Talk:OpenAI

location

seems to be based in sfbay area.Mercurywoodrose (talk) 06:10, 12 December 2015 (UTC)

 Done. Thanks, Gap9551 (talk) 17:49, 12 December 2015 (UTC)

thanks i couldnt find a ref, you did.Mercurywoodrose (talk) 19:22, 12 December 2015 (UTC)

See also itis

many articles like this have too many "see also"s. we shouldnt just place every related article here. it should be lists that include this article, and article directly related that have not been able to fit well into the actual article. other institutes should NOT be listed, but should be in the body of the article as reliable sources themselves make the link or connection. its more a style point, too many see alsos means we are doing the research for the reader on whats interesting to them. we could put a see also for "luddites" for people who read this and say "hell no i hate this", or a link to brain development articles, history of computing, other think tanks in the bay area, cool AI projects like Watson, the Singularity, roger penrose who says we cant develop AI, and the movie AI. the list goes on and on.Mercurywoodrose (talk) 19:30, 12 December 2015 (UTC)

I agree. I added just one originally, about the topic Existential risk from advanced artificial intelligence that closely matches the aims of the company. There are several institutes with similar goals, but they can also be found in Category:Existential risk organizations. I removed Allen Institute for Artificial Intelligence for starters, as they seem not be specifically aiming to reduce risks associated with AI, just to develop advanced AI in general. Gap9551 (talk) 00:47, 13 December 2015 (UTC)

the field and the company

This article is supposed to be about the company, not AIin general, but I see in the article a great deal of general discussion about the future prospects for AI. It doesn't belong here. DGG ( talk ) 22:33, 3 February 2016 (UTC)

DGG I originally added the content because the mainstream media coverage of OpenAI talks in great detail about the donors' motivating beliefs about the future prospects for AI. I know you're a busy admin; maybe you didn't have time to read the sources? Do you want to discuss this and the "promotional" tag some more, or would it satisfy your concerns if I just ask WP:THIRDOPINION for an opinion to avoid taking up too much of your time? (Of course, if anyone else on this page shares DGG's concerns and would like to elaborate on possible concerns, feel free to chime in.) Rolf H Nelson (talk) 20:34, 6 February 2016 (UTC)

Re deletion of the reference to the sex-bot article

Regarding deletion of the reference to the article "Re: Sex-Bots -- Let Us Look Before We Leap" ( http://www.mdpi.com/2076-0752/7/2/15 ), several points are in order.

First, the journal in which the article appears is relatively new, but it is not obscure, having recently published, for example, two articles by tech industry heavyweights -- "Can Computers Create Art?" ( http://www.mdpi.com/2076-0752/7/2/18 ) and "Art in the Age of Machine Intelligence" ( http://www.mdpi.com/2076-0752/6/4/18 ) -- and which have enjoyed between them some 9,400 page views.

And yes, the article in question is an opinion piece -- but at this point in time, opinion is all we have; i.e., there is no one who can say with authority where AI is leading us, much less AI-enabled sex-bots! So if someone -- and that someone, BTW, is yours truly, although I don't think I've broken any of the WP:SELFCITE guidelines -- takes the time to express his concerns in a carefully thought-out and articulated piece, and if that piece is in turn given careful scrutiny before being published -- and yes, "Let Us Look Before We Leap" underwent a thorough peer review at Arts, even though published by them as "Opinion" -- what more could we expect from a source cited in Wikipedia regarding the quite critical and quite speculative subject of AI?

And finally, regarding the argument that this Wikipedia article should be about the company and not AI in general, the fact is that OpenAI has, by its very charter, captured the subject of the desirability of requiring that all AI code to which the public is subject be open source (just as, for example, we now require public disclosure of the details of all pharmaceuticals), and thus likewise the quite understandable goal of someone who thinks that this is the correct approach: he has taken the time to articulate his arguments and have them published in a reputable journal; and he now wishes in turn to share them with a larger Wikipedia audience via said article.

Comments, please! I am obviously aiming at a re-instatement of the deleted content, but can certainly be dissuaded therefrom. Synchronist (talk) 04:06, 1 August 2018 (UTC)

I'll suspend judgement then on whether it's obscure. I removed the content based on its not meeting WP:RS; I'm happy to solicit other opinions though. We can always ask WP:DRR/3O for a third opinion if nobody else on talk has any thoughts on the matter. Rolf H Nelson (talk) 17:19, 5 August 2018 (UTC)

The use is inappropriate: it has no mention or apparent relevance to OpenAI, we can't put things together like this per WP:SYNTH. Whether someone wants to share it is irrelevant, this is about things that are related to OpenAI. Not just the use of the source, but the commentary "...one juried commentator has asked..." is not encyclopedic. K.Bog 01:32, 28 August 2018 (UTC)

Solving Rubik’s Cube with a robot hand

Just want to attract attention to a new article published by OpenAI October 15th, 2019, about how they made a system that learned to solve Rubik's Cube all by itself, using only one hand (a Shadow Dexterious Hand). Maybe someone wants to add a mention of this to the article. There is a blog post, Solving Rubik’s Cube with a Robot Hand, and a scientific paper of the same name: Solving Rubik’s Cube with a Robot Hand. --Jhertel (talk) 17:38, 19 October 2019 (UTC)

Re: GPT3 "Pre-training GPT-3 required several thousand petaflop/s-days of compute, compared to tens of petaflop/s-days for the full GPT-2 model."

The following discussion is closed. Please do not modify it. Subsequent comments should be made on the appropriate discussion page. No further edits should be made to this discussion.

---

A while ago I put up a tag saying copy edit was needed, and it was reverted with a summary stating "[t]his is a correctly used technical term". I've never seen the term petaflop be used in that particular way before.

These are the two glaring typographical irregularities that have gotten me stumped:

• petaflop/s-days: I assume this was supposed to mean either petaflops/day or petaflop-days, but both nouns are in their plural forms.
  • I have no idea what a petaflop-day is supposed to be.
  • petaflops/day means billions of operations per second per day, which would suggest that pre-training either GPT would require a computer to perform a certain amount of PFLOPS on one day, and more PFLOPS than that on the next day, and so on.
• "of compute"; I can't decide if it should be corrected to "of computation" or "to compute", so I've left that part as-is.

-- MrPersonHumanGuy (talk) 02:04, 30 August 2020 (UTC)

Update: I see someone has added something in parentheses to clarify that several thousand petaflop/s-days are "a unit equivalent to approximately 10²⁰ neural net operations". A thousand PFLOPS would be 10¹⁸ floating point operations a second. Or, since there's 86,400 seconds in a day, a petaflop would mean 8.64 × 10¹⁹ floating-point operations on a daily basis.

After some digging through the edit history, I've found the edit that introduced the odd writing. Below is the prose it replaced, but I've modified the notes and citations to prevent them from adding a list to the bottom of the talkspace:

Lambda Labs estimated that GPT-3 would cost US$4.6M and take 355 GPU years to train using state-of-the-art^[b] GPU technology.^[64] Another source lists training costs of US$12M and memory requirement of 350GB on an undisclosed hardware configuration.^[65] Yet another estimate by Intento calculated that GPT-3 training would take 1 or 2 months^[c] and might consume 432 MWh (1,555 GJ) of electricity if run 24/7. ^[66]

If the overwriting sentence was supposed to specify how many PFLOPS and days "of compute [sic]" it took to pre-train GPT-3, then petaflops and days should be separate words with separate amounts. -- MrPersonHumanGuy (talk) 18:44, 30 August 2020 (UTC)

Update 2: To quote the source the clarifier cited;

A petaflop/s-day (pfs-day) consists of performing 10¹⁵ neural net operations per second for one day, or a total of about 10²⁰ operations.

That is from the second footnote, which is for this sentence:

The total amount of compute, in petaflop/s-days,^[2] used to train selected results that are relatively well known, used a lot of compute for their time, and gave enough information to estimate the compute used.

I think it's a bit funny how the author(s) of the OpenAI blog AI and Compute used the word compute in place of computation all over the place, as if the verb is also a common noun. -- MrPersonHumanGuy (talk) 12:26, 31 August 2020 (UTC)

The discussion above is closed. Please do not modify it. Subsequent comments should be made on the appropriate discussion page. No further edits should be made to this discussion.

I'm a little late to the game, so this response is for all those students, science and non-science. @MrPersonHumanGuy, there is a reason why your high school science or chemistry teacher emphasized and stressed always paying attention to the use of units in calculations.

It is quite common in technical, and particularly science fields, to have complex units (qualifiers): foot–pounds vs. newton–meters. That is a units that are other than simple: inch, gallon, ton, calorie. So your misapprehension is probably a lack of exposure.

Firstly the OpenAI terminology Petaflop/s-day(sic), and pfs-day(sic). The notation is misleading, the "/" (division) should have been a dash as in a complex unit, and s-day, the dash should have been a "/" divisor, i.e. sec/day.

Pardon the scientific notation. Petaflop is understood to be 1 executed computer op-code with qualifier 10^15 per second, and s-days(sic) would be 8.64 * 10^4 seconds/day (i.e. 60 sec/min * 60 min/hr * 24 hr/day = 86,400 sec/day ).

So 1 petaflop–s-day = (1 * 10^15 op/sec) * (8.64 * 10^4 sec/day). Which reduces to 8.64 * 10^19 op/day. Approximately 10^20 op/day. Q.E.D. WurmWoode^T 02:23, 12 January 2022 (UTC)

Removal of Controversy Section?

The section I added about controversy concerning OpenAI is completely warranted. I can assure you the creation of OpenAI LP has generated controversy. Again just last week with the prica announcement of GPT-3 the no longer open company structure of OpenAI is debated. Could you elaborate your reasons to remove the entire Controversy section? HaeB Diff here: https://en.wikipedia.org/w/index.php?title=OpenAI&diff=975914486&oldid=975824355

I'd like to add the announced pricing of GPT-3 to the controversy section as well but before doing that and getting it removed again. This needs resolving imho. I've seen it in many wikipedia articles that the controversial things about a subject are being repeated in that section so imho it doesn't warrant a complete errasure. Additionally it is true that they are still filing as a non-profit which is controversial given how non-transparent they have been lately. — Preceding unsigned comment added by Juliacubed (talk • contribs) 07:35, 4 September 2020 (UTC)

I agree that turning for profit generated a lot of controversy and deserves a section. See https://techcrunch.com/2019/03/11/openai-shifts-from-nonprofit-to-capped-profit-to-attract-capital/ and https://www.technologyreview.com/2020/02/17/844721/ai-openai-moonshot-elon-musk-sam-altman-greg-brockman-messy-secretive-reality/ and https://www.wired.com/story/dark-side-big-tech-funding-ai-research/ Yannn11 16:17, 11 July 2021 (UTC)

There clearly should be a Controversy section. The name "OpenAI" is deliberately misleading - it suggests that all development is open source, which is clearly not the case. Removal of the Controversy section seems to me to have been vandalism. But now the page is protected so that it's difficult to add it back. Sayitclearly (talk) 11:32, 6 December 2022 (UTC)

This page is no longer protected, so you should be able to add the section back in. I would, but as a leader of a competing org I have a clear COI.

Stellaathena (talk) 20:58, 1 April 2023 (UTC)

For Profit owned by Non Profit? What?

This is a general encyclopedia for everyone. We need to explain this corporate/Organisation construct and who can possibly profit or not profit from this. The current article is bound to confuse, rather than to clear things up. Can we please get someone who knows about this legal construct and explain it? Thanks so much. --91.64.59.134 (talk) 20:48, 24 October 2021 (UTC)

A Commons file used on this page or its Wikidata item has been nominated for deletion

The following Wikimedia Commons file used on this page or its Wikidata item has been nominated for deletion:

• DALL-E sample.png

Participate in the deletion discussion at the nomination page. —Community Tech bot (talk) 14:37, 6 May 2022 (UTC)

Unnecessary emphasis on Elon Musk?

I think this page refers to Elon Musk somewhat gratuitously. In particular, it seems unnecessary to feature a relatively large portrait of Musk next to a classification of the article as belonging to a series related to Musk, and linking to a page with his honors and achievements. Musk was one of several co-founding donors to the openai project, and no longer has any involvement with it. I think it would be appropriate to remove the photo of Musk, and link to his honors and achievements. Nickstudenski (talk) 19:34, 10 June 2022 (UTC)

I agree and removed "Elon Musk series." Yannn11 18:30, 11 June 2022 (UTC)

The thehindu-ref for Musk in the infobox is fishy (date from 2015 with a bunch of updates), topic is him founding a rival AI. The role of Musk for me stays unclear after reading the wiki here. He talks a lot, so he is cited a lot. Besides the money and a seat on the board, what did he actually do. He didn't have that much influence, since he left. Or did he leave because of M$'s investment? There should be more to learn about it here. MenkinAlRire 21:22, 30 April 2023 (UTC)

Greg Brockman page

surely time for a wiki article on him. why not? he's an important player in OpenAI and thus in AI development. https://openai.com/blog/authors/greg/ https://www.forbes.com/profile/greg-brockman/ https://csuitespotlight.com/2022/08/23/ivy-league-dropout-greg-brockman-is-leading-the-ai-revolution/ JCJC777 (talk) 13:10, 4 January 2023 (UTC)

Definitely — Preceding unsigned comment added by Skogkatt88 (talk • contribs) 01:47, 6 February 2023 (UTC)

Kellycoinguy (talk) 11:03, 8 February 2023 (UTC) has he asked not to have a page?? Surely he's notable enough by now.

He had a page, but it was deleted twice (1st nomination, 2nd nomination). The page is now protected from creation, so only administrators can create it. There is a draft version at Draft:Greg Brockman, which was submitted for review on 15 March 2023. It may take many months before the draft gets actually reviewed.  --Lambiam 17:19, 17 April 2023 (UTC)

Proposed merge of GPT-4 into OpenAI

Per WP:NPRODUCT, For product lines that are produced and/or marketed by the same company, avoid creating multiple stubs about each individual product (e.g., PU-36 Explosive Space Modulator, Q-36 Explosive Space Modulator, R-36 Explosive Space Modulator, etc.) especially if there is no realistic hope of expansion. The relationship between a continuous line of products should be discussed within a single article. The general scope of GPT (language model) is currently covered within the OpenAI article, which seems like an apt location to merge this to until such a time where we have something beyond routine business announcements that cover this. — Red-tailed hawk _(nest) 04:21, 4 February 2023 (UTC)

Perhaps GPT-4 (and 5, 6, 7, etc.) article should be merged into ChatGPT rather than the main OpenAI article? Alternatively, perhaps both should be merged --> OpenAI (unless this results in something too voluminous and unwieldly for mere humans to process)? Cheers, Cl3phact0 (talk) 11:34, 6 February 2023 (UTC)

PS: Perhaps we should ask it what it thinks?

PPS: The same logic may also apply to the GPT-3 article.

Oppose merge of ChatGPT and GPT-3 into OpenAI. They deserve their own articles as they are major products and widely covered by reliable sources on their own. Support redirect of GPT-4 to OpenAI until GPT-4 is released and more information becomes available. --Ita140188 (talk) 12:14, 7 February 2023 (UTC)

For clarity, my suggestion (in reply to the original proposed merge) is to merge all GPT versions (3, 4, n) into a single main ChatGPT or GPT article (not OpenAI). Perhaps my alternative suggestion made this unclear (I will strike it out).

I agree with your opposition to merging them into OpenAI article, and also think that the Generative models sub-section could be shortened by moving any detailed information to a main GPT article (per above). Cheers, Cl3phact0 (talk) 12:32, 7 February 2023 (UTC)

I agree with your opposition. Keep GPT-2, GPT-3, and ChatGPT as separate articles, but merge GPT-4 into the OpenAI or GPT-3 article for now. Yannn11 15:46, 7 February 2023 (UTC)

Merge, but not to OpenAI but to main GPT article. I'll also support merge of all GPT-2/3/4/n to one article, probably Generative pre-trained transformer. Artem.G (talk) 20:10, 7 February 2023 (UTC)

Thanks for making it succinct. Agree. Cl3phact0 (talk) 22:54, 7 February 2023 (UTC)

Oppose, --- Tbf69 userpage • usertalk 19:56, 8 February 2023 (UTC)

Oppose We shouldn't merge them, there is substantial media attention on both. Mixed Biscuit (talk) 23:09, 18 February 2023 (UTC)

Keep it as is. We have a whole 2,000 word article on Apple's electric car project, slated for release in 2026. I think GPT-4 is notable enough that we can afford a separate article, and avoid the confusion of picking an appropriate redirect target.

NPRODUCT is a weak argument, since there is obviously "realistic hope for expansion"; it's more pertinent for almost-identical products like 1990s Macintoshes or whatever the heck "R-36 Explosive Space Modulators" are, but not for this. CRYSTAL would be a better argument, but I think we can (and should) make an exception in this case, since no merge target would provide valuable context to this article (MERGEREASON #5). I argue it fails MERGEREASON #4 as well.

Also tentatively oppose merging all GPT-X articles into Generative pre-trained transformer: This'll sound weird, but I've noticed the WP:RS "narrative" regarding LLMs is pretty "mercurial", as in: they give a pretty fascinating mirror into how societal views keep shifting regarding the dangers & opportunities of LLMs. See for example the difference in the "mood" of coverage between LaMDA (not OpenAI) and ChatGPT, or even between GPT-3 and ChatGPT. I worry that if we merge them all, we might lose those nuances. The Reception sections are the kind of curiosity-catnip you usually only find in Featured articles, and a merge would risk losing them.DFlhb (talk) 22:32, 8 February 2023 (UTC)

Strong Support for merge per WP:CRYSTAL: "short articles that consist of only product announcement information and rumors are not appropriate". We can mention GPT4 briefly in both OpenAI and Generative Pre-trained Transformer --TocMan (talk) 03:05, 21 February 2023 (UTC)

Alternative merge as proposed by @Artem.G. This GPT article is much better for all things GPT except for extremely notable stuff like ChatGPT. InvadingInvader (userpage, talk) 23:27, 21 February 2023 (UTC)

What makes ChatGPT extremely notable as opposed to GPT-2 or GPT-3? Yannn¹¹ 15:26, 27 February 2023 (UTC)

Merge to GPT-3. GPT-4 is the successor to it, but it doesn't have enough notability demonstrated for its own distinct article. SWinxy (talk) 01:52, 2 March 2023 (UTC)

Strong support merge because it's simply not notable enough yet. There is literally no information from OpenAI about it. I do believe it will eventually warrant an article, but WP:NOTJUSTYET. – Popo Dameron talk 05:26, 10 March 2023 (UTC)

Comment For the record, Microsoft announced yesterday that GPT-4 would come next week. Pointless to merge when we'll need to unmerge almost immediately. DFlhb (talk) 05:32, 10 March 2023 (UTC)

Oppose per Dflhb. GPT-4 will 100% be notable after it comes out in under a week. There's unconfirmed rumors that Bing is using it. Snowmanonahoe (talk) 20:52, 10 March 2023 (UTC)

Oppose as per DFlhb and Ita140188 . Don’t Get Hope And Give Up — Preceding undated comment added 11:01, 11 March 2023 (UTC)

Oppose It should be notable enough in a few days. — Omegatron (talk) 18:24, 13 March 2023 (UTC)

As it is released today, this discussion has no more sense, so I'm closing it and removing merge templates from both articles. Artem.G (talk) 17:35, 14 March 2023 (UTC)