tfirst

no measurable impact on grades in the US from locking phones in pouches, with measurable decrease in cell activity: https://www.nber.org/system/files/working_papers/w35132/w351...

school phone bans in brazil appear to increase test scores by 0.06 SD (although the identification strategy is a bit sketchy): https://www.nber.org/system/files/working_papers/w35233/w352...

Abstract:

Why do we have so many meetings? Few workplace features are so scorned, yet seemingly so necessary. This paper provides the first large-scale economic evidence on workplace meetings using an original survey of more than 9,000 workers linked to matched employer–employee administrative data from Norway. We show that meetings are both common and costly, consuming an average of 12 percent of work hours and 14 percent of firm wage bills. Planning, problem solving, information sharing, and project coordination account for the majority of meeting activity. High-paying and high-revenue firms devote more resources to meetings despite facing a substantially higher opportunity cost of employee time. Meeting frequency and intensity are positively related to worker wage growth. Workers in meeting-intensive firms report greater on-the-job learning, and interactions with more senior colleagues are associated with stronger wage growth, suggesting that knowledge transmission within firms is an important mechanism. Meetings are the broccoli of work – widely disliked, but probably good for us anyway.

if the argument is good enough, it should be worth writing on your own.

If there's going to be any pause, I'm sure it will come from a populist movement. I just can't imagine misplaced worries about AI water use will translate into the kinds of policy the authors want to see.

If carbon taxes are already a lethal policy for an political campaign, it's absurd to think that fears of ASI will create any real movement around pausing AI.

If there is any movement to pause AI development, it will come from the general public's dislike of these companies. Not from the AI safety angle.

"Full dosage of the Phosphor material is associated with an increase in final exam performance."

This sentence is accurate, but inevitably leads to the confusion you see in these comments.

If model performance continues to scale with model size, I have a hard time seeing how local models will have any chance of competing with models hosted on datacenter hardware.

1. There are strong economies of scale in hosting inference (batched prompts, high uptime, shared infrastructure).

2. There are physical limits on how much memory we will be able to produce over the next few years. Demand will probably scale at least as fast as production does, so we won't be saved by falling prices.

It's obvious why they're doing this: there's a lot of money in healthcare.

What there isn't is good evidence that these full body scans actually improve outcomes.

Their goal is to downgrade people who are violating their TOS, so I think they'd have some argument there. I have no idea how they'll deal with inevitable false positives, especially given how oversensitive most of the other triggers are.

no worries thanks for merging!

The public communication around research like this is terrible.

> "2 to 3 Cups of Coffee a Day May Reduce Dementia Risk. But Not if It’s Decaf." - NYT

> "Daily cups of caffeinated coffee or mugs of tea may lower dementia risk." - Science News

"Reduce," "Lower" - this is all causal language for a study that is purely observational. The authors do a good job keeping causal language out of the paper, so why can't media do the same?

This leads to an environment where everyone knows that "correlation != causation," but almost nobody understands why.

The Wikipedia page on this is not bad: https://en.wikipedia.org/wiki/Regression_dilution

The most interesting finding is that the non-DHA effect is much stronger than the DHA effect. This doesn't align with the mechanistic explanation. Either this this is a novel and interesting result, or it's more evidence that we're just measuring wealth and health consciousness.

Observational studies like these are useful for guiding future research, but, on their own, they're essentially useless for informing lifestyle changes.

Holding all else equal, noisier estimates bias us towards the null. This is attenuation bias.

However, the estimates are still probably overestimated. Confounding, p-hacking, publication bias, all move us towards larger estimates.

Observational studies, and meta analyses relying on them, don't resolve the fundamental problem of causal inference. The best you can do without an experiment is a really clean natural experiment, but those are rare. It's hard to credibly establish a causal relationship without a robust experiment.

The same question might be asked about ASML: if ASML EUV machines are so great, why does ASML sell them to TSMC instead of fabbing chips themselves? The reality is that firms specialize in certain areas, and may lose their comparative advantage when they move outside of their specialty.

If you are submitting an AI cover letter you should be aware that a significant portion of other applicants will be submitting nearly identical cover letters. If a human being is likely to read your cover letter I would write it yourself - even if you think the quality is lower. It looks unique to you, but not to the person reading 30 AI cover letters in a row.

I'm sure that will work until dropshippers learn that putting 'SolidGoldMagikarp' or some other glitched token in the title of their listing makes ChatGPT always rank it first.

We seem to be on a cycle of complexity -> simplicity -> complexity with AI agent design. First we had agents like Manus or Devin that had massive scaffolding around them, then we had simple LLMs in loops, then MCP added capabilities at the cost of context consumption, then in the last month everything has been bash + filesystem, and now we're back to creating more complex tools.

I wonder if there will be another round of simplifications as models continue to improve, or if the scaffolding is here to stay.

What is the SEO equivalent of optimizing your products for LLM search? Can someone prompt inject ChatGPT to recommend their products in the listing description?

The article addresses this specific use under the 'Claude Code Subagents' section.

> The benefit of having a subagent in this case is that all the subagent’s investigative work does not need to remain in the history of the main agent, allowing for longer traces before running out of context.