balint stewart

Data science, policy, microbes

[balint stewart @BalintStewart]

@93vintagejones @wself definitely has an overpowering LLM smell, the too-neat aphorism "proceduralized itself to the point where procedure no longer mediates trust but replaces it" is particularly egregious. Odd because Will Self has been a critic of AI generated prose - Is this some kind of test?

[balint stewart @BalintStewart]

@So8res We still watch humans playing chess though, pretty much exclusively

[balint stewart @BalintStewart]

@JuliaEMcCoy Kinda true, but thinking properly/building effective world models requires some memorisation. There's no way around it. @daisychristo has written extensively and V persuasively about this

[balint stewart @BalintStewart]

I used to be skeptical about ensemble methods in ML (because of the danger of the models' amplifying one another's error, rather than cancelling them out) but the wisdom of the crowds does seem to work beautifully in most instances.

[balint stewart @BalintStewart]

A small but underappreciated way to reduce LLM hallucination is to feed outputs from one model into a different model for review (this works a lot better than getting the LLM to mark its own homework).

[balint stewart @BalintStewart]

@mattwridley To read is to commune with another mind. You can't do that with video and audio.

[balint stewart @BalintStewart]

For exactly the reasons Voss explains here: labelling allows the listener to learn much more quickly (Perel has only 1 counselling session per episode) AND allows her to build rapport, even (maybe esp) if the client corrects her labeling of their emotions

[balint stewart @BalintStewart]

Faacinating methodological parallels between high stakes hostage negotiation and therapy - emotional Jedi @EstherPerel uses 'labeling' (describing someone's feelings out loud to them) in practically every episode of 'Where Should We Begin?' https://t.co/OxE3NLuFfB

[balint stewart @BalintStewart]

Interesting takedown of various explanations for massive decline in adolescent mental health, i'd liked the idea that a decline in 'locus of control' developed through unsupervised play was important but the dates don't line up to be sufficient w/o tech https://t.co/hfWZgkyF4S

[ADR UK @adr_uk]

We're piloting workshops to highlight the linked administrative datasets available for public good research across the UK, with @UKStatsAuth💡

Interested in hosting an event like this for your research organisation? Find out more and get in touch ⬇️
https://t.co/724dSc9ZWW

[Anthony Finkelstein @profserious]

I was unaware of this work on Reproducible Analytical Pipelines https://t.co/By3cmRBHZp which is very impressive

[ADR UK @adr_uk]

Are you interested in a PhD studentship on quantitative research using linked #AdministrativeData? ADR UK are happy to announce we will be funding @ESRC Doctoral Training Partnerships later this year! Check out the upcoming opportunities: https://t.co/KjZjlDUaAW #DataDrivenChange

[balint stewart @BalintStewart]

@shayda_k @adr_uk @StatsRegulation @QUBelfast @DrMaryLouCowan @KohlrabiCons Keep up the good work Shayda!

[ADR UK @adr_uk]

Our new 'What is research-ready data?' roundtable report details a discussion exploring what we mean by research-ready #data, how we can address the challenges, and the importance of transparency
👉 https://t.co/UotkaxXA8Z
#DataDrivenChange #AdministrativeData

[balint stewart @BalintStewart]

@eLife These look like the opposite of Bigger's dormant 'persister' cells

[IOE Library @IOELibrary]

This guest post by @BalintStewart explains how @adr_uk is revolutionising access to government-held data to enable research on the lives of children and young people for better policy decisions #linkeddata https://t.co/Fpo58lWOtJ

[ADR UK @adr_uk]

“It is inspiring to see momentum grow for better, broader, safer use of health data across so many sectors.”

We welcome the recommendation from @bengoldacre’s review that #TrustedResearchEnvironments become the norm for #DataAccess for research🔢

https://t.co/vDcDSEeMLv (1/5)

[Daniel Martin @DanGMartin1]

The Feynman Lectures on Physics is so beautiful in its new website form. What better light reading?

https://t.co/OdsdRwIzsz

[Moshe Vardi @vardi]

English :-)(

[Ewan Birney @ewanbirney]

Gather around again for a tweet-thread about "models", their types, what they are (and aren't) and how to use and comment and/or critique them.