Conversation

Twitter is reinforcement learning for humans

220

jimmah

@jamesdouma

Replying to

@shivon

It needs a better reward function.

2:01 AM · Apr 28, 2022Twitter for iPad

Replying to

and

I just want twitter to reward not being a jerk. If it could do that, it would be a productive medium for discussion that attracts most of the population.

Replying to

and

it has way too Many negative rewards 🥲but positive ones are big and its crazy that people like you James can "reward" massivly Millions of people.

Replying to

and

+ Better training models, better annotated data, better inputs, better forcing functions, better checks and balances, better simulations. When does the machine learning student become the human master?? 🤔

Replying to

and

I keep thinking what is the best neutral reward function and I keep landing on empathy. If we can understand where people are coming from it will bring people together and we need that!