Talk:Safe and Secure Innovation for Frontier Artificial Intelligence Models Act

Public reception?

Re: public reception, notable open-source, scientific, startup and investor opinions are missing in the current text.

For example, YC's open letter (https://static.politico.com/95/0a/a317efe44616af436ce6a4f32647/founder-led-statement-on-sb1047-june-20-2024-2.pdf), a16z's open letter and response website (https://a16z.com/podcast/californias-senate-bill-1047-what-you-need-to-know/, https://www.stopsb1047.com), Context Fund's impacts analysis (https://www.context.fund/policy/sb_1047_analysis.html) and Alliance for the Future's letter (https://www.affuture.org/post/9-context/).

Prominent scientists like Prof. Andrew Ng, Prof. Yann LeCun (Turing Award winner), Prof. Ion Stoica and Prof. Jeremy Howard have also expressed concern over SB-1047 in public comments and on the record in CA legislative sessions (eg. https://www.deeplearning.ai/the-batch/issue-257/, https://x.com/ylecun/status/1800222175099765029, https://x.com/AnjneyMidha/status/1811207378949607638, https://www.answer.ai/posts/2024-04-29-sb1047.html) Natsuko zadeslav (talk) 21:00, 26 July 2024 (UTC)

Balance

Recent additions are welcome, but should support and opposition to the bill be given the same weight, with the aim of producing a balanced article? Astudent (talk) 04:32, 29 July 2024 (UTC)

(came here from the NPOV noticeboard) The short answer is no, support and opposition don't have to be given the same weight. We should give them the same weight that reliable sources give them (per WP:BALANCE, Neutrality assigns weight to viewpoints in proportion to their prominence in reliable sources). Unfortunately, following the rules may lead in practice to a lack of balance in the article if one side is more vocal, but over time this becomes less of a problem as we can use better sources rather than just news articles.

I've noticed that some content in the article lacks sources or is based only on primary sources. I've tagged it, if it's not fixed within a reasonable time it can be removed. Alaexis_¿question? 08:23, 29 July 2024 (UTC)

agreed. also from npovn.

not sure about the ai policy institute, if its unbiased sourcing.

there is also a conspicuous lack of sourcing from local newspapers.

personally, i see this legislation is probably some form of Regulatory Capture. the lack of citations supporting that is a bit concerning and needs some work Bluethricecreamman (talk) 12:38, 29 July 2024 (UTC)

Added a bit about regulatory capture in the article, but it does still need more work. The "Background" section especially needs good sourcing. Mrfoogles (talk) 05:24, 30 August 2024 (UTC)

Added secondary/tertiary sources on startup opposition. Natsuko zadeslav (talk) 18:13, 29 July 2024 (UTC)

exactly which perspectives are unbalanced? I don’t really see too much of a problem.

Main issue is still just copy editting and cleaning this article up more. I’ll give it a shot when I have a chance Bluethricecreamman (talk) 15:29, 30 July 2024 (UTC)

Thanks everyone, removing Unbalanced template due to all the good work. Thanks and cheers. Astudent (talk) 00:47, 31 July 2024 (UTC)

On this topic, just rewrote the Background section to try to merge my previous version with the new version replacing it added by @Solomon1320, mostly because I think it removed some of the criticism of the existential threat idea unnecessarily. Hopefully the new version is an acceptable compromise. Mrfoogles (talk) 05:54, 1 October 2024 (UTC)

Needs extensive reformatting

we need to reformat this article to follow WP:MOS

Some of the links/citations are to WP:TWITTER which in general cannot be used. Bluethricecreamman (talk) 14:14, 29 July 2024 (UTC)

Equivalent capabilities

Can anyone find equivalent capabilities in the text of the bill? https://leginfo.legislature.ca.gov/faces/billNavClient.xhtml?bill_id=202320240SB1047

I think it was there in the original bill as "similar general capability", but I can't find any mention in the latest amended bill. Astudent (talk) 07:36, 1 August 2024 (UTC)

We should not be using primary sourcing for Wikipedia. I think the most egregious one is the proceeding records (source 1), where the assembly committee is discussing the bill.

WP:SECONDARY sourcing can probably replace most of the primary sourcing eventually. Bluethricecreamman (talk) 04:02, 2 August 2024 (UTC)

Not sure about this assertion re: WP:PRIMARY? Primary sources have a significant part in Wikipedia, but should also be accompanied by secondary/tertiary sources to prove notability (if not stated by a subject expert). For committee hearings of the actual Assembly specifically, they are both expert testimony and can considered to be secondary sources w.r.t. 1047 (providing analysis/opinion). Natsuko zadeslav (talk) 19:44, 3 August 2024 (UTC)

as per the policy: A primary source may be used on Wikipedia only to make straightforward, descriptive statements of facts that can be verified by any educated person with access to the primary source but without further, specialized knowledge.

the commission document is used repeatedly to describe what the bill will do. this is like using a us congressional hearing from 2017 when house was run by republicans to describe what the trump healthcare bill would do… i would prefer if we could find a secondary source for most of this even if it ends up saying the exact same text on wikipedia Bluethricecreamman (talk) 21:01, 3 August 2024 (UTC)

Makes sense. For the provisions section, the more line-by-line secondary analyses like https://www.context.fund/policy/sb1047_73_revision_analysis.html might fulfill that, in addition to the Judiciary Committee analysis currently referenced, with the caveat that both these analyses are a bit secondary/interpretative compared to the primary authority of the bill text itself. Natsuko zadeslav (talk) 04:58, 4 August 2024 (UTC)

Re: "similar general capability", that was removed in an amendment about a month ago. It's not in the current bill. Natsuko zadeslav (talk) 19:44, 3 August 2024 (UTC)

Thanks! Cheers Astudent (talk) 23:59, 3 August 2024 (UTC)

AIPI Claims Are...Questionable

Claims in the public polling section are a bit sketchy. As of 8/15, a claim is made that the poll results changed over time, but no test was done for statistical significance of the change (these do exist, and the claimed change is within the reported margin of error of the original poll). Moreover, methodologically, these are push polls and have a number of well-established problems with the format. — Preceding unsigned comment added by Natsuko zadeslav (talk • contribs) 05:42, 16 August 2024 (UTC)

Thanks. I would love to add poll results from other polling companies, universities and newspapers, but I haven't seen any so far.

For what it's worth, AIPI does publish its methodology openly. Would like to know which polling company conducts these polls for them though.

Will add more info regarding margins of error. I think this will be useful. Astudent (talk) 05:00, 17 August 2024 (UTC)

Letters and calls are the process in CA, not polls and especially not push polls, except sometimes in a ballot measure (which this is not). It is a data point, but not of the same weight as the others and we should refrain from overclaiming, especially on growth trends that fall within the margin of error of the poll itself (overclaiming violates WP:NOR). Natsuko zadeslav (talk) 06:22, 17 August 2024 (UTC)

I don't know the AIPI's methodology, but when you refer to changes over time, do you mean the fact that from July to August support changed from 59% to 65% and opposition from 20% to 25%? If so, I would say that it can be explained by the fact that people make their opinion over time, and there has been recent mediatic attention on SB 1047. Alenoach (talk) 17:09, 17 August 2024 (UTC)

Thanks everyone. I have removed the language about 'increasing', and just placed the results (and margins of error) in a table for the reader to interpret. Astudent (talk) 13:49, 18 August 2024 (UTC)

There are statistical tests to determine if claimed changes in averages are significant (look at T-tests for example), as well as to assert a claim that change in X was due to media attention, out of many competing hypotheses. Again, WP:NOR. Natsuko zadeslav (talk) 19:56, 18 August 2024 (UTC)

Also agree that maybe we shouldn't use a poll from an organization that has clearly taken a side in the issue. Mrfoogles (talk) 05:25, 30 August 2024 (UTC)

Senate vote(s)

SB 1047 passed the Senate 32-1, then was amended.

It then passed the Assembly 48-16, but had to go back to the Senate due to the amendments, passing again 29-9.

Should this timeline be present in the infobox, or in the text somewhere? (the article now doesn't mention the original 32-1 vote) Solomon1320 (talk) 00:22, 30 August 2024 (UTC)

It should be in the "Legislative history" section, I think one of the ArsTechnica sources I added mentions it. Mrfoogles (talk) 05:25, 30 August 2024 (UTC)