Tessera

Folgen

Michiel Bakker

@mbakker.bsky.social

LLMs and AI Safety. Assistant Prof @MIT. Research Scientist @GoogleDeepMind. CS PhD @MIT. He/him. www.miba.dev

3 Posts331 Follower 69 Folgt

Rachit Dubey @rachitdubey.bsky.social · 4mo ago

🚨New preprint and our results are rather concerning.. We find the "boiling frog" equivalent of AI use. Using large-scale RCTs, we provide *casual* evidence that AI assistance reduces persistence and hurts independent performance. And these effects emerge after just 10–15 minutes of AI use! 1/

🇺🇦 paul q. brussée @paulbrussee.bsky.social · last yr.

'AI wordt de fundering van economie, defensie, zorg, onderwijs en alles wat onze samenleving draaiende houdt. Als we die systemen niet zelf bouwen en begrijpen, raken we de regie over onze toekomst kwijt.’ @mbakker.bsky.social @jelleprins.com @nrc.nl #technologie www.nrc.nl/nieuws/2025/...

Opinie | Nationaal actieplan voor de AI-transitie is nu noodzakelijk

Technologie: Nederland dreigt controle te verliezen over de ontwikkeling van AI: een technologie die straks álles raakt – van innovatie tot defensie en dagelijks werk. Michiel Bakker en Jelle Prins he...

nrc.nl

Michiel Bakker @mbakker.bsky.social · last yr.

This was such a fun project to work on, with the incredible @taylor-sorensen.bsky.social 🥳

Taylor Sorensen@taylor-sorensen.bsky.social · last yr.

🤔🤖Most AI systems assume there’s just one right answer—but many tasks have reasonable disagreement. How can we better model human variation? 🌍✨ We propose modeling at the individual-level using open-ended, textual value profiles! 🗣️📝 arxiv.org/abs/2503.15484

Michiel Bakker @mbakker.bsky.social · last yr.

New preprint with @lukethorburn.com, Andrew Konya, Ariel Procaccia and others 🚨

Luke Thorburn@lukethorburn.com · last yr.

🔔 (new paper!) You might have heard of the "Habermas Machine", an AI-human pipeline that is really good at finding common ground between ideologically diverse groups, at least in lab settings. But can this kind of approach help in real world conflicts?

A diagram of the process used in the paper that this thread announces. The effort consisted of four dialogue cycles with peacebuilders: three uninational dialogues cycles conducted respectively with
Israeli Jews, Palestinians from the West Bank and Gaza, and Palestinian citizens of Israel, followed by a final joint dialogue involving
all three groups. Each dialogue cycle involved a process to find common ground among participants followed by a deliberation on the
results of that process. The common ground process started with a collective dialogue, then bridging-based ranking was used to
identify common ground ’bridging statements’ that were distilled into articulate ’collective statements’ via LLM and reviewed by
human experts before being shared back with participants for a final vote.

Beth Simone Noveck @bethnoveck.bsky.social · last yr.

Starting soon! Join us at 4 pm EST for a discussion The Peacemaking Machine with MIT & Deepmind's MH Tessler & @mbakker.bsky.social discussing how AI can transform democratic deliberation w/ @ bethnoveck @burnescenter.bsky.social @thegovlab.org rebootdemocracy.ai/events

bit.ly

David Rand @dgrand.bsky.social · 2y ago

Everyone going to SJDM this weekend, come to our special session on using #LLMs in #JDM research on Monday at 9:45am (location = Empire Complex)! w/ @kobihackenburg.bsky.social Hope Schroeder and myself - presentations from us but also hopefully lots of discussion/Q&A with all of you!

Niklas Henderson @14henderson.bsky.social · 2y ago

Cool preprint creating "supernotes": LLM-generated community notes that synthesise info from several existing notes.

An example post, along with three community notes proposed to provide additional context, and the Supernote produced by our framework (caption taken from preprint paper)

arxiv cs.SI@arxiv-cs-si.bsky.social · 2y ago

Supernotes: Driving Consensus in Crowd-Sourced Fact-Checking https://arxiv.org/abs/2411.06116