Talk:Google Brain

This edit request by an editor with a conflict of interest has now been answered.

Started Hi, I represent AlphaChip and am requesting edits to address what we believe are inaccuracies in the article, specifically regarding contested claims about our research. Requested changes: We request removal or modification of the following passage in the article: "However, this claim is contested because claimed results, especially fast chip design, were not properly supported by specific empirical data and found inconsistent with subsequent published research.[55][56][57][58] The paper does not report run times of prior and proposed methods on specific inputs, lacks head-to-head comparisons to sufficiently advanced implementations of prior methods, and is difficult to replicability due to proprietary training and test data." Rationale: The research code and data have been made open source and publicly available on GitHub at https://github.com/google-research/circuit_training, directly addressing the concerns about proprietary data and replicability. We also request modification of: "At least one initially favorable commentary has been retracted upon further review,[59] and the paper is under investigation by Nature.[60]" Rationale: This information is now outdated. Nature completed their investigation and published an addendum in September 2024 upholding AlphaChip's claims.^[1] Suggested replacement text: "At least one initially favorable commentary was retracted upon further review.[59] Following an 18-month investigation, Nature published an addendum in September 2024 that upheld the paper's claims.

Ag77777 (talk) 20:26, 13 October 2025 (UTC)

Hi @Ag77777, do you have a link to Nature's addendum? Thanks Encoded  Talk 💬 20:03, 28 December 2025 (UTC)

Hi, here is a link to the Nature Addendum: https://www.nature.com/articles/s41586-024-08032-5 Ag77777 (talk) 20:09, 28 December 2025 (UTC)

I assume the addendum closes the investigation? Is there any way to get confirmation on this? If so, it would be factually incorrect to say that it's still under investigation so that should be updated. According to the original article on Nature's website it says a correction was issued, I think it would not be appropriate to remove the section on the wiki page but I would support updating it to say a correction and/or addendum was released. Thanks, Encoded  Talk 💬 19:34, 29 December 2025 (UTC)

The request also says that "Nature completed their investigation and published an addendum in September 2024 upholding AlphaChip's claims" which makes it sound like Nature wrote it, I'd suggest rewording since it appears to have been written by Google's team. Would "Nature completed their investigation and published an addendum written by members of the Google Brain team in September 2024 upholding AlphaChip's claims" be appropriate? Encoded  Talk 💬 00:06, 30 December 2025 (UTC)

I do not believe the current state of affairs warrants any change in the article. There is a well known and completely uncontroversial method for showing a new placement method is a breakthrough, or at least an improvement. You simply run it on well-known and public designs, and show it gets better results, or it runs faster, or some other improvement. Here, for example, is a tiny fraction of table V in the paper RePlAce: Advancing Solution Quality and Routability Validation in Global Placement^[1] It straightforwardly shows their method gives better results, although often with worse runtimes:

More information Benchmark, Best Known ...

Table Caption (Optional)

Benchmark	Best Known		RePlace
	Wire Length	CPU time	Wire length	CPU time
Adaptec-1	74.20	13.13	73.01	14.18
BigBlue-3	264.48	33.15	255.07	68.62
NewBlue-7	978.07	246.00	335.19	889.18

Close

In particular, releasing the source code does not show the new method is an improvement, nor do results on unpublished designs. The current status as of late 2024 is summarized here:"Updates Spark Uproar". Until and unless the AlphaChip method is directly and transparently compared with other existing methods, and shown to be better, then claims of the method's superiority will continue to be controversial. LouScheffer (talk) 21:28, 28 December 2025 (UTC)

Thank you LouScheffer and Encoded for the thoughtful feedback.

Encoded, you're correct that the addendum was authored by the original research team, not Nature independently. The research was approved, published, and supported by Nature, however – a rarity. This was also reviewed by independent committee. I appreciate that clarification. The addendum represents Nature's completion of their investigation and their decision to publish a correction addressing the concerns raised, rather than a retraction.

LouScheffer, I believe the Nature addendum is still relevant to include because:

1) The article currently states "the paper is under investigation by Nature"[60] - this is factually outdated since Nature concluded their investigation and published an addendum in September 2024.

2) The addendum represents Nature's editorial decision after reviewing the concerns raised. While it doesn't settle the technical debate about comparative performance, it does resolve the question of whether Nature found sufficient issues to retract or substantively correct the paper.

I'd propose updating the existing sentence to reflect what actually occurred:

Change: "At least one initially favorable commentary has been retracted upon further review,[59] and the paper is under investigation by Nature.[60]"

To: "At least one initially favorable commentary has been retracted upon further review,[59] and following an investigation, Nature published an addendum in September 2024.^[2]"

This simply updates factually incorrect information (the investigation is no longer ongoing) without claiming the addendum resolves the technical concerns you've outlined. Ag77777 (talk) 17:01, 6 January 2026 (UTC) Ag77777 (talk) 17:01, 6 January 2026 (UTC)

Note for reviewers: an additional request is in progress at Talk:Criticism of Google that may be relevant. Encoded  Talk 💬 19:48, 29 December 2025 (UTC)

Ag77777 I don't see anything that specifically says the investigation has ended. Equating the addendum with closure seeps into WP:OR. Also, your revision would remove the fact that there was an investigation, which I believe is wrong, for the historical record. Nonetheless, I put the info about the addendum (with some extra context) at the end of the section, so it's chronological. I also changed some earlier tenses. Let me know what you think, so we can close this. STEMinfo (talk) 10:01, 7 January 2026 (UTC)

Clarified that the update was published, but did not resolve the controversy. This involved moving some post-update references down from the previous paragraph, and explaining the issues involved and why the update does not settle the dispute. All straight from the references cited. LouScheffer (talk) 13:06, 7 January 2026 (UTC)

@STEMinfo, your tense-related edits are welcome. However, clicking through the addendum URL suggests that the date and the title in the reference may have been hallucinated, so I corrected those. The addendum seems to have some 20 authors, not 3, so I added "et al". You may also recall that Nature needs to be italicized per MOS:NAMESANDTITLES.

Since you apparently added the reference without checking if it matches the source, I encourage you to check the "fully reproduce" narrative you added (see my clarification request in the article). The addendum is behind the paywall, so I couldn't easily check it beyond the metadata.

Something else doesn't add up here: a 2022 open-access "Authors' correction" (with the same author list) announced that they had open-sourced a software repository to reproduce the methods described in their 2021 paper.^[3] So, it's unclear if the new 2024 source verifies additional distinct claims (per WP:V) for code availability and reproducibility - perhaps, you can include a relevant quote in the reference. Shouldn't the 2022 open-access source be cited for reproducibility claims?

SFsticated 06:51, 8 January 2026 (UTC) SFsticated 06:51, 8 January 2026 (UTC)

FYI the addendum is available via WP:TWL Czarking0 (talk) 04:27, 9 January 2026 (UTC)

FYI this COI user is pushing a similar line at Talk:Criticism of Google I found similar results. Czarking0 (talk) 04:20, 9 January 2026 (UTC)

I'm going to close this request. Several others have reviewed and made changes to update the article accordingly. The current request is outdated. Any additional changes will require a new, properly sourced request. STEMinfo (talk) 02:05, 27 January 2026 (UTC)