I think I’ll be using more Claude Sonnet 4.6 until Anthropic works out the kinks with Opus 4.7

It's extremely disappointing to see this from someone capable of much better thinking.

If we support a standard or vague agenda to slow or shape automation then most people will end up worse off than if we did nothing. At least half the Acemoglu ideas seem bad to me.

However, it will become politically untenable to “do nothing” and we need a positive agenda.

Very cool competition that the Financial Times held in 1914 to forecast the border changes in WWI.

‘Precisely seven years after the competition closed to entries, the FT announced: “No map sent in is even approximately accurate as compared with the final lines.”’

www.update.news/p/ai-just-is...

There's a periodic poll of academic economists run by UChicago, and it recently asked them about AI and economic growth. Lots of uncertainty, but this result is slightly more bullish than I would have expected. kentclarkcenter.org/surveys/ai-a...

If you care to explain the sense in which consent is violated, the harm done, the norms most uphold or should, I’m happy to listen.

I mean, I didn’t consent to you making your post above, but it’s obviously unreasonable for me to consent to any comment you might make about me in advance.

‎Gemini - direct access to Google AI Created with Gemini

I regret that you didn’t like the conversation and I regret if I’m missing something important about it.

I appreciate the irony of sharing this and I’m not trying to provoke you but
I asked Gemini if it was a good conversation based on the YouTube transcript g.co/gemini/share...

I’m probably guilty of tweeting as a vibesy sensible politics person.

What’s the right ratio of vibes to deep substance when posting?

Apologies for the belated reply…

I’ll just say I love podcasts like this one… much better than formal debates and much better than if the host does no pushback.

Of course, if the pushback is of sufficiently low quality I’d also prefer they said nothing. And I guess that’s your assessment of it.

What should/can I do about that?

You reached a lot of new people by going on that podcast.

It’s probably great for you and your ideas to do some interviews where the interviewer has their own views and you have some debate.

Wow, I can’t believe that is a take people have.

Much respect to those who make predictions and publicly score them. 🫡

A Rosetta Stone for AI Benchmarks YouTube video by Epoch AI

Individual AI benchmarks saturate too quickly to give us a long-run trend of AI progress.

We can solve this by "stitching" them together. This lets us forecast AI capabilities, quantify algorithmic improvements, and detect accelerations in AI progress.

www.youtube.com/watch?v=vwiD...

Could look at outcomes like wellbeing, ambition, self-understanding, political views?

- find a natural experiment to study students randomly exposed to a more/less humanities curriculum

- or perhaps actively creating art is more important so study that.

I’m sure the impact of exposure to the humanities is varied and multiple, so no experiment could capture it all. But some examples of interesting studies might be:

- Explicit randomized experiments comparing the impact of e.g. reading a novel vs. non-fiction book.

Noting that a lack of scientific evidence of benefits is not proof of a lack of benefits. Disturbing as it is to be skeptical of claimed benefits it would be more disturbing if we did not seek what scientific evidence can be had.

If Michael Baum says it inspired him that’s certainly worth noting.

If anyone has more scientific evidence about the benefits of the humanities I’m eager to see it.

Ending the kidney shortage is a no-brainer.

I just contacted my elected officials, Click here to take @action and do the same.

actionbutton.nationbuilder.com/share/SPK-QE...

We Let an Immigrant Live in Our Home You should too, especially if you have young children

The ideal that all parents, but especially mothers, spend max time with their kids leads to a stigma against au pairs.

More people should consider it.

www.alexnowrasteh.com/p/we-let-an-...

Trustworthy Data Visualization (Kieran Healy, Duke University) | posit::conf(2025) YouTube video by Posit PBC

I watched @kjhealy.co's keynote yesterday. It is well worth your time if you often wonder about what a good visualisation should do, and if you are curious about how they fit or don't in a truth-challenged environment. It is funny too. I also like the 3 words slides! #rstats
youtu.be/ZamPCbvBAgE?...

Also note that internet poker pros earned a steady income playing 12 tables simultaneously… they weren’t deeply analyzing their opponents, they steadily skim money off players that are playing far from game-theory-optimal.

podcast.clearerthinking.org - transcript BARRY: And where the features that are relevant are easily quantified. It's easy to decide what bets to make at a casino. It is and should be formulaic. It's not so easy to decide how to play poker, because it isn't just about probabilities. You have to understand the people you're playing against and try to, in effect, see through the backs of their cards. So that adds some complexity, but you can still calculate expected values. You know that staying in on some hands is bad, chasing the odds are so far against you, and on other hands, you almost certainly have the best hand at the table. You can calculate that just by knowing the odds of various cards in a deck. But that doesn't make you a winner at poker. That's a precondition. If you can't do that, you'll never be a winner at poker. If you can do that, you at least have a chance to be a winner at poker. Because especially as that kind of calculation gets more and more automatic, you can devote yourself to trying to read the other people at the table. And my

People often say that being good at poker requires reading tells and psychological insight, but this is mostly false.

Poker AIs destroy humans despite not seeing physical tells, and with a very limited model of individual players.

@clearerthinking.bsky.social @spencrgreenberg.bsky.social

Title: A Justification for 80% Power Abstract: Cohen’s heuristic reason for choosing 80% power (balancing Type I and TypeII errors) conveniently arrives at approximately the same number as an approachwhere one maximizes the marginal gain in power per standard error reduction. Ihave yet to see someone point this out, and this is interesting because it providesa non-arbitrary justification for 80% power.

a derivation of the result

I think this is kind of neat and I don't think anyone else has noticed it (I've looked and I can't find anyone who has) osf.io/preprints/so...

Maybe I should back off "justification" language, but it's at least a remarkable coincidence. I still think someone else *must* have noticed it...

Making normies care about something is a great skill/vocation.

Sadly this isn’t the skill/vocation that gets politicians elected, nor does it lead to writing popular op-eds.

Why no polls in BlueSky?

Wondering if anyone on this site wants to comment on this from @ctulocal1.bsky.social

The social sciences face a replicability crisis. A key determinant of replication success is statistical power. We assess the power of political science research by collating over 16,000 hypothesis tests from about 2,000 articles in 46 areas of the discipline. Under generous assumptions, we show that quantitative research in political science is greatly underpow- ered: the median analysis has about 10% power, and only about 1 in 10 tests have at least 80% power to detect the consensus effects reported in the literature. We also find substantial heterogeneity in tests across research areas, with some being characterized by high power but most having very low power. To contextualize our findings, we survey political methodologists to assess their expectations about power levels. Most methodologists greatly overestimate the statistical power of political science research.

The pretty draft is now online.

Link to paper (free): www.journals.uchicago.edu/doi/epdf/10....

Our replication package starts from the raw data and we put real work into making it readable & setting it up so people could poke at it, so please do explore it: dataverse.harvard.edu/dataset.xhtm...

Suggestion for authors: go back to forecasters now and ask them if they’d like to update their 2030 forecasts.