Honestly, why bother about the wider world when this provides an excellent filter for those both principled enough to quit Twitter and competent enough to join Mastodon? Sure, you'll have a high false negative rate (i.e., some great folks won't join), but the people who do will probably be more interesting to you than the average person.
I think all your criticisms of randomized controlled trials (apart from the ethical one) apply to your suggested approach, no?
RCTs are the gold standard for demonstrating efficacy precisely because of the controls. I agree that they're not going to give you enough info about potential side-effects - a thorough monitoring program is definitely necessary for that.
But without RCTs, showing that a drug works would be much harder. How would you weed out the snake oil?
RCTs were not required and were rarely performed prior to 1962, and yet that period from roughly 1940 to 1960 is still today widely known as "the Golden Age of drug development." How did they know, back then, that new drugs work? Simple: They have sufficiently powerful (in a statistical sense) effects in a patient population. Very powerful effects require a very small population -- this applies to ALL drugs that are curative -- whereas moderate effects would require a large population. Reasonably sized safety trials and extensive postmarketing surveillance are entirely sufficient.
> How would you weed out the snake oil?
The current paradigm is "better 10,000 people die of neglect than 1 person die of snake oil or quackery." I'm okay with a little bit of snake oil if it means that more drugs are being introduced more quickly, and I trust practitioners and patients to, more often than not, determine the therapeutic regimens that are right for them.
I see! My intuition was that it would be hard to evaluate efficacy without RCTs - I assumed low effect sizes, which I guess is not necessarily true for many drugs.
Agreed that there should be a faster way to trial drugs and get them to market. Though I've heard that the "slowness" of the FDA is actually overstated; not sure how true that claim is.
Those were also small molecule drugs which are relatively simple to produce compared to modern large molecule drugs (biologics such as modern insulin, mrna vaccines, humira, many cancer drugs, etc.)
You might be surprised that the modern processes for true biologicals like peptides and mRNA are actually simpler than the chemical processes for classic drugs. E.g. solid state peptide synthesis. Once you can make a 10 peptide one, you can make a 1000 unit one in an identical way with more iteration.
Compare that to establishing a bespoke chemical process for each drug, or messing with purification of natural biologicals, or fusing antibodies to protein or peptides to grow inside a whole cell.
They should be actually cheap. They aren't because new.
Humira is quite a bit harder to manufacture cleanly than a peptide or mRNA, and still harder than a plain solid state synthesized and folded protein. It requires a lot of post translational modification, which means cellular machinery, which means a whole bacterial cell you now have to grow, feed and remove from the product.
Not a dev, but I've often seen software that acknowledge their dependencies in About sections and on their websites. Is that particularly intrusive? Especially given that you could always use the GPL too if you want.
This goes beyond that. I didn't see that list of requirements; some places only say "attribution."
I would have no problem listing a library in the About dialog, but dictating the use of special icons and then all of those other requirements goes further than I've seen in any similar product.
As a non-business person, I imagine that folks would typically want some correctness guarantees, which seem near-impossible with today's LLMs. What kind of business use-cases do they enable? Probably a failure of imagination on my part, but I can't think of many tasks that can be done consistently well by AI that can't also be done with simpler software.
I’m both a business person and an engineer. I recently did a short workshop where a marketing guy took us through using GPT and Midjourney to iterate through new business ideas. In the 4 hours of the workshop, a bunch of mostly non developer people generated about 500 business ideas using GPT, took the top ideas, and turned them into business pitches. A couple of people in the room thought the idea their promoting came up with was good enough to run with.
This sort of stuff is fuzzy, even without AI. In many ways it’s about exploring the knowledge space rather than seeking specific answers. None of this requires correctness because there are no correct answers to begin with.
Even in computing, correctness is not always important. I recently asked GPT to produce a bunch of names from the Austin Powers universe in order to populate a demo database. I asked for them in JSON format and in 20 seconds I had what I needed. There is no wrong answer for that, either.
> The haiku at the end is not a haiku, technically
Yeah, haikus are 17 japanese mora, and so if it doesn't come from the mora region of japan it's not a true haiku, it's just a short poem.
> I imagine that folks would typically want some correctness guarantees
Us programmers really value correctness, but the real world doesn't actually care about it in quite a few cases. In the real world, most data comes from humans, and humans make mistakes all the time, so basically any data might have some amount of error. There's a high tolerance for a few minor mistakes here and there.
In the real world, tasks are often given to interns, and they make small mistakes all the time, but it mostly ends up being okay. If AI manages to do the same, well, that's a huge swath of tasks it can do.
> I can't think of many tasks that can be done consistently well by AI that can't also be done with simpler software
Here's the thing though: "simpler software" probably isn't actually cheaper, available, or usable. It's probably not simpler to operate since, you know, english is simpler than any DSL us programmers have come up with until now.
I've dabbled in comms work, and it's very useful for that. One thing that I despise doing is writing copy and LLMs are really good for that - writing dozens of social media posts is tedious and LLMs can rephrase things so that they sound less repetitive. I'm a good writer but I'm an amazing editor so having something to start with also solves the 'staring at a blank page/doc' problem.
Likewise, generating polite answers to stupid questions. Instead of worrying 'how do I tell an ego-driven exec/politician/candidate that what they want is stupid or not possible?' I can have an LLM spit out the stupid corporate drivel the C-Suite all want from their servants. It's also good for dumbing things down but not making it sound like you're dumbing things down and therefore is good for rewriting information for different audiences.
You don't (right now) use LLMs for generating facts but they have a great deal of utility in massaging language.
My intuition is that browser plugins based on some federated system (e.g. Mastodon, Matrix) could enable this. Why complicate browsers even more? (No expertise here, so I'm probably missing the big picture)
To simplify it for normal people. My family all use browser extensions to block ads/tracking and shore up security because I set it all up for them. My non-technical friends don't. They understand video game mods because adding mods is fun. But the idea of modding other kinds of programs makes no sense to them, because those are for getting work done, not for tinkering with.
So in order for it take off, it needs to be presented as just a native feature of the web. In people's heads, the logic should go like: "My email software lets me forward emails. My calendar software lets me invite people to calendar events. My contact software lets me merge duplicate contacts. My web browser lets me annotate web pages with other people. What a strange world it would be if things didn't work like that?"
Why does everyone find much stronger genetic effects than parental in adoption studies?
“Little intergenerational correlation in education was observed in the absence of genetic similarity between parent and child—that is, among adoptees.”
“By examining parent-offspring resemblance in a sample of offspring that are among the oldest of any adoption study of IQ to date, we have effectively tested for the presence of parenting effects that would have persisted for more than a decade after the conclusion of the typical rearing period. No such persistence is found to occur in our unique sample.”
Did you actually read your own references? From the abstract of the study of adopted Korean Americans:
"I find large effects on adoptees' education,
income, and health from assignment to parents with more education and from
assignment to smaller families. Parental education and family size are significantly more correlated with adoptee outcomes than are parental income or neighborhood characteristics."
I'm not sure, that's not my area of expertise! I think it's pretty well-replicated that IQ is more predicted by genetics than any other factor; however, my understanding is that variation in IQ within an ethnic population is much greater than variation between populations. If a population mean is significantly different from others (which appears to be the case with Ashkenazi Jews and Asians), I'd expect the effect to be primarily environmental, not genetic.
I don't know enough about the inheritance of intelligence to be sure of this at all, though.
