ta20211004_1
To address our favorite topic: while I use LLMs to assist on coding tasks a lot, I think they're very weak at this. Claude is much more likely to suggest or expand complex control flow logic on small data types than it is to recognize and implement an opportunity to encapsulate ideas in composable chunks. And I don't buy the idea that this doesn't matter since most code will be produced and consumed by LLMs. The LLMs of today are much more effective on code bases that have already been thoughtfully designed. So are humans. Why would that change?
Usually they have fun examples of problems these things can solve, but we don’t actually think they’re doing something fundamentally different from a greedy algorithm or trial and error optimization? And they’re slow as slow can be. So, not that exciting?
Yeah, you've got it. A mapping from words to vectors such that semantic similarity between words is reflected in mathematical similarity between vectors.
An idea of how you might train this thing: lets say the words "king" and "queen" are being embedded. In your training data there are lots of examples where "king" and "queen" are interchangeable, for example in the sentence "The ___ is dead, long live the ____", either word is appropriate in either slot, so each time we see an example like this we nudge "king" and "queen" a little closer together in some sense. However you also find phrases where they are not interchangeable, such as "The first born male will one day be ____". So when you see those examples you nudge "king" a little closer in some sense to other words which appropriately complete the sentence (which does not include "queen" in this case).
In this way, repeated over a giant training set with thousands of words, concepts like "male/female" and "royalty", "person/object" and tons of others end up getting reflected in the relationships between the vectors.
These vectors are then useful representations of words to ML models.
That's my 2c.
It's not showing me the 'reply' button on your comment.
That seems like fair criticism. It's not my area of expertise. I'm just here to post the source document because in light of what the study actually claims, the media reaction has been ridiculous.
I have no idea if any of the recommendations at the end of the study were actually put into action.
To me a charitable interpretation of the tone of this presentation, which focuses on highlighting negative outcomes rather than celebrating positive outcomes, was to identify areas for improvement. Again, interpreting charitably, this means the company is interested in identifying and reducing negative impacts.
Again, I don't know if they ever made any of these changes. I would like to see more transparency from the company.
Is there a better way to independently measure the mental health effect of a single variable on a person's life than just asking them?
https://s.wsj.net/public/resources/documents/teen-mental-hea...
p24. Teens are about twice as likely to state that Instagram improves their mental health than to state that it harms it.
p28. Teens who were already unsatisfied with their mental health tend to think Instagram harms their mental health more than teens who are satisfied with their mental health. Still, teens who are not satisfied with their mental health (in the USA) are more likely to think Instagram is a positive influence on mental health than a negative one. This is not true in the UK.
If you're the kind of person who hates when statistics are abused for a narrative, this should be something you dislike. The "damning internal documents" actually show that 80% of teens have a neutral or positive reaction to Instagram, and teens are two times more likely to have a positive reaction than a negative one.
Zuckerberg's comments about Instagram having an overall positive effect on teen mental health look a lot less like an obvious lie from this angle.
My personal opinion/guess: Instagram and social media does deliver harmful content to some people and can have negative effects on people's mental health. However, in the discussion we've been having this week, every possible positive effect has been discounted to 0 value. That's not really fair, doesn't reflect the actual results of the leaked studies, and certainly doesn't match my anecdotal experience using social media (including fbook, reddit, hn and insta).
That seems pretty reasonable. I am sure the Facebook recommendation algorithms can be improved. I would be in favour of more transparency here. Without transparency we simply can't know to what extent Facebook is trying to improve things, or if they even are.
To be clear, I do think recommendation algorithms are useful in content aggregator sites (which Facebook basically is). I just think we'd be better off if we could see what they're doing.
> And we had safe elections prior to 2020 as well. There were no insurrections in the current era here either...until January of 2021. I'd argue that we're just starting to see the 2nd and 3rd order effects of modern social media, and we can't necessarily just look to the past or to make conclusions about the present.
I don't thing this argument is falsifiable. It's totally possible. But I think we'd expect to see this in many countries where Facebook is available if it was a single cause of these kinds of outcomes. And so far that is just not happening.
The Rohingya genocide? While I agree that Facebook does not do enough to moderate content in non English communities, what percentage of the blame do you think Facebook actually deserves here? Recall this genocide was funded and executed by the military of a sovereign state.
It's _forcing_ increased polarization in politics? Be fair. We all have a choice whether or not to engage with content on Facebook, and whether we accept what we see there as truth. Facebook is available in every democracy in the world. In my home country of Canada, we just had a safe and fair election between a reasonable number of moderate parties. We had no insurrection. Many countries are able to do the same. There are a few exceptions.
You're using the term "facilitated" constantly in this post because really all Facebook has done in these cases was serve as a communication medium for blameworthy people with bad ideas. You seem to have no interested in the people who actually _did the things_.
Do you think anything you pointed out here would be difficult to do given any other existing communication medium?