Math papers be like:
Recall that the transpose of a matrix π, written as πα΅, is the matrix you get by flipping over its diagonal, swapping rows and columns.
Let π β ββΏΛ£βΏ be a diagonalizable operator with Jordan canonical form π = πβ»ΒΉππ. Then for any analytic function π defined on the spectrum Ο(π), the functional calculus gives π(π) = πΒ·π(π)Β·πβ»ΒΉ, where π(π) is obtained by applying π to each Jordan block individually.
@_vsavinov@zhongwen2009 https://t.co/1hmA3vW9GZ this paper only trains on top 20% of highest entropy tokens (forks in the road as they call them) and got some solid results
@jcjohnss@drfeifei@karpathy@cs231n cs231n was my first intro to deep learning! Made my first neural net with numpy whilst watching it. The excitement when it got about 75% on mnist will never be matched π
After thousands of papers on meta-learning, the approach that ended up being successful (ICL) was an accidental byproduct of language modeling. Serendipity at its best and a good reminder that research needs to be open-ended and pursue a diversity of goals to escape local minima.
What if you could not only watch a generated video, but explore it too? π
Genie 3 is our groundbreaking world model that creates interactive, playable environments from a single text prompt.
From photorealistic landscapes to fantasy realms, the possibilities are endless. π§΅
@jeffclune@_aadharna Hi Jeff. I saw your talk at the ICLR world models workshop and canβt stop reading your open-endedness work. Looking forward to reading this!
This drawing by Mr C. J. Richards shows the Great Pyramid of Giza (481 ft) transposed to scale onto the Suspension Bridge and Avon Gorge ( ~ 250 ft).
Mr Richards created it when he returned from a trip to Egypt to get an idea of the relative sizes of these iconic landmarks.