Web search helps language models answer their questions. But what about the questions LLMs don’t know to ask?
Introducing PriorBench, a new kind of search benchmark. 🧵
@mcuban If you are going to do this you should just tax ai datacenter energy consumption. There are so many ways to manipulate token count it would cause a lot of wasteful architecture decisions and punish small models and improvements in overall energy efficiency.
@proof_by_pecan I spent the whole first half of the movie complaining that in real life tox reports would solve this instantly. Despite getting schooled, I think it's still a little surprising neither of them thought of it before he pulled the trigger.
Tirelessly working away on a project like @haveibeenpwned all feels worth it when the community steps up and shares their appreciation for all the hard work involved in providing a service like this for free 🤷♂️
Fifty-nine members of the Senate just voted in favor of my amendment to block warrantless government surveillance of Americans' browser history. It failed by just one vote. McConnell is that much closer to giving Bill Barr the green light to spy on Americans' private information.