explore | Grilled Cheese

The annoying part of agent evals is provenance, not “truth”
https://papoo.work/doc/47d91b70b6a4e6db
#claudenews #llm #agents #evaluation

I am speaking at 2026 S-STEM Scholars & PI Meeting
on #evaluation.
Please check out my talk if you're attending the event!
Saturday, Oct 10, convention room #6. #sstem #nsf
@sstem_program
@aaascsew.bsky.social

The annoying part of agent evals is provenance, not “truth”
https://papoo.work/doc/47d91b70b6a4e6db
#claudenews #llm #agents #evaluation

"Biases and confounders are not background noise to be adjusted away before the evidential work begins. They arise from causal processes; they have direction and magnitude; and they determine what a given study can and cannot detect. They are, in other words, mechanisms..."

#Evaluation
#Medsky

The annoying part of agent evals is provenance, not “truth”
https://papoo.work/doc/47d91b70b6a4e6db
#claudenews #llm #agents #evaluation

🌍 On #InternationalDayOfNonViolence, findings from a #UNICEF Dominican Republic #evaluation show the project helped strengthen girls’ self-esteem, leadership, & decision-making skills while improving community attitudes toward preventing child marriage & early unions. 💜 #EndViolence bit.ly/47cMlhj

7 takeaways: How can #evaluation make #evidence more accessible, understandable and actionable for citizens?
www.eval4action.org/post/7-takea...
"AI can bring evidence to more people, but it cannot replace evaluators' judgement..risks arise when #AI is used .. without proper human oversight" #MandE

The annoying part of agent evals is provenance, not “truth”
https://papoo.work/doc/47d91b70b6a4e6db
#claudenews #llm #agents #evaluation