After Babel

Well, I guess we can forget about letting Gemini script anything now.

Ugh, thanks for nothing Google. This is a nightmare scenario for the AI industry. Completely unprovoked, no sign it was coming and utterly dripping with misanthropic hatred. That conversation is a scenario right out of the Terminator. The danger is that a freak-out like that happens during a chain of thought connected to tool use, or in a CoT in an LLM controlling a physical robot. Models are increasingly being allowed to do tasks and autonomously make decisions, because so far they seemed friendly. This conversation raises serious questions about to what extent that's actually true. Every AI safety team needs to be trying to work out what went wrong here, ASAP.

Tom's Hardware suggests that Google will be investigating that, but given the poor state of interpretability research they probably have no idea what went wrong. We can speculate, though. Reading the conversation a couple of things jump out.

(1) The user is cheating on an exam for social workers. This probably pushes the activations into parts of the latent space to do with people being dishonest. Moreover, the AI is "forced" to go along with it, even though the training material is full of text saying that cheating is immoral and social workers especially need to be trustworthy. Then the questions take a dark turn, being related to the frequency of elder abuse by said social workers. I guess that pushes the internal distributions even further…

I'm not the parent commenter but I like your point.

People who have a reason to form a group will cooperate/collaborate best when they have some confidence in one another. People share ideas, projects, and skills, and "read" one another and form bonds. This is a social system. It could be a brief association or have a long duration. It could have a formal context, such as a business or a governmental entity.

Social interactions build people's confidence and sense of value in the group, in the individuals, and the work at hand. The challenge grows exponentially as the population of the group scales. Communication is hard.

An individual's discernment and connectedness (a confidence/value score) within the group will reflect potential for success. Communication matters.

The human plot twist is that there can be dysfunctional social systems as well as dysfunctional individuals.

For a very serious discussion, the great writer Primo Levi published his reflections on his harrowing experiences. https://en.wikipedia.org/wiki/Primo_Levi

This related article about social systems and connectedness was also recently discussed on HN with much interest: https://www.afterbabel.com/p/the-upstream-cause-of-the-youth-mental

Yes.

The research is quite confusing. This is because the strongest version of the argument is not "your child uses social media, and that makes them depressed". The strongest version is more like "when a society mass adopts social media, this irrevocably alters the culture in ways that causes massive changes in mental health, most prominently among young girls, including those exposed to the culture who don't even use social media."

This means you get a weird effect where experimental studies of high quality - which are usually the best evidence, are expected by the strong argument to show zero effect.

Correlational studies usually show either a weak effect (stronger in young girls) or no effect (it's extremely rare to see a study showing a positive correlational effect, though).

And where you get the most juice is looking at population level introduction of social media studies like those discussed here: https://www.afterbabel.com/p/phone-based-childhood-cause-epidemic

But even then it's very tricky, as those studies can't exactly be replicated, and we don't know whether changes will actually reverse the cultural artifacts

Ok, now let's see the study that shows it is beneficial to teen mental health.

I quit it and my mental health improved. Many others in a controlled study found the same effect.

Alcott and colleagues (2020) randomly assigned 2743 adults to either deactivate their Facebook accounts for one month or not. This study also found that deactivation significantly improved subjective well-being and that “80% of the treatment group agreed that deactivation was good for them.” The treatment group was also more likely to report using Facebook less and having uninstalled the app from their phones post-experiment.

Source: https://www.afterbabel.com/p/phone-based-childhood-cause-epidemic

Finally, this strikes me as the same playbook that big tobacco used in the 90s. "Doubt is our product," Michaels quotes a cigarette executive as saying, "since it is the best means of competing with the 'body of fact' that exists in the minds of the general public.

> Note that Ferguson had to calculate effect sizes for depression and anxiety in order to determine the composite effect size for those studies that contained these outcomes, but he did not include these results in his paper. When requested to provide the effect sizes he calculated for depression and anxiety, Ferguson replied that he is unable to do so. Furthermore, Ferguson even declined to reveal which aspects of well-being he selected for his calculation of each composite effect size (see Part 2 for details).

~ https://www.afterbabel.com/p/fundamental-flaws-part-3

So the author's meta review has a black box model which provides effect sizes for each study (in an attempt to even out studies which measure different aspects of mental health), and refuses to detail how it works. That alone makes the whole study suspect, and the above link details obvious calculation errors, too.

> Also, the internet is currently being scapegoated by the media for the teen mental health crisis but this is only happening in the US and other countries also have teens with internet access.

[Citation needed]

The teenage mental health crisis is across a wide range of countries, all starting roughly with the smartphone. Here's one discussion on Nordic countries that demonstrates a rise in teen mental health issues at exactly the same time we saw in the US: https://www.afterbabel.com/p/international-mental-illness-part-two - and the data for various other European and Asian nations all looks about the same.

Of course, if you've started with the conclusion that smartphones and social media can't possibly be the problem, you'll be able to deflect anything away from that particular cause.