@krishnanrohit Not sure on this — @AdamUnikowsky has a lot of great writing on AI & Law, and he’s used a Claude for almost all of it. E.g. https://t.co/oss6vf0ycZ. I imagine that’s what’s prompted the remark here!
🎲 Game on! Our new MScAC squad is levelling up with board games on day two of orientation! Special shout-out to our MScAC alums Nathan and @RJRedelmeier
who are tonight's hosts, bringing all the fun for our competitors📷 #UofTMScAC#OrientationWeek#UofTBackToSchool#UTogether
The Polish poet Wisława Szymborska was born on this day in 1923.
In “On Statistics,” published in The Atlantic in 1997, Szymborska takes the language of data and uses it to measure some of the messiest aspects of human nature:
@cohere Makes the point that this technology is incredibly promising for productivity, “can drive the growth we need to see”
Do think people said the same about all the tech of the past few decades amidst stagnation, but AI definitely is unique.
Really excited to be at #CollisionConf this week, and thanks to #Intuit Canada and the @UofTMScAC program for this opportunity!
Starting strong — a talk with @aidangomez about the real world impact of AI.
@cohere Two things to watch out for: don’t get trapped in a proof of concept death cycle, and make sure that your employees have access to this technology because they need it and want it at work (something Intuit has done very well IMO).
@J_Jankovic@RohanAlexander@TorontoSRI That's valid.
I took it as a worthwhile reminder that while we'd really like to think that research is a strong driver of policy, to the point that mistakes in single papers can drive mistaken public policy, it's unlikely to be that clear/causal.
(Fun story all the same!)
Really enjoying @RohanAlexander's @TorontoSRI talk on Improving Reproducibility in Quantitative Social Sciences. Appreciate the wide swath of background, and especially this anecdote: https://t.co/qaS4I5XZPt
(Though note https://t.co/ORhqnr4noN )
Highlights https://t.co/mpg7oirOdp -- submit code, LLM tests with a number of prompts/approaches, then creates github issues if/as necessary for issues in code.
Maybe this is easier to integrate than copilot-esque code assistance? More comprehensive, probably.
When using LLMs to build that testing, there's a lot of variance based on prompting. I wonder if something like https://t.co/ui8ObY1NH5 could come in here? (Package to better optimize that prompt engineering approach)