Kirk Roberts
@kirk_roberts
Followers
545
Following
2K
Statuses
1K
Joined January 2009
I've got a News and Views article out in Nature Medicine today. Grateful for the opportunity to share my thoughts on clinical uses of LLMs.
Large language models for reducing clinicians’ documentation burden. News and Views from @kirk_roberts @McWilliamsSBMI #LLMs #AI
0
3
26
RT @NRider70: How would receiving a diagnosis 3 years early impact patients with inborn errors of immunity (IEI)? Another terrific collabor…
0
7
0
@proftatonetti @drnigam @williamhersh @stevenbedrick It's been used by LLM-based methods before, e.g.
0
0
3
RT @Gregory47922793: Our systematic review on question answering systems for clinicians has just been published in JAMIA. We investigate th…
0
5
0
@zhizhid That's true. Currently I think they should just give you 3 pages to spell out your responses and changes. Original docs can't be changed. If that won't communicate the ideas properly because too much needs to change then probably a fresh submission makes more sense anyway.
0
0
1
@yoavgo Unlike other VR headsets, if you're wearing these and someone next to you tries to talk to you, you have no excuse to ignore them. But if you try to talk to someone else while wearing them, they will rightly slap you. Introvert's nightmare. No thanks.
0
0
1
@OptaJoe @OptaAnalyst How about not counting games the team still would have won without that player's goals? Because the alternative way of looking at your stat is "players who scored goals when their team was going to win anyway". Or "players who only score when their team is dominating".
0
0
2
RT @JamesMartinSJ: A superb two-minute explanation of one of the most commonly asked questions of Christian leaders. From our @Outreach int…
0
51
0
@thamar_solorio At a minimum there should be a central tracking system (@WHOSTP should care about this?) that provides # peer reviews/venue, ideally even conf chair and journal editor history. That would at least allow for identifying the worst offenders. Shame may be more powerful than rules.
0
0
4
Not just psychology that has a replication problem...
New blog: Common Flaws in NLP Evaluation Experiment "experimental flaws we have encountered while reproducing earlier experiments, including code bugs, UI problems, inappropriate exclusion of data, reporting errors, and ethical lapses. Pretty depressing."
0
0
1
@EhudReiter In medicine there's usually fewer decisions and the choices are simpler. With human trials, the protocols are written out and debated for months before any experiment. And yet even these often fail to replicate. Science is hard and we need to reconcile ourselves to that.
0
1
2
RT @AsmaBenAbacha: We are excited to announce the 6th edition of the #ClinicalNLP Workshop @ #NAACL2024 in Mexico City. ➡️The Call for Pape…
0
10
0
@nicktatonetti I scrolled to the bottom thinking it was a joke or something. The advice is all awful. Really I feel bad for the guy. Must have had a really toxic relationship with his advisor (or... maybe everyone).
0
0
3
This ad came up on my feed again, so thought I'd share my reply. I'll let you know when they block me :)
@DrOracleAI A reminder to those interested in what's in the medical literature: evidence. There are rigorous standards for medical evidence and this AI tool doesn't meet them. Go make an AI for crypto nerds, they're easy to deceive. But those who understand medicine know this AI is a scam.
0
0
3
@DrOracleAI A reminder to those interested in what's in the medical literature: evidence. There are rigorous standards for medical evidence and this AI tool doesn't meet them. Go make an AI for crypto nerds, they're easy to deceive. But those who understand medicine know this AI is a scam.
0
0
2
@zhizhid @johannes_hage @tugot17 I've been there :). It's amazing except there's an even better supercomputer next door.
0
0
4
@heartsalve Yep. The LLM rule is a corollary to your more general rule. Except there's never been a non-determinism hype cycle :)
0
0
2
Friends don't let friends re-implement already-solved computational problems with LLMs.
1
0
3