Review · Speculative decoding1 / 7
RightSignalWeek of Oct 19 – 25 · review
Review · Gradient descent2 / 7
Why might gradient descent end with a decent but not perfect answer?
Review · Huffman coding3 / 7
A file uses all 256 byte values equally often. How much does Huffman coding save?
Review · Mixture of experts4 / 7
A model has 64 experts of 1 billion parameters each and uses top-4 routing. About how many expert parameters work on each token?
Review · HyperLogLog5 / 7
Why don't repeat visits from one person raise a HyperLogLog count?
Review · Floating-point numbers6 / 7
A 32-bit float has 24 significant bits. What happens when you store 16,777,217 (2²⁴ + 1)?
Review · Inter-frame video compression7 / 7
Why can one burst of lost packets on a video call cause smearing for a second or more?
Almost there.
7 left unanswered. Swipe back to finish them.
SwipeScroll or use arrow keys