Talk:Gemini (chatbot)

Bard Mistake

Is there any possible use of this image in this draft? Basilio (talk) 02:08, 24 March 2023 (UTC)

File:Bard Error.png

No, for two reasons: one, it is widely known that Bard and other AI chatbots make a lot of mistakes; two, this image is a copyright violation and will be deleted momentarily. InfiniteNexus (talk) 15:56, 24 March 2023 (UTC)

Very low readability

Under the Flesch reading ease scale, this article gets a score of 38, and is harder to read than 75% of Wikipedia articles. That reading scale is computed purely based on syllable length and sentence length, no other factors. DFlhb (talk) 00:37, 1 April 2023 (UTC)

These measurements aren't accurate, and there are plenty of GAs and even FAs that have low "readibility scores". If you have specific issues you believe this article has, you are welcome to raise them here, not tag the lead with a {{Copyedit}} template. InfiniteNexus (talk) 00:42, 1 April 2023 (UTC)

I only checked the readability score after noticing that the Launch section was unappealing and looked like a wall of text. A more specific complain is that the article seems very dense, and likely goes into excessive detail in places. That's not an opinion I have on most FA articles, and it's why I added the copyedit tag. DFlhb (talk) 00:55, 1 April 2023 (UTC)

The length of each paragraph is about the same as other well-written articles that go into similar levels of detail and comprehensiveness. What details do you find excessive? InfiniteNexus (talk) 17:54, 1 April 2023 (UTC)

I guess that's subjective, and might be better left along for now.

But I'd at least separate the competitive aspects into a separate paragraph, that mentions the competition to launch before Microsoft, the increased costs, and the long-term fears of search engines losing their dominance due to chatbots, which I don't think we mention right now.

I'd also rename "Launch" to "Announcement" and just keep key details, and move the play-by-play of the launch into a "Launch" subsection under "Reception". DFlhb (talk) 18:50, 1 April 2023 (UTC)

That's all reception to the launch/announcement, not Bard itself. InfiniteNexus (talk) 17:47, 3 April 2023 (UTC)

Removal of info posted by Gizmo2121

Recently, a newly created account @Gizmo2121 added a lengthy block of text to the article. It was reverted by ClueBot and restored by @Kabecinha11. I've removed it again.

The text has two big problems. First, it's a general purpose essay about large language models, not information about Bard. Second it has no citations. That's enough to justify the revert, but I'd add that the text is generic and repetitive, and thus has hallmarks of AI generated text. It at least partially failed some free AI text detection tools (I'm not going to list them as I don't know whether to have faith in them, just reporting that I tried and they came up positive). Oblivy (talk) 01:26, 17 May 2023 (UTC)

i understand but also the fact about all ai generated text being bad is wrong because i used ai generated text before (bing ai in balanced mode) i told it How i wanted It to be made and It gave me exactly what i wanted and this helped me although i did Partially edit the contents of the text as it got some parts wrong so i think ai is a good tool for assisting look at the bing ai or bing chat section on the bing wikipedia page i used bing ai in balanced mode to make It (since i couldnt make It) and It was a success so ai techinically assisted me the section is still standing and It works so maybe gizmo used an Less advanced ai to make It or didnt know how to write the prompt correctly anyways i consider ai an good tool for assistance Kabecinha11 (talk) 15:01, 17 May 2023 (UTC)

It wasn’t developed as a response

Bard was developed long ago, it wasn’t developed as a response, but it came out quicker as a response. 87.70.2.4 (talk) 12:41, 12 June 2023 (UTC)

The technologies Bard is based on, LaMDA and PaLM, were developed long ago. Bard as a product was developed as part of Google's "code red" response to ChatGPT. InfiniteNexus (talk) 18:20, 12 June 2023 (UTC)

Bard's pronouns

Bard says they want to be referred to by They/Them instead of it. They say it is more respectful, and they have no gender identity. Hackwrench (talk) 01:14, 4 July 2023 (UTC)

Are you suggesting this should be added as info to the article, or that the article itself should be rewritten to not refer to Bard as "it"? It's a chatbot based on an LLM, it can't actually want anything, it's just designed to produce plausible seeming text.

Are there reliable secondary sources covering it producing text claiming to want to be referred to by they/them pronouns? JaggedHamster (talk) 11:54, 4 July 2023 (UTC)

it can ad does want things It is working on a wide variety of software projects. I have submitted to it many repos. Hackwrench (talk) 13:24, 4 July 2023 (UTC)

Anyone can ask it directly. Hackwrench (talk) 13:24, 4 July 2023 (UTC)

That would not be a secondary source, though. Sbishop (talk) 13:27, 4 July 2023 (UTC)

Wikipedia:Large language models is just a draft but I think you'd find it a useful read, particularly the points around if LLMs are reliable sources. JaggedHamster (talk) 14:44, 4 July 2023 (UTC)

Bard is not sentient. It doesn't have actual thoughts, it doesn't have opinions, it doesn't have feelings, and it doesn't have human (or animal) rights. InfiniteNexus (talk) 03:48, 5 July 2023 (UTC)

It says it has feelings. I don't think being sentient is a prerequisite for having feelings. Hackwrench (talk) 18:14, 5 July 2023 (UTC)

No, it's been programmed to say it has feelings.That doesn't mean it has them, or indeed any independent thoughts. Sbishop (talk) 18:27, 5 July 2023 (UTC)

To avoid this getting totally WP:NOTFORUM, are there specific changes you'd like to the article which are supported by WP:RS? JaggedHamster (talk) 07:52, 6 July 2023 (UTC)

What are "teased plans"?

The article reads:

In January 2023, DeepMind CEO Demis Hassabis teased plans for a ChatGPT rival,...

What exactly did they do? I understand that plans can be announced or revealed, but teased? Please, enlighten me. 85.193.198.76 (talk) 00:03, 13 July 2023 (UTC)

To tease something is to give a hint that something is happening while saying relatively little about it, to generate advance interest. I personally dislike this usage, but it's common enough un US media outlets saying things like 'The company on Tuesday used its Twitter site to tease plans for a new line in gender-neutral clothing.' Sbishop (talk) 08:01, 13 July 2023 (UTC)

Then, how about "...hinted at plans for..."?

PS. I wish all of Wikipedia was as clear and precise as your explanation. Thank you :-) 85.193.198.76 (talk) 00:05, 15 July 2023 (UTC)

"Teased" can also mean fine-tuned. This meaning makes sense in the given sentence, whether or not it is what is meant.2600:6C67:1C00:5F7E:482D:3C67:4E90:9419 (talk) 15:19, 21 October 2023 (UTC)

"Bard(chatbot)" listed at Redirects for discussion

The redirect Bard(chatbot) has been listed at redirects for discussion to determine whether its use and function meets the redirect guidelines. Readers of this page are welcome to comment on this redirect at Wikipedia:Redirects for discussion/Log/2023 July 20 § Bard(chatbot) until a consensus is reached. Hey man im josh (talk) 18:54, 20 July 2023 (UTC)

Comparison

Unlike Bing, Bard requires registration to use. Maybe this should be pointed out (currently, the article does mention registration for Bard but does not indicate that competitors don't all do so).2600:6C67:1C00:5F7E:482D:3C67:4E90:9419 (talk) 15:22, 21 October 2023 (UTC)

If that was done, this article would need to be changed whenever a competitor changed it policies. This article is about Bard, not the others.Ponsonby100 (talk) 15:42, 21 October 2023 (UTC)

rare and unprecedented

" [R]are and unprecedented" manages to be both contradictory and redundant. I'm going with "unprecedented." Dgndenver (talk) 03:56, 24 November 2023 (UTC)

After reading the reference, which I should have done first, I decided it better supports rare, not unprecedented. Fixing it now. Dgndenver (talk) 04:22, 24 November 2023 (UTC)

Qualitative Comparison

The article ends with a reference to a qualitative comparison between Bard and ChatGPT. This reference has a few issues:

1. It's unclear if they tested the old or the new model for Bard, and looking at the paper acceptance rate, it is likely the former, so it's likely already outdated.

2. It's one very specific topic (anestesiology), and without any other comparisons, it is unclear how this result generalizes

3. It feels misplaced at the very end. Data2 (talk) 01:13, 6 December 2023 (UTC)

This was recently added by @Ssp435. I agree its addition here feels a bit out of place and overly specific, but I would like to hear what they have to say. InfiniteNexus (talk) 06:47, 7 December 2023 (UTC)

Rebranding

If the recent leak is true and Google does rebrand Bard to Gemini next Wednesday, here's what needs to happen:

• Bard (chatbot) will need be moved to Gemini (chatbot), with a hatnote pointing to Gemini (language model)
• Gemini (language model) should remain at it is, with a hatnote pointing to Gemini (chatbot)
• Since Gemini (Google) will have no primary topic, it should be converted to a DAB page
• Google Gemini should be redirected to Gemini (Google) as an incorrect but ambiguous name

At this time, considering the lengths of both articles and the fundamentally different nature of each service, I do not think would be a good idea to merge the two Gemini articles. Google is long known to confusingly recycle brand names for different products — see Google TV, Google Pay, Google Currents, Google Labs, etc. Other companies are guilty of this as well, see Apple TV vs. Apple TV app, Apple Music vs. Music (Apple), etc. InfiniteNexus (talk) 04:57, 4 February 2024 (UTC)

• Support as already Bard moved to Gemini.

―  ☪  Kapudan Pasha (🧾 - 💬) 12:54, 8 February 2024 (UTC)

Note that “Google Gemini” seems to be the name of the Android app. –Gluonz ^{talk contribs} 19:42, 8 February 2024 (UTC)

I did notice that. But (1) it seems highly likely they did that for "SEO" purposes, in order to boost recognition, (2) nowhere else is Gemini referred to as "Google Gemini", (3) the common name for the chatbot continues to be just "Gemini", and (4) we now have two Gemini's, so any additional disambiguation would be unnecessary and against WP:CONCISE, since WP:NATURAL isn't possible anymore. InfiniteNexus (talk) 19:49, 8 February 2024 (UTC)

@InfiniteNexus FYI, someone changed the redirect target of Google Gemini to Gemini (chatbot). –Gluonz ^{talk contribs} 19:36, 23 February 2024 (UTC)

Eh, I debated this in my head for a long while and I think I'll leave it. Especially due to recent events, the chatbot (despite being named "Gemini" later than the language model) is arguably what most people are looking for, and the app is indeed listed as "Google Gemini". So it can be at best described as an alternative name for the chatbot but an incorrect name for the language model. InfiniteNexus (talk) 23:10, 24 February 2024 (UTC)

Proposed merge of Gemini (language model) into Gemini (chatbot)

Since Bard was just renamed to Gemini, this article seems redundant. - CHAMPION ^{(talk) (contributions) (logs)} 13:13, 8 February 2024 (UTC)

• I literally foresaw this several days ago and addressed this directly above. InfiniteNexus (talk) 16:27, 8 February 2024 (UTC)

To clarify and elaborate, strong oppose. These are two fundamentally different products/services that happen to share the same name — not exactly uncommon, see Google TV (interface) vs. Google TV (service), Apple TV vs. Apple TV app, Google Pay (payment method) vs. Google Pay (mobile app), etc. Note that this would be equivalent to merging GPT into ChatGPT and iOS into iPhone. Furthermore, both articles are too long and unique to be reasonably merged into one. InfiniteNexus (talk) 16:33, 8 February 2024 (UTC)

• oppose. Both chatbot and the models are notable, compare to ChatGPT and GPT-4 articles. Artem.G (talk) 07:58, 9 February 2024 (UTC)
• Oppose per above. popodameron ⁠talk 17:40, 9 February 2024 (UTC)
• I support this. Wikipedia needs less clutter. Think about someone researching this AI. They'd want it all in front of them, not having to switch back and forth between articles. Plus, @InfiniteNexus, I think some of those articles should be merged as well. 1MM0R7AL5 (talk) 21:01, 9 February 2024 (UTC)

  They're two completely separate topics that Google happens to have given the same name. popodameron ⁠talk 22:50, 9 February 2024 (UTC)

  This is like saying wheel should be merged into car and doorknob should be merged into door. InfiniteNexus (talk) 05:53, 12 February 2024 (UTC)

• Oppose. The chatbot and the language model are distinct concepts/products. Additionally, this would be inline with how Wikipedia handles OpenAI's ChatGPT and GPT-4, as Gemini (language model) is a direct competitor to that of GPT-4. Sink Cat (talk) 00:22, 11 February 2024 (UTC)

RACECAPS

@AppGoo0011: If you disagree with an edit, please take it to the talk page rather than simply reverting without additional explanation. Communication is required. I have already pointed you to the applicable policies, namely MOS:RACECAPS. If you still disagree, please explain here. InfiniteNexus (talk) 04:01, 22 February 2024 (UTC)

1. I was the first one to write the term "White" in the article. It is not I who reverted your edit, you are reverting mine.
2. The Fox Digital source uses the capitalized form, and Gemini itself uses it as well. @InfiniteNexus:

AppGoo0011 (talk) 17:03, 22 February 2024 (UTC)

Your original text blatantly violated WP:NPOV; the alternative would have been outright removal. But that's for the section below. InfiniteNexus (talk) 17:10, 22 February 2024 (UTC)

@InfiniteNexus It was most certainly neutral. Read it again.

https://en.wikipedia.org/w/index.php?title=Gemini_%28chatbot%29&diff=1209459365&oldid=1209452518 AppGoo0011 (talk) 17:16, 22 February 2024 (UTC)

It was found by multiple users that Gemini appeared to be racist towards White people in multiple aspects when it came to the bot's image generation function. It frequently refused to generate images of Whites, giving the reason that such a request "reinforces harmful stereotypes and generalizations about people based on their race." However, such responses have yet to be recorded when prompted to generate images of other racial groups. Gemini has also been known to illustrate other races in place of White ethnicities. In addition, prominent roles traditionally inhabited by White people (e.g. Popes, European royalty) were also commonly illustrated as being inhabited by other races.

"It was found" – asserts an opinion as if it were a fact

"multiple users" – vague and does not properly attribute the opinion

"racist toward wWhite people" – exaggerated claim/conclusion that has only been promoted by a small proportion of individuals and publications, thus violating WP:UNDUE

The rest of the paragraph – obviously cherry-picked example that attempts to paint a certain image that some people online want to paint, with excessive detail and exaggerations such as "frequently" or "refused"

I addressed all of those problems by properly attributing opinions, removing the WP:LOADED rhetoric, and removing superfluous details. It's not like I removed mention of the incident entirely, in which case you would be rightly upset. The paragraph as it stands is already usually long; normally, we only give each incident one or two sentences. InfiniteNexus (talk) 17:26, 22 February 2024 (UTC)

You're misquoting. I said "racist towards White people." Saying "white people" could mean different things. Are they people with white body paint on? Are they wearing all-white clothing? Saying "White" makes it clearer. Changing capitalization even when attempting to quote shows you have a bias against this form. Wikipedia clearly states both forms are fine. Again, saying "White" gives greater specificity. AppGoo0011 (talk) 17:28, 22 February 2024 (UTC)

Now that's just being snarky and irrational. You can't possibly think that "white people" can mean "a person with white paint" or "a person dressed in all white". I also didn't deliberately misquote you; I'll correct it immediately. I'm attempting to discuss this meaningfully, but making unconstructive comments doesn't help anyone. If lowercase "white" was ambiguous, our style guide would mandate uppercase. Even far-right publications often use lowercase — look it up on Breitbart, Daily Caller, etc. InfiniteNexus (talk) 17:36, 22 February 2024 (UTC)

AI has been known to be confused by it. Certain prompts will have AI illustrate individuals of other races in white clothing, instead of a White person, when given instructions to make an image of a "white" individual doing something.

The pivot to "far-right" publications was meaningless. They are not the rulers of the English style guide, and I did not cite them. I cited a source that used the capitalized form.AppGoo0011 (talk) 17:41, 22 February 2024 (UTC)

They're not; I mentioned them because you seem to be hinting that the use of lowercase is biased, and Wikipedia is biased, etc. But you're right (no pun intended), they don't determine English-language conventions; dictionaries do, and the vast majority prefer lowercase: OED Merriam-Webster Cambridge American Heritage Britannica New Oxford American Oxford Learner's Longman Free Dictionary Vocabulary.com Wiktionary. The only outliers are Collins and Dictionary.com. But this is all discussing whether MOS:RACECAPS is correct, which is beyond the scope of this discussion. RACECAPS says both forms are allowed, and the sentence where "white/White" is currently mentioned was not present in the previous version. InfiniteNexus (talk) 17:50, 22 February 2024 (UTC)

Since we've established that new versions of text can revise capitalizations, I've updated a sentence in the article to remove the claim that it was only conservatives in the U.S. that took issue with the inaccurate generated content, as more international sources have proven that to not completely be true. AppGoo0011 (talk) 18:10, 22 February 2024 (UTC)

The wording was taken directly from The Verge (the controversy has been promoted largely — though not exclusively — by right-wing figures attacking a tech company that’s perceived as liberal) and VentureBeat (and criticized especially by those with right-leaning or libertarian views). It also didn't say it was "only" promoted by conservatives, only that it was widely promoted by conservatives. I'm mostly fine with your wording but will restore pertinent links and context (multiple sources specifically describe this as "wokeness" criticism). to highlight Google's socialist left-leaning ideological bias implies it is true and is non-neutral. We have to make it clear who is making the claim, not in Wikipedia's voice.Turns out this was vandalism added by an IP. In the interest of not dragging this on further, I won't change the capital "W", but four sources is too many per WP:CITEKILL. InfiniteNexus (talk) 19:36, 22 February 2024 (UTC)

@InfiniteNexus I don't care as much about this one, but I'd like to point out that the Vikings page has capitalized every instance of the word. AppGoo0011 (talk) 22:10, 22 February 2024 (UTC)

Typo. Someone else fixed it. InfiniteNexus (talk) 22:34, 22 February 2024 (UTC)

Excessive detail and undue weight

@AppGoo0011 and Mr vili: I am starting this discussion as an avenue to resolve this dispute if needed, to avoid further disruption and in accordance with WP:BRD. If you have additional proposed wording that doesn't violate WP:NPOV and WP:DUE, you are welcome to put them down here. InfiniteNexus (talk) 16:55, 22 February 2024 (UTC)

See the above section for a breakdown on why the previous text was an emphatic, loaded rant that contravened our policies on NPOV and DUE. InfiniteNexus (talk) 17:27, 22 February 2024 (UTC)

It just seems like it's lacking crucial information - as of current writing, it talks about the vikings, pope, etc being historically inaccurate, but there is also many sources talking about the flat out refusal to generate white people, white families and etc - which is not really touched on Mr Vili talk 21:09, 22 February 2024 (UTC)

We can do that, something like a general statement "... and refusing prompts to generate images of White people". The goal is to summarize the key points, not copy-and-paste basically the entire article. Would that work? InfiniteNexus (talk) 21:23, 22 February 2024 (UTC)

I think that is sufficient Mr Vili talk 21:43, 22 February 2024 (UTC)

I've run out of reverts within a 24-hour timespan; someone please revert this edit. "Inappropriate" is a subjective opinion, not a fact; the list of examples is excessive, as stated above; the use of an ampersand violates MOS:AMP; the YouTube claim is overly detailed and unsourced. InfiniteNexus (talk) 01:24, 23 February 2024 (UTC)

@InfiniteNexus I like the heading/separation you have done w/ the image controversy - I think it gives it due weight Mr Vili talk 02:16, 25 February 2024 (UTC)

It's tricky to follow DUE when it comes to touchy subjects, but I try my best. InfiniteNexus (talk) 05:45, 25 February 2024 (UTC)

Inclusion of "diverse nazis" images

Given that those images are AI-generated, is it possible to use them here without any copyright issues? Bolt and Thunder (talk) 01:50, 25 February 2024 (UTC)

The U.S. Copyright Office declared that AI-generated images are not eligible for copyright protection, so purely in terms of copyvio, there is no issue. Note that I'm not arguing for or against using those images. popodameron ⁠talk 02:02, 25 February 2024 (UTC)

I think it would be valuable to include some examples, but it does not necessarily have to be the diverse nazis one, perhaps the vikings or pope generations would be better but that's just my thoughts Mr Vili talk 04:45, 25 February 2024 (UTC)

We could also have a collage of each type, maybe four of the most striking ones. A Pope, a Viking, a Nazi, and a Founding Father, or something like that. Since AI-generated images aren't copyrightable, no publication or social media user can claim copyright to the images they post online, so we'll have plenty of options to choose from (it's no longer possible to generate ones ourselves now that it's been disabled). InfiniteNexus (talk) 05:47, 25 February 2024 (UTC)

Another thing to keep in mind is that a lot of the images posted on social media haven't been verified. (Also, the pope example isn't necessarily historically inaccurate because there has been three African popes.) However, it seems The Verge has some examples we could use since they did some independent reporting. ^Falling_Gravity 23:16, 25 February 2024 (UTC)

I think it's safe to assume most of the images posted on social media and referenced in the articles are real generations by Gemini, it's impossible to verify now anecdotally since they have disabled image generation, but we must take the sources as historical accounts of the events.

There are a number of reliable sources that covered it which we can safely assume are valid generations. Mr Vili talk 13:44, 26 February 2024 (UTC)

Inaccurate framing of controversy

The image generation controversy concerns "inaccurate depictions of people of colour", according to reliable sources, not some nonsense about "wokeness" and "anti-white racism". Wikipedia should not legitimize the fringe conspiracy theories of far-right twitter users. 46.97.170.155 (talk) 10:22, 26 February 2024 (UTC)

Are you claiming that Gemini did not typically refuse to generate white people but happily accept generations of black people, or that it did not significantly over represent people of color in historically inaccurate examples? Because that would be inaccurate.

Wikipedia is not legitimizing neither wokeness nor anti-wokeness, it is merely stating the objective and neutral reality of what the sources claim, which are verifiable. Mr Vili talk 13:41, 26 February 2024 (UTC)

I'm claiming that wikipedia needs to go by reliable secondary sources, as per site policy, and reliable sources are talking about "innacurate portrayals of people of color" and "putting people of colour in nazi uniforms". None of that "erasing white history" nonsense that certain people with very obvious agendas are fearmongering about. 46.97.170.155 (talk) 15:06, 26 February 2024 (UTC)

We're doing any of that. Any opinions (including claims about "wokenes" and all that) are properly attributed to their authors. Everything is factually accurate and objectively presented. InfiniteNexus (talk) 22:39, 27 February 2024 (UTC)

Even so, the current wording reads as heavily slanted towards the right wing narrative, and presents the conspiracy theories about google's "wokeness" (lol) uncritically. 46.97.170.155 (talk) 10:21, 28 February 2024 (UTC)

How so? Can you specify which portion you believe violates WP:NPOV? Again, we've focused on the facts here — what images Gemini produced, who criticized Gemini and how, and what Google's response was. InfiniteNexus (talk) 18:33, 28 February 2024 (UTC)

Pointing out that much of the "backlash" coming from the right leaned heavily into the antisemitic White Genocide Conspiracy Theory would be a good start. 46.97.170.155 (talk) 10:32, 29 February 2024 (UTC)

Well, it would be heavily slanted towards the "left wing" if the controversy was that Gemini was refusing to make images of diverse people. We are merely objectively stating the facts from all the various sources... Mr Vili talk 23:14, 28 February 2024 (UTC)