior to this, which made this work basically as intended.
There are a couple examples linked in another comment[1], so they do exist. The author of the game has stated[2] they excluded about 1/4 of all possible configurations to avoid including puzzles that required too much trial and error. Perhaps symmetry leads to more ambiguity than asymmetry, and therefore more than ~1/4 of symmetrical configurations were excluded?
Your comment made me curious about how often symmetry occurs in the full set of all possible 5x5 configurations. I took a shot at calculating this as an exercise, but I am a bit rusty when it comes to combinatorics...
First consider mirror symmetry via the center column. There are 2^5 configurations of the center column, and for each of those, there are 2^10 configurations of the left two columns. Since we are mirroring the right two columns from the left two, the number of configurations exhibiting mirror symmetry via the center column is 2^5 * 2^10 = 2^15. Rotating these 90 degrees gives us mirror symmetry via the center row, which is another 2^15 configurations. Mirror symmetry via the corner-to-corner axes, which also have 5 squares, is another pair of 2^15 configurations. So now we're at 2^17 configurations for mirror symmetry for the four axes.
Radial symmetry is slightly harder to describe, but it involves similar partitioning. You can partition the 5x5 grid into two 12-square subsets excluding the center square:
x x x x x
x x x x x
x x . o o
o o o o o
o o o o o
For any given configuration of the x-subset, you flip and reverse that to get the configuration of the o-subset. There are 2^12 possible configurations of the x-subset. Since there are two possible values of the center square, that gives us 2^13 configurations of two-subset radial symmetry. I believe rotating 90 or 180 degrees simply produces another configuration that has already been accounted for.
There is also four-subset radial symmetry:
a a a B B
a a B B B
D a . c B
D D D c c
D D c c c
however, I think these would all be special cases of two-subset radial symmetry. If I pick a random configuration for the a-partition and apply it to the other subsets, it matches a configuration that would appear in the two-subset group:
a a a B B X X o B B X X o o X
a a B B B o X B B B o X X X X
D a . c B => D X . c B => o X . X o
D D D c c D D D c c X X X X o
D D c c c D D c c c X o o X X
So between mirror symmetry and radial symmetry we have: 2^17 + 2^13 = 139,264. There are a total of 2^25 = 33,554,432 configurations irrespective of symmetry, so that's 17/4096 or roughly 0.415% that are symmetric...a bit more than 1/256.
EDIT: And by some hilarious bit of fate, I just went to go knock out a few puzzles to reset my brain, and the first one I completed[3] exhibits mirror symmetry in one of the diagonal axes. I'm pretty sure this is the first one I've hit in over 1000 solves.
If you concatenate all the clues_*.txt files, in order, to a single file, you can search through it to get the number of a pattern using standard tools, e.g.
I don't want to ruin the game for the community so will hold off on saying too much. I will say this- this game is very well designed so it's pretty straightforward to understand how it works without using anything more than the network tab of chrome :). No js, frontend puppeteering, etc. are needed. I might write it up when the game is close to completion!
Many people (more than the average rate for the prior generations) _did_ have problems. Perhaps more importantly, the only way to address those problems when they arose was to replace not only the keyboard itself but the entire top case of the machine due to the way the parts were integrated. This process costed many hundreds of dollars when the machines were out of warranty, and the company eventually acquiesced to social pressure and lawsuits by creating an extended warranty program.
That's not to say your situation is unique...there are probably many machines out there that have not had problems, including one owned by my wife. But there are also an unusually high number of machines that did.
I'm a native English speaker and nobody told me this (and I didn't manage to pick it up) until I was nearly 40. "Cost"'s past tense is also "cost."
There's another, newer, largely fatuous, verbed "cost" that means "to calculate the cost of something." That's the one that gets used in the past tense ("the projects have all been costed.")
"I've costed a keyboard replacement for my computer, and the total is more than the computer cost in the first place."
Probably similar to whatever its normal frequencies are for you, but perceptually louder. That seems to be my experience when I'm in a location with minimal background noise...
Sony sued Quad9 in Germany back in 2021 and won an early injunction[1], however Quad9 successfully appealed[2]. Interestingly, Sony is now going after them in Italy according to that second article.
reply