I've been worried that people trust me too much. A lot of the time I'll link a quote or a chart and give the source but most people never click it.
Makes me think I how great I could be at sharing misinformation if I tried. I'm going to make up a lot of nonsense and attribute it to random tech leaders.
Really appreciated how you held space for the nuances here rather than rushing to a neat verdict.
Your breakdown felt like an invitation to think more slowly and honestly about what we’re actually rewarding and what unintended behaviors might surface as a result.
It also quietly reminds us that every design choice carries a deeper ethical current, shaping not just outputs but values over time.
Thank you for mapping the terrain so clearly while leaving room for questioning. It reads less like a prescription and more like an open-ended conversation which is exactly what we need more of.
A little touch but seeing you do thjis--
chosen_seqs = [
"I love deep (learning) focus!",
"Cameron is great at explaining stuff",
"AGI is coming very soon...",
]
# Rejected prompt-response sequences
rejected_seqs = [
"I'm not a fan of deep (learning) focus",
"Cameron doesn't know what he's talking about",
"AGI is fake and LLMs can't reason!",
]
made me so happy. t's very cool to see you inject more personality here
LOL - I've been waiting for someone to notice this random stuff. I also put pictures of my dog into a lot of random papers / articles :)
Haha nice. Has anyone ever caught it before?
I've been worried that people trust me too much. A lot of the time I'll link a quote or a chart and give the source but most people never click it.
Makes me think I how great I could be at sharing misinformation if I tried. I'm going to make up a lot of nonsense and attribute it to random tech leaders.
I never make anything up! Just try to put random personal references into code / pictures haha
Always love to read your content! Great work!
Thanks so much!
Really appreciated how you held space for the nuances here rather than rushing to a neat verdict.
Your breakdown felt like an invitation to think more slowly and honestly about what we’re actually rewarding and what unintended behaviors might surface as a result.
It also quietly reminds us that every design choice carries a deeper ethical current, shaping not just outputs but values over time.
Thank you for mapping the terrain so clearly while leaving room for questioning. It reads less like a prescription and more like an open-ended conversation which is exactly what we need more of.