Man, even the brocast community appears to be reading @shannonvallor.bsky.social 's book.

Notable new NBER study on genAI and labor: despite widespread adoption of AI chatbots in Danish workplaces, their impact on earnings and hours worked is negligible. Productivity gains average just 3%, challenging the narrative of AI-driven labor market disruption

www.nber.org/system/files...

*sweating in ACT*

This is really cool work from @apartresearch.bsky.social - DarkBench to test different models for various dark patterns.

Would love to see more benchmarks like these!

www.apartresearch.com/post/uncover...

@lujain.bsky.social has published an excellent new paper exploring anthropomorphic behaviours in LLMs. Notable finding - majority of these behaviours occur after multi-turn interactions

arxiv.org/abs/2502.07077

Paris AI action summit kicking off today

A friend pointed out the side-by-side comparison of the changes to the Google RAI principles.

First two are the originals, 3/4/5 are the new ones.

Goodbye, redlines on weapons, surveillance, and human-rights abusing tech...

web.archive.org/web/20230804...

ai.google/responsibili...

Behold the power of an open-weight model release.

This was an interesting blog post about labor and economic impacts of agentic systems. I want to stop on this point about 'collecting job task data', as I'm seeing it as the basis for extrapolations that an o3 + will inevitably be as good at any kind of task

www.strangeloopcanon.com/p/what-would...

Something my grandmother wrote in 1973 after the Watergate scandal to her local newspaper (the Washington Post). Feels very relevant for this week.

From an excellent CJR report by @klaudia.bsky.social and @aisvarya17.bsky.social on how ChatGPT misrepresents sources.

This feature of LLMs in particular is the root cause of so many problems when adopted for tasks where accuracy matters.

www.cjr.org/tow_center/h...

H/t @bridainep.bsky.social

Oh no.

Sections on evidence collecting and indicators (9 and 10) look very strong to me. There are also excellent proposals for post-market monitoring, incident reporting, and whistleblowing. And requirements for executive and board level governance look great.

That said, this sections on the nature and drivers of risk are brilliant and get us closer clarity on how to define these abstract concepts. I would, however, question how likelihood and severity are determined - who gets to make that call? How will marginalised groups be involved?

The COP requires GPAI model providers to test and mitigate systemic risks - so how does the COP define systemic risk?

This is one issue I have - why exclude surveillance, privacy, human agency, quality of service harms, economic security and human rights? These should surely be included.

Love the intended use and AUP policy requirements. These look great.

First, great to see some stringent requirements in place for copyrighted data and energy disclosure. This does need to be made public if rightsholders are to use this (I also wonder what kinds of mechanisms rightholders need to query if their works are in a training data set?)