@emollick Right now Claude usage inside non-SW enterprises is held back far more by harness design and frankly Anthropic being far behind on features and product design for the enterprise than it is by the model.
@John_Hempton One technique I've used for longer conversations is to ask it to save down a "key notes" file before the context window gets full and then restart the convo by having it read the key notes file.
@BrendanPorath You can’t have JMO without also being okay with JDB. Golf being the only game where AARP members can occasionally compete with the youths is a fun distinction and I would argue we shouldn’t get too upset at the occasional senior citizen invite.
@SethBurn@TheZvi Introducing the concepts of seed point tiebreakers and whether to take two Sweet 16 teams or not would truly befuddle Claude Opus and 5.4. They are struggling with more simple concepts like future lockouts depending on whom you pick today.
@TheZvi After extensive testing, I can confirm that current frontier models fail my personal benchmark - figuring out the logic of NCAA Tournament survivor pools. I am not convinced that will remain the case in subsequent years.
@SliceTransverse The amazing thing is if I asked you to do it in real terms as an analyst, might take another hour. I assume it is seamless in your tool?
@JosephLaMagna Take I am workshopping: Hideki is what we thought Colin Morikawa was back in the day - good enough ballstriker that when he spikes with the putter (or more recently, driver), he wins.
I wrote about the Set Piece Revolution in the Premier League, why I think it's still just getting started, and my worries that it will require significant reforms to roll back
@johnarnold Judge Lopez also has a debtor-friendly approach and the bigger issue is non-random assignments of cases to those two judges. Lawyers would find a nexus to file in SDTX because in Delaware or NYC you couldn’t be assured of a debtor-friendly judge.