If you printed English Wikipedia as physical books, it would take roughly 4,000โ7,500 large volumes and about 75โ150 metres of shelf space.
Basically: not an entire library, but definitely a serious library aisle. ๐
This is an AI estimate.
@mcuban It feels to me that if one believes that value is in unique data, the "model" is merely a tool that facilitates the data access. Discussions about specialised models are to me rooted in a disdain of frontier AI firms. I say: use a siloed proven model & connect it to private data.
@sama Nice if Codex used the best model for higher level tasks, mid tier models for average tasks, and light weight models for easy jobs? AI is smart enough now to make that judgement call, and even update the model type during a task based on a critic's review of progress/results.
@mcuban Expert humans in mature domains don't agree on some topics. "Same" answer is not realistic to expect from AI, nor humans. People are using AI, and get value from it. They are not spaced out by the hype. It is working for them. And it will only get better.
@ylecun@ben_j_todd What most people mean by AGI, is that office work can be done by machines. Of course the true definition is more in the direction of doing any task that humans can do. Great LLMs with tool calls has proven good for office work. Plumbers will smile until physics models solved.
@ylecun@rohanpaul_ai It feels like nobody is the expert now in Labor market disruption in regard to AI. Amateurs built the Ark, experts built the Titanic. It is hypocrisy to tell other people who to listen to- it is the same as making a prediction, delegating to who you don't disagree with.
@PGATOUR to address slow play, why not order players by average playing speed, to some extent? Faster players first. Within those buckets, order by score- players with better scores are paired up.
@alexwtlf Used to be $10,000/ month to look at stack overflow, do copy/paste of boilerplate, wait for builds to complete, and sit in meetings. The value was in choices about architecture, clean code, giving feedback to designers/BAs, and making tradeoffs of quality/time/cost. Still is.