You can (and should) do RL from human feedback during pretraining itself! In our new paper, we show how training w/ human preferences early on greatly reduces undesirable LM behaviors, including under adversarial attack, w/o hurting downstream performance. https://t.co/YZSGnrT6lD
I’ve just found this in the bathroom at the Kennedy School and heard it’s circulating via email. Harvard folks: please do NOT sign this petition!
This letter is likely to lead to adverse consequences for sex workers and is not the best way to go about combating trafficking.
@mattieuMattieu Have each rectangle vertex be a node n in a graph, and give each edge (n1, n2) a weight that is the distance between n1 and n2. Then do Set TSP where each rectangle's 4 vertices is a set
Are you an #onlyfans creator? Sign up for our paid ($50/up to 75 min) interview study.
We want to understand your digital experiences!
https://t.co/BEJ2AECDh4
Blog posts on mindfulness in grad school (https://t.co/zkka1LdzwX), support from software in grad school (https://t.co/CCYMcwvVUX), and a pep talk for members who are several years into a long degree (https://t.co/uc5E48Ve1e)!
Margo St. James, who sought to decriminalize prostitution and make life better for sex workers, has died at 83. An erstwhile sex worker herself, she would begin speeches by saying, "Nice to see so many familiar faces." Obit by @kseelye https://t.co/79FCySOlpd
CDS Faculty Sam Bowman recently published his research on how our models learn bias. Read about the press coverage for the paper as well as Professor Bowman’s thoughts about the research in our latest blog post:
https://t.co/g8mzvD5xJF