Kids are going to red team anything that's thrown at them. The limits of what ever prompts used to keep the model safe for kids are going to be thoroughly tested, and cleverly bypassed.

How much does this imply it would cost to scrape the web indiscriminately?
If it's only a few billion it may be within reach of large AI companies, granting them the exclusive privilege to web scraping.

What about scraping a subset of the internet, such a single website or small group of sites?
Would this still be cost prohibitive?

Its not confusing that these elements from super novae are called "metal" despite their non-traditionally metallic properties. Exploding suns are, arguably, the most metal thing to ever exist since the dawn of time. *riffs on guitar* (As confirmed by the telescope)

The issue presented in the article is not with the verbose definitions within legal texts. Precision is, indeed, important when defining laws. The issue is, however, with center-embedding, which involves nested definitions, often with multiple layers, in the legal text, making it difficult to parse and which can even create ambiguity.

An example of center-embedding:
"The officer who the witness whom the prosecutor questioned identified failed to appear in court."

This sentence may describe an officer who failed to appear in court, the officer may have been identified by a witness, and the prosecutor may have questioned that witness. There are many ways that center-embedded sentences could be re-written. Their clauses may even be independent and therefore separable into individual sentences.

Are the embedded structures the product of an afterthought?
What happened? "The officer failed to appear in court"
Which officer? "The officer, who the witness identified, failed to appear in court."
Which witness? "The officer, who the witness, whom the prosecutor questioned, identified, failed to appear in court."

Are they the only way to link the definition to the noun in the sentence?
Surely not, especially given that many legal documents "declare their variables", so to speak, in a prepended "definitions" section.

Are they meant to make legal documents more difficult to parse and understand by laymen?
This is the most likely answer unless it's just a weird, old habit of the legal profession. Either way, we can do better.

Has anyone forked it?

This is just a procedural language's subroutine. Like Basic's GOSUB, its a jump statement to a block that jumps back.

Likely pay-walled https://www.wsj.com/articles/c... This was reported on August 17th, why are the guardian, washington post, npr, and cnbc only reporting on this now?

Can't help but hear "Google Meat" when said out loud. Not too far fetched, they really have their hand in everything! Google Meetings I think may be a better name for the service.

When engineering physical structures calculations can be made to ensure they hold up to their use case in typical and statistically likely situations . When they are built if the design, materials, and craftsmanship adhere to those precalculated standards, the builders are not liable for acts of malice or acts of god. In software engineering only known potential exploits are calculable and forseeable. The software engineers should only be held liable for breaches and failures which are calculable and forseeable which do not involve innovative exploits, nor involving security policy circumvention where breaches of human trust are beyond industry security standards. Assuming they were comissioned to engineer such a thing to standards. Liability would then fall upon the employer. Ie. Fire IT if the network gets infected with the "iloveyou" virus in 2022, unless an employee disregarded policy or circumvented procedure in order to load the link in that email. And if the IT profressional was hired on Fiver to whip up the network security policy for a large enterprise overnight, then it falls on the enterprise.

Clearly we should be having 1 meeting instead of 5.