@karpathy We wrote a paper on this last year based on data from a randomized controlled trial calling this the "jagged frontier" of AI. https://t.co/vTbYZryoJI
Jagged Intelligence
The word I came up with to describe the (strange, unintuitive) fact that state of the art LLMs can both perform extremely impressive tasks (e.g. solve complex math problems) while simultaneously struggle with some very dumb problems.
E.g. example from two days ago - which number is bigger, 9.11 or 9.9? Wrong.
https://t.co/dUrR6wm8GC
or failing to play tic-tac-toe: making non-sensical decisions:
https://t.co/XarwfUBtod
or another common example, failing to count, e.g. the number of times the letter "r" occurs in the word "barrier", ChatGPT-4o claims it's 2:
https://t.co/xpffK2r0pv
The same is true in other modalities. State of the art LLMs can reasonably identify thousands of species of dogs or flowers, but e.g. can't tell if two circles overlap:
https://t.co/HCXxBxosAu
Jagged Intelligence. Some things work extremely well (by human standards) while some things fail catastrophically (again by human standards), and it's not always obvious which is which, though you can develop a bit of intuition over time. Different from humans, where a lot of knowledge and problem solving capabilities are all highly correlated and improve linearly all together, from birth to adulthood.
Personally I think these are not fundamental issues. They demand more work across the stack, including not just scaling. The big one I think is the present lack of "cognitive self-knowledge", which requires more sophisticated approaches in model post-training instead of the naive "imitate human labelers and make it big" solutions that have mostly gotten us this far. For an example of what I'm talking about, see Llama 3.1 paper section on mitigating hallucinations:
https://t.co/pjuxoIOJCY
For now, this is something to be aware of, especially in production settings. Use LLMs for the tasks they are good at but be on a lookout for jagged edges, and keep a human in the loop.
@stevenstrogatz@m_j_wiener > see his tweets below
I wasn't able to find the tweets without dinner digging. Should be easier to find by following backwards from this tweet
https://t.co/DRoCucENdH
@MarcosCarreira@m_j_wiener You guys realize what a precious thing it is to be able to do math together. I love it that we can understand each other, admit our mistakes if and when we make them (nobody’s perfect!) and figure out the truth together.
@patrickc When John Templeton said that, he was talking about learning from your mistakes instead of thinking "this time it's different" and making the same mistake again.
@kaushikcbasu Roger Penrose discovered Penrose tiles in the 1970s. These tiles use simple shapes to achieve a non-repeating pattern ("aperiodic tiling").
People observed later that this concept of aperiodicity had appeared centuries earlier with Girih tiles used in the darb-i imam shrine.
@ChShersh Michael J. Flynn, creator of the foundational 1966 Flynn's taxonomy for classifying computer architectures (SISD, SIMD, MISD, MIMD) based on instruction/data streams, passed away on December 24, 2025, at age 91.