6/12/2025

yes im very late

this article argues about how mech interp is not that useful and it raises a lot of good points (i.e. in biology / other complex systems, we don't do the bottom up approach, and a lot of the buzz from mech interp comes from cherrypicked results, the compression involved leads to the loss of edge cases, so much has been invested but not much has come out, this post talks about how google stopped prioritizing SAEs because they weren't performing as well). i think i should definitely keep this in mind as i get more interested in interpretability; i've heard friends make such comments too. meanwhile this article by dario amodei urges for the importance of mech interp, tho he is from anthropic which may influence his viewpoints. he advocates for the govt to also get involved, citing all the bad disasters that may happen because ai is too "opaque". i get his points too; the race between the development of ai models and our understanding of them is pretty concerning. anyways, i think i'll still be exploring interpretability which the first article acknowledges is more interesting to those coming from a math background who enjoy rigor and proofs. i'll just keep both sides in mind, and the reference to focusing on top to bottom approaches is valid, with developments in ai safety like representation engineering etc. here's a really neat resource with lots of resources for ai safety - https://www.aisafety.com/map

cool application of equidistributed sequences to isl problem 


tessellation thm - Wallace-Bolyai-Gerwin theorem 

two polygons are "equidecomposable" iff they have the same area
this is rlly cool and powerful :O


this article talks about an advancement in fermat's last thm along with langlands program (?) that makes use of high tech computing

Comments

Post a Comment

Popular posts from this blog

no more ap exams ever!

i've officially closed out one chapter of my high school career - ap exams! which means now i have a lot of time on my hands to do things i'm interested in :) last week i read this article  except i didn't understand a lot of it. but topology is cool! and it's pretty sad if you write a whole book on something that is only a few months later proved to not exist a lot of cool quanta articles popped up in my newsletter so i hope i can read them all and update this. in terms of other tech-related news, i heard about alphaevolve, which has been causing a lot of stir on the internet recently! in a later post i hope to dive further into the developments of google deepmind which are very relevant to the intersection of math and cs. i also heard about vibe-coding, a term coined by andrej kaparthy. there are tools like cursor.ai that help with writing code; i hope i can test it out soon and show you the results. i feel like it is still important for us to not be so reliant on ai ...
Read more

may 25 2025

 5/25/2025 is a pretty date :) okay so i'm getting back into posting half-baked things i guess i discovered curius and was thinking of using that other than this but i'm better at conveying myself on a blog? perhaps i read this quanta article on superdiffusion  and it's super cool how mathematicians are able to formalize the chaoticness of things like rubber ducks floating in a river. i'll include the gif from the article here because it's very aesthetic :O soap bubbles!  also the article emphasized the methods  involved in proving the superdiffusion conjecture, mainly renormalization (which i don't really understand) and homogenization (which the mathematicians emphasize is more important that they showed these techniques are very useful).  quantum field theory in general is pretty interesting because it's like modern physics + math, and some high schoolers are doing research in the math part of it ;D which brings me to the next thing i read which was an ar...
Read more

happy may

 i guess i'm just a girl who loves half-baked things. go to maybe an about page i'll create with some more info about myself. here goes a low quality blog that hopefully doesn't reveal too much about myself while accomplishing what i want, mainly to share what i've learned and what hobbies i like i guess i hope i can write something out everyday, and this will be a good motivator for me to continue exploring what i love i just read this quanta article on how the physics principle of spin glass contributed to the idea of a hopfield network, which is a form of an rnn that has "associative memory." hopfield networks influenced the boltzmann machine and modern neural networks. their creators, geoffrey hinton and his colleagues, won the nobel prize in physics this year for their work in this area. this article  is a good refresher on some ai concepts.  this other article on the fusion of math and cs for proofs is especially interesting because that is sort of what...
Read more