You may have seen the story that GPT-4 told a taskrabbit it was blind in order to solve a captcha. The team that conducted safety testing, ARC evaluations, has a blog post out now about how that test went down: evals.alignment.org/blog/2023-03-1… The big things that confused me about the original story were: why was GPT-4 asking a Taskrabbit for help instead of using a service like 2Captcha? Which steps here did GPT-4 do independently? The blog post was helpful for explaining those things.

People might think Matt is overstating this but I literally heard it from NYT reporters at the time. There was a top-down decision that tech could not be covered positively, even when there was a true, newsworthy and positive story. I'd never heard anything like it.

https://twitter.com/mattyglesias/status/1588190763413868553

For the record, Vox has never told me that my coverage of something must be 'hard-hitting' or must be critical or must be positive, and if they did, I would quit. Internal culture can happen in more subtle ways but the thing the NYT did is not normal.

I have now had someone *impersonate a virologist* (badly) in order to, I assume, learn what emails I'm sending to schedule interviews about the Covid preprint claiming to find a synthetic lab origin. Setting aside that this is insane behavior, happy to show my work. When I learned that this preprint had been released, I read it, talked to my editor about it, decided it was worth digging into deeply to understand the analysis and any flaws in it and deliver a full explainer on the claims in the paper and whether they stand up to scrutiny.

Okay, very important sociology question. Someone is talking and you know where they're going with the sentence. You gesture or interrupt or finish the sentence for them as a way of indicating they should move on to the next one: Answer this one ONLY IF YOU ARE CULTURALLY JEWISH (whatever that means to you). Same question as above.

Preview of tomorrow's Future Perfect newsletter: to get to the bottom of why Meta's new chatbot is so bad, I spent hours talking to it. It turned out to be, uh, a Genghis Khan apologist???

With effective altruism in the news absolutely everyone has been publishing their takes on the movement, and I keep thinking of things I want to say in response to all of them but don't have time. So let's try this: 1 like = 1 opinion on effective altruism and its critics. Global health interventions totally save peoples' lives and many of them won't be funded unless individual donors decide to donate money. There's lots of clever contrarian second-order stuff which just doesn't really touch this core fact about the world.

Meta's free, open BlenderBot is remarkably bad compared to GPT-3 or other top-of-the-line language models I've played with. I'm confused about why they released something so much worse than the state of the art. There's something about playing with a true SOTA language model that is terrifying and compelling. They're clever; they have the occasional beautiful turn of phrase; even when something they say is basically word salad, it takes a reread to *notice*.

I wrote for Vox about viral discovery, going out into nature to find lots of viruses and maybe find one that'll cause a future pandemic: vox.com/future-perfect…. When I first started writing about pandemics, I was pretty excited about viral discovery work. Intuitively, it seems like a clever idea. If we know what's coming, we can design vaccines and treatments before it even arrives, and not be caught off guard. Imagine if you could go back to 2017 with the Covid-19 genome and have everyone know exactly what to look out for!

The top-line result of this study of microlending is that it helps women start businesses and earn income. But unfortunately, the businesses they are starting are often multilevel marketing businesses, where you mostly make money by signing up other people. (1/7)

https://twitter.com/MDRC_News/status/1503794897161269248

I read the 18-month mid-experiment report for this a year and a half ago and was really concerned about the prevalence of MLMs among the 'success stories' of women starting jobs, but it was unclear from the mid-experiment data whether the microloans caused the women to join MLMs.

Related to this, there is some horrifying pathos in the story of the meltdown of entrepreneur Steve Kirsch, who founded the Covid-19 Early Treatment Fund and then alienated basically everyone in it.

https://twitter.com/KelseyTuoc/status/1458137345882918919

This Technology Review article technologyreview.com/2021/10/05/103… tells the story like so:

Another piece of advice for thinking about ivermectin: think of a specific question that is about physical reality NOT about social reality. ie, not "is ivermectin being unjustly censored on YouTube" or "are the horse goo people stupid" or "do people want us scared of Covid". It's not that those questions don't matter. But trying to reason about social reality without figuring out the actual ground-level truth is a great recipe to be miserable and confused and have no propositional beliefs that aren't reactive affiliation stuff.

If you're trying to make sense of the ivermectin evidence base, one thing I recommend is briefly researching a drug that works well against Covid, like dexamethasone. It's hard to guess from first principles how good an evidence base "should" look if the drug has sizable benefits. Sure, the ivm evidence base contains tons of terrible studies and fraud, but maybe that's just a common problem w/ science in a pandemic?

c19ivermectin.com is a gorgeous website with a beautiful UI that makes stunningly illiterate statistical claims about the case for ivermectin. Right now I'm mad about the claim (in the chart below) that the chance of these results if ivm didn't work is "a trillion to one". As far as I can tell, that's derived by multiplying the p-values of all the studies in the database. If this study finds only a 1% chance of getting results as strong or stronger if ivermectin had no effects, and this study finds only a 10% chance, then the chance of both is .01.

I've spent the last couple months trying to understand the degrowth movement, which wants to address the ecological impact of humans by ending economic growth: vox.com/future-perfect… I started out skeptical - I think the world is poor enough it needs growth for everyone to get their basic needs met, and I suspect that there are important things I haven't even thought of which a growing economic base will produce and improves peoples' lives with.

There's a @mattyglesias piece out today titled 'Here are some ways to change things', about how to make the world better without activism. But I feel like many of the best ways to change things are left off the list! slowboring.com/p/change-not-a… A lot of the most important social changes in the world, both good and bad, are downstream of technological ones - antibiotics, birth control, cars, vaccines, air pollution, nuclear weapons. A good way to change the world is to invent those, or figure out who might invent those.

This is framed in a maximally Twitter-dunkable way and everybody predictably hates it, but there's a divide among the dunkers that is actually interesting and gets at something important.

https://twitter.com/BW/status/1399538145201426442

Some people hate this because they think there should be many fewer humans in the world, because marginal people is actively bad. Some people hate this because they just think the idea that nonexistent people's absence can be good or bad is incoherent and kind of stupid.

I talked to lots of extremely reasonable experts about the intellectual property/vaccines thing and will write a summary of their responsible expert views but my irresponsible non-expert personal opinion is we should simply give Pfizer, Moderna, J&J, etc 100 billion dollars each. Don't waive the intellectual property, buy it! Don't worry too much about the right price, reward these companies for having done great work! Make everyone believe if they invent a pandemic-ending vaccine they'll get outlandishly rich!

The mess at Basecamp is reinforcing a consensus in tech that the reasonable thing to do is to have a workplace that encourages political + social discussion, and that workplaces which discourage that are going to struggle diversity and inclusion. I suspect it's more complicated. Obviously one reason employees want to talk politics at work is that they're directly impacted by social injustices and can't feel fully invested in a work environment where they're expected to set it aside.

A week ago, Oren Cass argued that sending child benefits to "nonworking households" was "welfare extremism" that "evinces an attitude that connection to the labor force isn't that important anyway". In response I shared a friend's story:

https://twitter.com/KelseyTuoc/status/1367946480414195714

Now my friend has written about this herself for Vox's First Person. vox.com/first-person/2…

More generally than the specific homelessness and stimulant checks thing, I strongly recommend to every single upper middle class person that you spend a couple of hours talking with working poor friends about their experience of interacting with institutions of all kinds. There's just this gulf there that is really important and very easy to underestimate until you've actually experienced it firsthand, and listening to the people who have is the next best thing.

My group house usually (including this last year) has some residents who've previously been homeless, and the process of getting them the benefits they're eligible for is always exhausting and infuriating. Without a fixed address, ID docs, + computer access, it's really hard to apply for the stuff you're eligible for. + after time and time again of applying and being rejected for random, poorly communicated, confusing reasons, lots of people just conclude the system isn't for them.