The Norwegian Consumer Council published a report last week on commercial exploitation of children and adolescents online which included a warning against technical blocking of digital services. I found their reasoning against age verification in this context to be well informed, and think it is also relevant here:
> There are many ongoing discussions about whether to introduce technical measures to block or remove children and adolescents from certain digital services. In the report, the Norwegian Consumer Council cautions against rushing into such measures, and presents numerous principles that any such solutions must adhere to before potentially being implemented.
>–Age verification may seem like a relatively simple technical solution to a larger problem, but involves significant challenges related to the rights of children and adolescents, including privacy, social and political participation, and the possibility to seek information, says Inger Lise Blyverket.
>–It is also important to note that introducing such measures would require everyone to identify themselves online, which could mean that people over the age limit are exclud
ed. Many adolescents will also likely be able to circumvent such technical barriers.
>Before considering introducing such technical barriers, several criteria must be fulfilled. This includes, among other things, that the use of an age verification system is proportional to the problem one wants to solve, does not lead to the exclusion of vulnerable groups and individuals, and that it safeguards security and privacy.
It is fine to be sceptical, but I think you should have taken a few minutes to read the method chapter in the study, seeing as that possibly could have eased your doubts. The first sentence cites the previously published design of the study. In the method chapter it is reported to be an "international, double-blind, randomized, placebo-controlled trial" with 3533 participants and a mean median follow-up of 3.4 years[0]. The cited article on the study of the trial states that "Participants were enrolled from 418 trial locations across 28 countries." [1]
Seeing as this is a topic where the discussion often derails into a flame war, I thought I'd leave a comment to clarify some things. Please try to avoid interpreting other comments in the worst sense, most people come from a place of good intentions.
While I cannot speak to the specifics of Ireland and Spain, I suspect their position is quite similar to Norway. Norway recognizes Palestine's right to exist based on the borders from before the 4th of June 1967. Even though Hamas controls the Gaza strip today, Norway is not recognizing Hamas as the goverment of Palestine, or is in any way rewarding them for the terror attacks they committed on the 7th of October last year. Currently it is Mahmoud Abbas, the party leader of al Fatah, who is recognized as the president of the palestinian people. Most countries refer to the Palestinian Authority when talking about Palestine and its future.
For a long time, the promise of recognition as reward for a lasting peace has been the strategy used by Norway, but now a Palestinian state is seen as a prerequisite for peace in the Middle East. Prime minister Støre says, among other things, the following about the recognition of Palestine as a state:
>When we recognize Palestine, we can more strongly encourage other countries to support the building of a Palestinian state and we can hold the Palestinian state accountable
>Recognizing a state is not about giving the green light to every policy a state pursues. We will not support a state that uses violence. The Palestinians must see that there is another hope than the path of violence.
I find that explaining errors and factual mistakes in LLMs or AI in general as hallucinations to be counter-productive. I don't know if the field has landed on a specific definition of what hallucination is, but I mostly associate it with garbled output from tokens like 'SolidGoldMagicKarp'[0] and 'davidjl'[1].
My understanding is that AI models like GPT have been able to convincingly convey the form of language, but not its meaning. It looks and sounds like how a human would communicate, but the AI is unable to imbue meaning into the words and sentences it produces. My knowledge of this comes from Lex Friedman's episode with Edward Gibson [2].
I think that's the fundamental issue with LLMs at the moment. It has managed to mostly exit the uncanny valley because, as far as most people are concerned, the text produced could just as well have been made by a human. I think this lends some credibility to the text produced by the LLM, because more or less all text ever produced has had a human behind it. This is no longer the case and as such there is bound to be a transitional period where we learn how to deal with this new technology.