- Post autonomous weapons / DOD mess, I think they made some changes to make it more suspicious of what the usage is, particularly for malware. They also knew the government would be watching like a hawk, so its hedged to be extra safe.
- Because the tasks are running longer and more autonomously, they've raised the "self-confidence" level so it just makes decisions and stands by them more firmly.
- I think they've also slightly lowered the temperature so the outputs are more deterministic, so even if something has left context, it can make the same decision again with higher likelihood that it guesses the same thing.
- Lowering the temperature also makes it easier to sneak through some cached outputs (I think this likely only happens for first answers).
- They are deeply afraid of making sycophantic AI that creeps into the area of "addiction" like what happened with GPT-4o and opening themselves up to further legal liability.
A la carte in AI is going to be the name of the game for a couple reasons:
- Avoids regulatory scrutiny (for now at least)
- Nobody is actually entrenched enough for customers to matter
- Weird "celebrity" culture in tech, and AI especially. Everyone is looking for a "whisperer" or a "godfather" or whatever.
- Investors still get paid out
Smart operational talent will probably adapt by demanding higher salary, signing bonuses, severance packages in lieu of equity. Distribution of the true "lottery tickets" will get more uneven.
I was always looking for them because I was the weird nerd pointing out proper em dash, en dash, and hyphen usage years and years ago.
It's really only devs / engineers I see doing this, probably in some quest to create an indistinguishable voice in the name of productivity or something.
I've noticed people using emdashes more in known non-AI text in what I assume is a smokescreen to maintain plausible deniability when they wholesale copy AI text.
It's so interesting to me that human writing is subtly changing to mirror AI writing.
More like they see the future as more multi-modal, and they're probably right to think that is the best value approach vs. throwing more money at large language models.
I don't know if these statistics are even kept, but the current social environment in the US feels like a ripe breeding ground for cults. I've had so many people in the past couple years be like "I just want to farm with my friends and family and get away from all this".
- Post autonomous weapons / DOD mess, I think they made some changes to make it more suspicious of what the usage is, particularly for malware. They also knew the government would be watching like a hawk, so its hedged to be extra safe.
- Because the tasks are running longer and more autonomously, they've raised the "self-confidence" level so it just makes decisions and stands by them more firmly.
- I think they've also slightly lowered the temperature so the outputs are more deterministic, so even if something has left context, it can make the same decision again with higher likelihood that it guesses the same thing.
- Lowering the temperature also makes it easier to sneak through some cached outputs (I think this likely only happens for first answers).
- They are deeply afraid of making sycophantic AI that creeps into the area of "addiction" like what happened with GPT-4o and opening themselves up to further legal liability.