A lot of them have rules forcing them to have some amount of exposure to indexes of a market, or all entries in a market.
There are MAJOR rule changes made to allow them to do this (90 day wait-time reduced to 5 days, financial stability requirements lowered or removed), which is why automated rules like that were created ("oh, if they make it to X, they were already vetted for Y, Z").
A lot of people are throwing a lot of trust and reputation on the bonfire to make this happen.
How about measuring the political bias in base reality?
> We want Claude to be seen as fair and
> trustworthy by people across the political
> spectrum, and to be unbiased and even-handed
> in its approach to political topics.
So, a lot of 'seems' rather than 'is.' They are training this to be non-offensive, based on some weird moral / media takes that shift year to year, political cycle to political cycle to political cycle. Not for bring forthright and fact based.
Wow.
This would be 'meh' if it was some weird individual. This is a 'whitehouse.gov' domain. This our our the Trump Administration folks. Acting like immature 13 year old, and making jokes no one under 40. And doing it using federal resources to do political party activities.
After decades of 'proprietary, respect, professionalism', this kind of childishness is just gross and embarrassing.
This is a false dichotomy presented to people. By framing this as 'Giant government, or giant business?" you are going to get crap answers.
None of these are one size fits all solutions, and there should be a mix. We have a working patch-work of laws in physical space, for a reason. It allow flexibility, and adjustments as we go, as the world changes. We should extended that to virtual space as well.
Age / Content Labeling and opt-in/ opt-out for some content. Outright ban on other kinds of content. A similar "I sue when you abuse my content" for copyright, impersonation, etc.
One size does not fit all, and is not how the real world works. Online shouldn't work much differently.
Great, please apply this to yourself, and livestream all of the time, leave the rest of us out of this.
People don't have one standard of behavior. I won't tell my kid jokes, I tell my wife. I won't complain about people in public, the way I vent to my sister (who gets it is just me venting, not how I feel all the time". I am not going to speak to a cop as I'm getting a speeding ticket, they way I will talk to one who is harassing a friend at a parade.
I won't talk to / about a co-worker in a meeting, the way I talk to someone he just (rightly, but very meanly) chewed out, and who needs a boss who listens, or will I talk to him in a meeting the way I will (a tad later) chew him out for making a coworker cry.
This take is so naive and emotionally / socially unintelligent about human behavior in various situations.
This is the fascist dream. Who defines 'Best Behavior?' Whoever owns the cameras, and the cops.
Criticize the president? Not best behavior.
Kiss someone of your sex/gender? Not best behavior.
Call AI stupid? Not best behavior.
Whistleblower on out a deadly chemical leak? Not best behavior.
Disagree with a politician? Not best behavior.
Defined LGBTQ+ people's rights? Not best behavior.
Criticize Isreal ? Not best behavior.
They are going to define "best behavior" in a way that never threatens their feelings, let alone threatens their powers, if you let them.
Results that are too good to be true! The probably are. If this study was reproducible (it is not) someone would have started using it, bragging about it, and we would have 'recovery vacations' to cruise lines that would de-age us by years.
Extraordinary claims require extraordinary proof, and this just isn't it.
There are MAJOR rule changes made to allow them to do this (90 day wait-time reduced to 5 days, financial stability requirements lowered or removed), which is why automated rules like that were created ("oh, if they make it to X, they were already vetted for Y, Z").
A lot of people are throwing a lot of trust and reputation on the bonfire to make this happen.