There's an opportunity to insert or remove a leap second twice a year. They only decide about 6 months in advance of each opportunity what to do (leap second, skipped second, or do nothing).
The Ouro looping results are interesting [1] and they are focused more on the improved reasoning from looping middle layers rather than the parameter efficiency aspect. They train 1.4 and 2.6B parameter models with 7T tokens. The training includes learning how many times to loop on any given token (there’s an early exit module). My guess as to why (as far as we know) looping is not in frontier models yet is that, at frontier training run scale, it’s probably going to require a lot of trial and error and at-scale research. While currently they already probably have a list of dozens or hundreds of of promising ideas that don’t complicate things as much. In the other hand, Ouro’s looping technique shows ability to compete well with models with 3x parameters which seems attention-getting to me. If there’s another 3x to be had down that path. It’s order of magnitude opportunity. Btw there is a great related work section in the paper.
I hesitate to propose ulterior motives, but given there have been several seemingly obtuse objections to projection from Rivian, perhaps the CEO is concerned that, if Rivian supports projection, it will harm the perception of the value of their software stack? Related, I think they licensed their stack to VW.
I've heard students react this way to seeing problems on exams that are not strictly of the types taught in class or in previously-assigned homework. "It's not fair! The teacher never showed us how to do problems like that!" This kind of thing was expected and assumed when I was in secondary school, that there would be combining of some concepts from the unit into a single problem. Very worried that there's both a cultural change supported by AI tools that will lead to the outsourcing of thought to the AI rather than outsourcing drudgery.
One problem here is that students that use AI to outsource thinking become people who cannot think. These people are not likely to be very useful to employers or even society. We have to figure out how to allow AI to outsource drudgery but not the thinking itself. It should be a better and better bicycle for the mind not a replacement for the brain.
There are many documented, exploited-in-the-wild font-file attacks (one example in 1]). Apple is re-writing their font interpreter specifically to improve security. [2]
Here a link to the best recent HN-featured long-form article on Japan rail network. Probably spent more time with this than any other item posted here in months.