Jacob Hoover Vigly

@postylem.bsky.social

jahoo.github.io

- Cast controlled generation as an inference problem, with the LM as a prior and verifiers and scorers as likelihood - Use Sequential Monte Carlo to sample from the resulting posterior Library w/ tutorials for setting up your own controlled generation inference problems: github.com/genlm/genlm-...

GitHub - genlm/genlm-control: Controlled text generation with programmable constraints.

Controlled text generation with programmable constraints. - genlm/genlm-control

github.com

I just realized that if the ‘gumbelmax trick’ (look it up) is defined by using argmax on Gumbels, and identical ‘expmin trick’ uses argmin on Exponentials, then using argmax on Betas (precisely equivalent to the previous two), gives what really should be called the ‘betamax trick’ 📹

If you're an NSF PI like me, you might be wondering: Will my grant be suspended or terminated next week? A few key things to know: First off, remember that NSF is **congressionally mandated** to evaluate broader impacts. And laws like CHIPS and Science supersede EOs. But also... 🧵 thread

Dr. David Miller 🏳️‍🌈@davidimiller.bsky.social · 2y ago

Found this gem in an arcane NSF doc...key to know: The Constitution and law SUPERSEDES exec orders! "In the event of a conflict between policies issued at a lower tier versus policies issued at a higher tier, the higher tier policy will take precedence" nsf-gov-resources.nsf.gov/files/Resear...

NSF froze payments to researchers. State AGs said it was unconstitutional for exec branch to pause congressionally mandated funds. Judge issued restraining order. Payments resumed. But some inconsistencies in what the NSF has said (thread): www.statnews.com/2025/02/02/n...

National Science Foundation restores payments after five-day pause, but worries over science funding persist

The NSF said on Sunday that its payment system was back on line, after it froze out researchers and sparked financial worries.

statnews.com