On Dec 29, 2022, at 5:17 PM, Steven Walling <steven.walling(a)gmail.com> wrote:  This is a cool idea but what would the goals of developing a Wikipedia-specific generative AI be? IMO it would be nice to have a natural language search right in Wikipedia that could return factual answers not just links to our (often too long) articles. OpenAI models aren’t open source btw. Some of the products are free to use right now, but their business model is to charge for API use etc. so including it directly in Wikipedia is pretty much a non-starter. They’re almost certainly using Wikipedia. The answer from ChatGPT is: “ChatGPT is a chatbot model developed by OpenAI. It was trained on a dataset of human-generated text, including data from a variety of sources such as books, articles, and websites. It is possible that some of the data used to train ChatGPT may have come from Wikipedia, as Wikipedia is a widely-used source of information and is likely to be included in many datasets of human-generated text.” Who knows but we would have to develop our own models to test this idea. _______________________________________________ Wikimedia-l mailing list -- wikimedia-l(a)lists.wikimedia.org, guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/wiki/Wikimedia-l Public archives at https://lists.wikimedia.org/hyperkitty/list/wikimedia-l@lists.wikimedia.org… To unsubscribe send an email to wikimedia-l-leave(a)lists.wikimedia.org

I think the simplest answer is yes its an artificial writer but its not intelligence as the name implies but rather just a piece of software that gives answers according to the methodology of that software. The garbage in garbage out format, it can never be better than the programmers behind the machine On Fri, 30 Dec 2022 at 09:56, Victoria Coleman < vstavridoucoleman(a)gmail.com&gt; wrote: -- Boodarwun Gnangarra 'ngany dabakarn koorliny arn boodjera dardoon ngalang Nyungar koortaboodjar' _______________________________________________ Wikimedia-l mailing list -- wikimedia-l(a)lists.wikimedia.org, guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/wiki/Wikimedia-l Public archives at https://lists.wikimedia.org/hyperkitty/list/wikimedia-l@lists.wikimedia.org… To unsubscribe send an email to wikimedia-l-leave(a)lists.wikimedia.org

Hi, just to remark that it superficially looks like a great tool for small language Wikipedias (for which the translation tool is typically not available). One can train the tool in some less common language using the dictionary and some texts, and then let it fill the project with a thousands of articles. (As an aside, in fact, one probably can train it to the soon-to-be-extint languages and save them until the moment there is any interest for revival, but nobody seems to be interested). However, there is a high potential for abuse, as I can imagine people not speaking the language running the tool and creating thousands of substandard articles - we have seen this done manually, and I would be very cautious allowing this. Best Yaroslav On Fri, Dec 30, 2022 at 4:57 AM Raymond Leonard < raymond.f.leonard.jr(a)gmail.com&gt; wrote: _______________________________________________ Wikimedia-l mailing list -- wikimedia-l(a)lists.wikimedia.org, guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/wiki/Wikimedia-l Public archives at https://lists.wikimedia.org/hyperkitty/list/wikimedia-l@lists.wikimedia.org… To unsubscribe send an email to wikimedia-l-leave(a)lists.wikimedia.org

On Dec 30, 2022, at 8:53 AM, Victoria Coleman &lt;vstavridoucoleman(a)gmail.com&gt; wrote: Anne, Interestingly enough what these large companies have to spend a ton of money on is creating and moderating content. In other words people. Passionate volunteers in large numbers is what the movement has in abundance. Imagine the power of combining the talents and passion of our community members with the advances offered by AI today. I was struck recently during a visit to NVIDIA how language models have changed. Back in my day, we would have to build one language model per domain and then load it in to the device, a computer or a phone, to use. Now they have one massive combined language model in a data center full of their GPUs which is there so long as you are connected. My sense is that within the guard rails offered by our volunteer community, we could use AI to force multiply their efforts and make knowledge even more accessible than it is today. Both for those who create and record knowledge as well as those who consume it. In the case of Chat GPT, our volunteers could use supervised learning for example to narrow down the mistakes the bot makes - which should be many fewer that the Open AI version since the Wikipedia version would be trained on good, clean Wikipedia content which is constantly reviewed by the community. Best regards, Victoria Coleman _______________________________________________ Wikimedia-l mailing list -- wikimedia-l(a)lists.wikimedia.org, guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/wiki/Wikimedia-l Public archives at https://lists.wikimedia.org/hyperkitty/list/wikimedia-l@lists.wikimedia.org… To unsubscribe send an email to wikimedia-l-leave(a)lists.wikimedia.org

On Dec 31, 2022, at 11:06 AM, Raymond Leonard &lt;raymond.f.leonard.jr(a)gmail.com&gt; wrote: Of relevance to this conversation: https://www.wired.com/story/large-language-models-artificial-intelligence/ <https://www.wired.com/story/large-language-models-artificial-intelligence/> On Fri, Dec 30, 2022 at 9:32 AM Neurodivergent Netizen &lt;idoh.idreamofhorses(a)gmail.com <mailto:idoh.idreamofhorses@gmail.com>> wrote: One concern I have is that all “oldbies” like myself have all seen bots basically decay after whomever is maintaining goes inactive. Of course, this could be mostly rectified by having the AI be open source. This leaves the “people” aspect; that is, not only does the AI need to be maintained, but interest needs to be maintained as well. From, I dream of horses She/her _______________________________________________ Wikimedia-l mailing list -- wikimedia-l(a)lists.wikimedia.org <mailto:wikimedia-l@lists.wikimedia.org>, guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines <https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines> and https://meta.wikimedia.org/wiki/Wikimedia-l <https://meta.wikimedia.org/wiki/Wikimedia-l> Public archives at https://lists.wikimedia.org/hyperkitty/list/wikimedia-l@lists.wikimedia.org… <https://lists.wikimedia.org/hyperkitty/list/wikimedia-l@lists.wikimedia.org/message/2UBNTXB72SIMB7NRXSLQNBYJNVFQAO4E/> To unsubscribe send an email to wikimedia-l-leave(a)lists.wikimedia.org <mailto:wikimedia-l-leave@lists.wikimedia.org>_______________________________________________ Wikimedia-l mailing list -- wikimedia-l(a)lists.wikimedia.org, guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/wiki/Wikimedia-l Public archives at https://lists.wikimedia.org/hyperkitty/list/wikimedia-l@lists.wikimedia.org… To unsubscribe send an email to wikimedia-l-leave(a)lists.wikimedia.org

On Fri, Feb 3, 2023 at 9:47 PM Gergő Tisza &lt;gtisza(a)gmail.com&gt; wrote: Without spending billions on LLM development to aim for a conversational chatbot trying to pass a Turing test, we could definitely try to catch up to the state of the art in search results. Our search currently does a pretty bad job (in terms of recall especially). Today's featured article in English is the Hot Chip album "Made in the Dark", and if I enter anything but the exact article title the typeahead results are woefully incomplete or wrong. If I ask an actual question, good luck. Google is feeling vulnerable to OpenAI here in part because everyone can see that their results are often full of low quality junk created for SEO, while ChatGPT just gives a concise answer right there. https://en.wikipedia.org/wiki/The_Menu_(2022_film) is one of the top viewed English articles. If I search "The Menu reviews" the Google results are noisy and not so great. ChatGPT actually gives you nothing relevant because it doesn't know anything from 2022. If we could just manage to display the three sentence snippet of our article about the critical response section of the article, it would be awesome. It's too bad that the whole "knowledge engine" debacle poisoned the well when it comes to a Wikipedia search engine, because we could definitely do a lot to learn from what people like about ChatGPT and apply to Wikipedia search. _______________________________________________ _______________________________________________ Wikimedia-l mailing list -- wikimedia-l(a)lists.wikimedia.org, guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/wiki/Wikimedia-l Public archives at https://lists.wikimedia.org/hyperkitty/list/wikimedia-l@lists.wikimedia.org… To unsubscribe send an email to wikimedia-l-leave(a)lists.wikimedia.org

Brainstorming on how to drive traffic to Wikimedia content from conversational media, UI/UX designers could provide menu items or buttons on chatbots' applications or webpage components (e.g., to read more about the content, to navigate to cited resources, to edit the content, to discuss the content, to upvote/downvote the content, to share the content or the recent dialogue history on social media, to request review/moderation/curation for the content, etc.). Many of these envisioned menu items or buttons would operate contextually during dialogues, upon the most recent (or otherwise selected) responses provided by the chatbot or upon the recent transcripts. Some of these features could also be made available to end-users via spoken-language commands. At any point during hypertext-based dialogues, end-users would be able to navigate to Wikimedia content. These navigations could utilize either URL query string arguments or HTTP POST. In either case, bulk usage data, e.g., those dialogue contexts navigated from, could be useful. The capability to perform A/B testing across chatbots’ dialogues, over large populations of end-users, could also be useful. In this way, Wikimedia would be better able to: (1) measure end-user engagement and satisfaction, (2) measure the quality of provided content, (3) perform personalization, (4) retain readers and editors. A/B testing could be performed by providing end-users with various feedback buttons (as described above). A/B testing data could also be obtained through data mining, analyzing end-users’ behaviors, response times, responses, and dialogue moves. These data could be provided for the community at special pages and could be made available per article, possibly by enhancing the “Page information” system. One can also envision these kinds of analytics data existing at the granularity of portions of, or selections of, articles. Best regards, Adam ------------------------------ *From:* Victoria Coleman &lt;vstavridoucoleman(a)gmail.com&gt; *Sent:* Saturday, February 4, 2023 8:10 AM *To:* Wikimedia Mailing List &lt;wikimedia-l(a)lists.wikimedia.org&gt; *Subject:* [Wikimedia-l] Re: Chat GPT Hi Christophe, I had not thought about the threat to Wikipedia traffic from Chat GPT but you have a good point. The success of the projects is always one step away from the next big disruption. So the WMF as the tech provider for the mission (because first and foremost in my view that?s what the WMF is - as well as the financial engine of the movement of course) needs to pay attention and experiment to maintain the long term viability of the mission. In fact I think the cluster of our projects offers compelling options. For example to your point below on data sets, we have the amazing Wikidata as well the excellent work on abstract Wikipedia. We have Wikipedia Enterprise which has built some avenues of collaboration with big tech. A bold vision is needed to bring all of it together and build an MVP for the community to experiment with. Best regards, Victoria Coleman On Feb 4, 2023, at 4:14 AM, Christophe Henner &lt;christophe.henner(a)gmail.com&gt; wrote: ?Hi, On the product side, NLP based AI biggest concern to me is that it would drastically decrease traffic to our websites/apps. Which means less new editors ans less donations. So first from a strictly positioning perspective, we have here a major change that needs to be managed. And to be honest, it will come faster than we think. We are perfectionists, I can assure you, most companies would be happy to launch a search product with a 80% confidence in answers quality. From a financial perspective, large industrial investment like this are usually a pool of money you can draw from in x years. You can expect they did not draw all of it yet. Second, GPT 3 and ChatGPT are far from being the most expensive products they have. On top of people you need: * datasets * people to tag the dataset * people to correct the algo * computing power I simplify here, but we already have the capacity to muster some of that, which drastically lowers our costs :) I would not discard the option of the movement doing it so easily. That being said, it would mean a new project with the need of substantial ressources. Sent from my iPhone On Feb 4, 2023, at 9:30 AM, Adam Sobieski &lt;adamsobieski(a)hotmail.com&gt; wrote: ? With respect to cloud computing costs, these being a significant component of the costs to train and operate modern AI systems, as a non-profit organization, the Wikimedia Foundation might be interested in the National Research Cloud (NRC) policy proposal: https://hai.stanford.edu/policy/national-research-cloud . "Artificial intelligence requires vast amounts of computing power, data, and expertise to train and deploy the massive machine learning models behind the most advanced research. But access is increasingly out of reach for most colleges and universities. A National Research Cloud (NRC) would provide academic and *non-profit researchers* with the compute power and government datasets needed for education and research. By democratizing access and equity for all colleges and universities, an NRC has the potential not only to unleash a string of advancements in AI, but to help ensure the U.S. maintains its leadership and competitiveness on the global stage. "Throughout 2020, Stanford HAI led efforts with 22 top computer science universities along with a bipartisan, bicameral group of lawmakers proposing legislation to bring the NRC to fruition. On January 1, 2021, the U.S. Congress authorized the National AI Research Resource Task Force Act as part of the National Defense Authorization Act for Fiscal Year 2021. This law requires that a federal task force be established to study and provide an implementation pathway to create world-class computational resources and robust government datasets for researchers across the country in the form of a National Research Cloud. The task force will issue a final report to the President and Congress next year. "The promise of an NRC is to democratize AI research, education, and innovation, making it accessible to all colleges and universities across the country. Without a National Research Cloud, all but the most elite universities risk losing the ability to conduct meaningful AI research and to adequately educate the next generation of AI researchers." See also: [1][2] [1] https://www.whitehouse.gov/ostp/news-updates/2023/01/24/national-artificial… [2] https://www.ai.gov/wp-content/uploads/2023/01/NAIRR-TF-Final-Report-2023.pdf ------------------------------ *From:* Steven Walling &lt;steven.walling(a)gmail.com&gt; *Sent:* Saturday, February 4, 2023 1:59 AM *To:* Wikimedia Mailing List &lt;wikimedia-l(a)lists.wikimedia.org&gt; *Subject:* [Wikimedia-l] Re: Chat GPT On Fri, Feb 3, 2023 at 9:47 PM Gerg? Tisza &lt;gtisza(a)gmail.com&gt; wrote: Just to give a sense of scale: OpenAI started with a $1 billion donation, got another $1B as investment, and is now getting a larger investment from Microsoft (undisclosed but rumored to be $10B). Assuming they spent most of their previous funding, which seems likely, their operational costs are in the ballpark of $300 million per year. The idea that the WMF could just choose to create conversational software of a similar quality if it wanted seems detached from reality to me. Without spending billions on LLM development to aim for a conversational chatbot trying to pass a Turing test, we could definitely try to catch up to the state of the art in search results. Our search currently does a pretty bad job (in terms of recall especially). Today's featured article in English is the Hot Chip album "Made in the Dark", and if I enter anything but the exact article title the typeahead results are woefully incomplete or wrong. If I ask an actual question, good luck. Google is feeling vulnerable to OpenAI here in part because everyone can see that their results are often full of low quality junk created for SEO, while ChatGPT just gives a concise answer right there. https://en.wikipedia.org/wiki/The_Menu_(2022_film) is one of the top viewed English articles. If I search "The Menu reviews" the Google results are noisy and not so great. ChatGPT actually gives you nothing relevant because it doesn't know anything from 2022. If we could just manage to display the three sentence snippet of our article about the critical response section of the article, it would be awesome. It's too bad that the whole "knowledge engine" debacle poisoned the well when it comes to a Wikipedia search engine, because we could definitely do a lot to learn from what people like about ChatGPT and apply to Wikipedia search. _______________________________________________ Wikimedia-l mailing list -- wikimedia-l(a)lists.wikimedia.org, guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/wiki/Wikimedia-l Public archives at https://lists.wikimedia.org/hyperkitty/list/wikimedia-l@lists.wikimedia.org… To unsubscribe send an email to wikimedia-l-leave(a)lists.wikimedia.org _______________________________________________ Wikimedia-l mailing list -- wikimedia-l(a)lists.wikimedia.org, guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/wiki/Wikimedia-l Public archives at https://lists.wikimedia.org/hyperkitty/list/wikimedia-l@lists.wikimedia.org… To unsubscribe send an email to wikimedia-l-leave(a)lists.wikimedia.org _______________________________________________ Wikimedia-l mailing list -- wikimedia-l(a)lists.wikimedia.org, guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/wiki/Wikimedia-l Public archives at https://lists.wikimedia.org/hyperkitty/list/wikimedia-l@lists.wikimedia.org… To unsubscribe send an email to wikimedia-l-leave(a)lists.wikimedia.org _______________________________________________ Wikimedia-l mailing list -- wikimedia-l(a)lists.wikimedia.org, guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/wiki/Wikimedia-l Public archives at https://lists.wikimedia.org/hyperkitty/list/wikimedia-l@lists.wikimedia.org… To unsubscribe send an email to wikimedia-l-leave(a)lists.wikimedia.org _______________________________________________ Wikimedia-l mailing list -- wikimedia-l(a)lists.wikimedia.org, guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/wiki/Wikimedia-l Public archives at https://lists.wikimedia.org/hyperkitty/list/wikimedia-l@lists.wikimedia.org… To unsubscribe send an email to wikimedia-l-leave(a)lists.wikimedia.org

From what I have seen the AIs are not great on citing sources. If they start citing reliable sources, their contributions can be verified, or not. If they produce verifiable, adequately sourced, well written information, are they a problem or a solution? Cheers, Peter *From:* Gnangarra [mailto:gnangarra@gmail.com] *Sent:* 04 February 2023 17:04 *To:* Wikimedia Mailing List *Subject:* [Wikimedia-l] Re: Chat GPT I see our biggest challenge is going to be detecting these AI tools adding content whether it's media or articles, along with identifying when they are in use by sources. The failing of all new AI is not in its ability but in the lack of transparency with that being able to be identified by the readers. We have seen people impersonating musicians and writing songs in their style. We have also seen pictures that have been created by copying someone else's work yet not acknowledging it as being derivative of any kind. Our big problems will be in ensuring that copyright is respected in legally, and not hosting anything that is even remotely dubious On Sat, 4 Feb 2023 at 22:24, Adam Sobieski &lt;adamsobieski(a)hotmail.com&gt; wrote: Brainstorming on how to drive traffic to Wikimedia content from conversational media, UI/UX designers could provide menu items or buttons on chatbots' applications or webpage components (e.g., to read more about the content, to navigate to cited resources, to edit the content, to discuss the content, to upvote/downvote the content, to share the content or the recent dialogue history on social media, to request review/moderation/curation for the content, etc.). Many of these envisioned menu items or buttons would operate contextually during dialogues, upon the most recent (or otherwise selected) responses provided by the chatbot or upon the recent transcripts. Some of these features could also be made available to end-users via spoken-language commands. At any point during hypertext-based dialogues, end-users would be able to navigate to Wikimedia content. These navigations could utilize either URL query string arguments or HTTP POST. In either case, bulk usage data, e.g., those dialogue contexts navigated from, could be useful. The capability to perform A/B testing across chatbots’ dialogues, over large populations of end-users, could also be useful. In this way, Wikimedia would be better able to: (1) measure end-user engagement and satisfaction, (2) measure the quality of provided content, (3) perform personalization, (4) retain readers and editors. A/B testing could be performed by providing end-users with various feedback buttons (as described above). A/B testing data could also be obtained through data mining, analyzing end-users’ behaviors, response times, responses, and dialogue moves. These data could be provided for the community at special pages and could be made available per article, possibly by enhancing the “Page information” system. One can also envision these kinds of analytics data existing at the granularity of portions of, or selections of, articles. Best regards, Adam ------------------------------ *From:* Victoria Coleman &lt;vstavridoucoleman(a)gmail.com&gt; *Sent:* Saturday, February 4, 2023 8:10 AM *To:* Wikimedia Mailing List &lt;wikimedia-l(a)lists.wikimedia.org&gt; *Subject:* [Wikimedia-l] Re: Chat GPT Hi Christophe, I had not thought about the threat to Wikipedia traffic from Chat GPT but you have a good point. The success of the projects is always one step away from the next big disruption. So the WMF as the tech provider for the mission (because first and foremost in my view that?s what the WMF is - as well as the financial engine of the movement of course) needs to pay attention and experiment to maintain the long term viability of the mission. In fact I think the cluster of our projects offers compelling options. For example to your point below on data sets, we have the amazing Wikidata as well the excellent work on abstract Wikipedia. We have Wikipedia Enterprise which has built some avenues of collaboration with big tech. A bold vision is needed to bring all of it together and build an MVP for the community to experiment with. Best regards, Victoria Coleman On Feb 4, 2023, at 4:14 AM, Christophe Henner &lt;christophe.henner(a)gmail.com&gt; wrote: ?Hi, On the product side, NLP based AI biggest concern to me is that it would drastically decrease traffic to our websites/apps. Which means less new editors ans less donations. So first from a strictly positioning perspective, we have here a major change that needs to be managed. And to be honest, it will come faster than we think. We are perfectionists, I can assure you, most companies would be happy to launch a search product with a 80% confidence in answers quality. From a financial perspective, large industrial investment like this are usually a pool of money you can draw from in x years. You can expect they did not draw all of it yet. Second, GPT 3 and ChatGPT are far from being the most expensive products they have. On top of people you need: * datasets * people to tag the dataset * people to correct the algo * computing power I simplify here, but we already have the capacity to muster some of that, which drastically lowers our costs :) I would not discard the option of the movement doing it so easily. That being said, it would mean a new project with the need of substantial ressources. Sent from my iPhone On Feb 4, 2023, at 9:30 AM, Adam Sobieski &lt;adamsobieski(a)hotmail.com&gt; wrote: ? With respect to cloud computing costs, these being a significant component of the costs to train and operate modern AI systems, as a non-profit organization, the Wikimedia Foundation might be interested in the National Research Cloud (NRC) policy proposal: https://hai.stanford.edu/policy/national-research-cloud . "Artificial intelligence requires vast amounts of computing power, data, and expertise to train and deploy the massive machine learning models behind the most advanced research. But access is increasingly out of reach for most colleges and universities. A National Research Cloud (NRC) would provide academic and *non-profit researchers* with the compute power and government datasets needed for education and research. By democratizing access and equity for all colleges and universities, an NRC has the potential not only to unleash a string of advancements in AI, but to help ensure the U.S. maintains its leadership and competitiveness on the global stage. "Throughout 2020, Stanford HAI led efforts with 22 top computer science universities along with a bipartisan, bicameral group of lawmakers proposing legislation to bring the NRC to fruition. On January 1, 2021, the U.S. Congress authorized the National AI Research Resource Task Force Act as part of the National Defense Authorization Act for Fiscal Year 2021. This law requires that a federal task force be established to study and provide an implementation pathway to create world-class computational resources and robust government datasets for researchers across the country in the form of a National Research Cloud. The task force will issue a final report to the President and Congress next year. "The promise of an NRC is to democratize AI research, education, and innovation, making it accessible to all colleges and universities across the country. Without a National Research Cloud, all but the most elite universities risk losing the ability to conduct meaningful AI research and to adequately educate the next generation of AI researchers." See also: [1][2] [1] https://www.whitehouse.gov/ostp/news-updates/2023/01/24/national-artificial… [2] https://www.ai.gov/wp-content/uploads/2023/01/NAIRR-TF-Final-Report-2023.pdf ------------------------------ *From:* Steven Walling &lt;steven.walling(a)gmail.com&gt; *Sent:* Saturday, February 4, 2023 1:59 AM *To:* Wikimedia Mailing List &lt;wikimedia-l(a)lists.wikimedia.org&gt; *Subject:* [Wikimedia-l] Re: Chat GPT On Fri, Feb 3, 2023 at 9:47 PM Gerg? Tisza &lt;gtisza(a)gmail.com&gt; wrote: Just to give a sense of scale: OpenAI started with a $1 billion donation, got another $1B as investment, and is now getting a larger investment from Microsoft (undisclosed but rumored to be $10B). Assuming they spent most of their previous funding, which seems likely, their operational costs are in the ballpark of $300 million per year. The idea that the WMF could just choose to create conversational software of a similar quality if it wanted seems detached from reality to me. Without spending billions on LLM development to aim for a conversational chatbot trying to pass a Turing test, we could definitely try to catch up to the state of the art in search results. Our search currently does a pretty bad job (in terms of recall especially). Today's featured article in English is the Hot Chip album "Made in the Dark", and if I enter anything but the exact article title the typeahead results are woefully incomplete or wrong. If I ask an actual question, good luck. Google is feeling vulnerable to OpenAI here in part because everyone can see that their results are often full of low quality junk created for SEO, while ChatGPT just gives a concise answer right there. https://en.wikipedia.org/wiki/The_Menu_(2022_film) is one of the top viewed English articles. If I search "The Menu reviews" the Google results are noisy and not so great. ChatGPT actually gives you nothing relevant because it doesn't know anything from 2022. If we could just manage to display the three sentence snippet of our article about the critical response section of the article, it would be awesome. It's too bad that the whole "knowledge engine" debacle poisoned the well when it comes to a Wikipedia search engine, because we could definitely do a lot to learn from what people like about ChatGPT and apply to Wikipedia search. _______________________________________________ Wikimedia-l mailing list -- wikimedia-l(a)lists.wikimedia.org, guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/wiki/Wikimedia-l Public archives at https://lists.wikimedia.org/hyperkitty/list/wikimedia-l@lists.wikimedia.org… To unsubscribe send an email to wikimedia-l-leave(a)lists.wikimedia.org _______________________________________________ Wikimedia-l mailing list -- wikimedia-l(a)lists.wikimedia.org, guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/wiki/Wikimedia-l Public archives at https://lists.wikimedia.org/hyperkitty/list/wikimedia-l@lists.wikimedia.org… To unsubscribe send an email to wikimedia-l-leave(a)lists.wikimedia.org _______________________________________________ Wikimedia-l mailing list -- wikimedia-l(a)lists.wikimedia.org, guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/wiki/Wikimedia-l Public archives at https://lists.wikimedia.org/hyperkitty/list/wikimedia-l@lists.wikimedia.org… To unsubscribe send an email to wikimedia-l-leave(a)lists.wikimedia.org _______________________________________________ Wikimedia-l mailing list -- wikimedia-l(a)lists.wikimedia.org, guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/wiki/Wikimedia-l Public archives at https://lists.wikimedia.org/hyperkitty/list/wikimedia-l@lists.wikimedia.org… To unsubscribe send an email to wikimedia-l-leave(a)lists.wikimedia.org _______________________________________________ Wikimedia-l mailing list -- wikimedia-l(a)lists.wikimedia.org, guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/wiki/Wikimedia-l Public archives at https://lists.wikimedia.org/hyperkitty/list/wikimedia-l@lists.wikimedia.org… To unsubscribe send an email to wikimedia-l-leave(a)lists.wikimedia.org -- Boodarwun Gnangarra 'ngany dabakarn koorliny arn boodjera dardoon ngalang Nyungar koortaboodjar' <http://www.avg.com/email-signature?utm_medium=email&utm_source=link&utm_campaign=sig-email&utm_content=emailclient> Virus-free.www.avg.com <http://www.avg.com/email-signature?utm_medium=email&utm_source=link&utm_campaign=sig-email&utm_content=emailclient> _______________________________________________ Wikimedia-l mailing list -- wikimedia-l(a)lists.wikimedia.org, guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/wiki/Wikimedia-l Public archives at https://lists.wikimedia.org/hyperkitty/list/wikimedia-l@lists.wikimedia.org… To unsubscribe send an email to wikimedia-l-leave(a)lists.wikimedia.org

I'm not so sure Getty's got a case, though. If the images are on the Web, is using them to train an AI something copyright would cover? That to me seems more equivalent to just looking at the images, and there's no copyright problem in going to Getty's site and just looking at a bunch of their pictures. But it will be interesting to see how that one shakes out. Todd On Sat, Feb 4, 2023 at 11:47 AM Subhashish &lt;psubhashish(a)gmail.com&gt; wrote: _______________________________________________ Wikimedia-l mailing list -- wikimedia-l(a)lists.wikimedia.org, guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/wiki/Wikimedia-l Public archives at https://lists.wikimedia.org/hyperkitty/list/wikimedia-l@lists.wikimedia.org… To unsubscribe send an email to wikimedia-l-leave(a)lists.wikimedia.org

“Not citing sources is probably a conscious design choice, as citing sources would mean sharing the sources used to train the language models” This may be a choice that comes back to bite them. Without citing their sources, they are unreliable as a source for anything one does not know already. Someone will have a bad consequence from relying on the information and will sue the publisher. It will be interesting to see how they plan to weasel their way out of legal responsibility while retaining any credibility. My guess is there will be a requirement to state that the information is AI generated and of entirely unknown and untested reliability. How soon to the first class action, I wonder. Lots of money for the lawyers. Cheers, Peter. *From:* Subhashish [mailto:psubhashish@gmail.com] *Sent:* 05 February 2023 06:37 *To:* Wikimedia Mailing List *Subject:* [Wikimedia-l] Re: Chat GPT Just to clarify, my point was not about Getty to begin with. Whether Getty would win and whether a big corporation should own such a large amount of visual content are questions outside this particular thread. It would certainly be interesting to see how things roll. But AI/ML is way more than just looking. Training with large models is a very sophisticated and technical process. Data annotation among many other forms of labour are done by real people. the article I had linked earlier tells a lot about the real world consequences of AI. I'm certain AI/ML, especially when we're talking about language models like ChatGPT, are far from innocent looking/reading. For starters, derivative of works, except Public Domain ones, must attribute the authors. Any provision for attribution is deliberately removed from systems like ChatGPT and that only gives corporations like OpenAI a free ride sans accountability. Subhashish On Sat, Feb 4, 2023, 4:41 PM Todd Allen &lt;toddmallen(a)gmail.com&gt; wrote: I'm not so sure Getty's got a case, though. If the images are on the Web, is using them to train an AI something copyright would cover? That to me seems more equivalent to just looking at the images, and there's no copyright problem in going to Getty's site and just looking at a bunch of their pictures. But it will be interesting to see how that one shakes out. Todd On Sat, Feb 4, 2023 at 11:47 AM Subhashish &lt;psubhashish(a)gmail.com&gt; wrote: Not citing sources is probably a conscious design choice, as citing sources would mean sharing the sources used to train the language models. Getty has just sued Stability AI, alleging the use of 12 million photographs without permission or compensation. Imagine if Stability had to purchase from Getty through a legal process. For starters, Getty might not have agreed in the first place. Bulk-scaping publicly visible text in text-based AIs like ChatGPT would mean scraping text with copyright. But even reusing CC BY-SA content would require attribution. None of the AI platforms attributes their sources because they did not acquire content in legal and ethical ways [1]. Large language models won't be large and releases won't happen fast if they actually start acquiring content gradually from trustworthy sources. It took so many years for hundreds and thousands of Wikimedians to take Wikipedias in different languages to where they are for a reason. 1. https://time.com/6247678/openai-chatgpt-kenya-workers/ Subhashish On Sat, Feb 4, 2023 at 1:06 PM Peter Southwood < peter.southwood(a)telkomsa.net&gt; wrote: From what I have seen the AIs are not great on citing sources. If they start citing reliable sources, their contributions can be verified, or not. If they produce verifiable, adequately sourced, well written information, are they a problem or a solution? Cheers, Peter *From:* Gnangarra [mailto:gnangarra@gmail.com] *Sent:* 04 February 2023 17:04 *To:* Wikimedia Mailing List *Subject:* [Wikimedia-l] Re: Chat GPT I see our biggest challenge is going to be detecting these AI tools adding content whether it's media or articles, along with identifying when they are in use by sources. The failing of all new AI is not in its ability but in the lack of transparency with that being able to be identified by the readers. We have seen people impersonating musicians and writing songs in their style. We have also seen pictures that have been created by copying someone else's work yet not acknowledging it as being derivative of any kind. Our big problems will be in ensuring that copyright is respected in legally, and not hosting anything that is even remotely dubious On Sat, 4 Feb 2023 at 22:24, Adam Sobieski &lt;adamsobieski(a)hotmail.com&gt; wrote: Brainstorming on how to drive traffic to Wikimedia content from conversational media, UI/UX designers could provide menu items or buttons on chatbots' applications or webpage components (e.g., to read more about the content, to navigate to cited resources, to edit the content, to discuss the content, to upvote/downvote the content, to share the content or the recent dialogue history on social media, to request review/moderation/curation for the content, etc.). Many of these envisioned menu items or buttons would operate contextually during dialogues, upon the most recent (or otherwise selected) responses provided by the chatbot or upon the recent transcripts. Some of these features could also be made available to end-users via spoken-language commands. At any point during hypertext-based dialogues, end-users would be able to navigate to Wikimedia content. These navigations could utilize either URL query string arguments or HTTP POST. In either case, bulk usage data, e.g., those dialogue contexts navigated from, could be useful. The capability to perform A/B testing across chatbots’ dialogues, over large populations of end-users, could also be useful. In this way, Wikimedia would be better able to: (1) measure end-user engagement and satisfaction, (2) measure the quality of provided content, (3) perform personalization, (4) retain readers and editors. A/B testing could be performed by providing end-users with various feedback buttons (as described above). A/B testing data could also be obtained through data mining, analyzing end-users’ behaviors, response times, responses, and dialogue moves. These data could be provided for the community at special pages and could be made available per article, possibly by enhancing the “Page information” system. One can also envision these kinds of analytics data existing at the granularity of portions of, or selections of, articles. Best regards, Adam ------------------------------ *From:* Victoria Coleman &lt;vstavridoucoleman(a)gmail.com&gt; *Sent:* Saturday, February 4, 2023 8:10 AM *To:* Wikimedia Mailing List &lt;wikimedia-l(a)lists.wikimedia.org&gt; *Subject:* [Wikimedia-l] Re: Chat GPT Hi Christophe, I had not thought about the threat to Wikipedia traffic from Chat GPT but you have a good point. The success of the projects is always one step away from the next big disruption. So the WMF as the tech provider for the mission (because first and foremost in my view that?s what the WMF is - as well as the financial engine of the movement of course) needs to pay attention and experiment to maintain the long term viability of the mission. In fact I think the cluster of our projects offers compelling options. For example to your point below on data sets, we have the amazing Wikidata as well the excellent work on abstract Wikipedia. We have Wikipedia Enterprise which has built some avenues of collaboration with big tech. A bold vision is needed to bring all of it together and build an MVP for the community to experiment with. Best regards, Victoria Coleman On Feb 4, 2023, at 4:14 AM, Christophe Henner &lt;christophe.henner(a)gmail.com&gt; wrote: ?Hi, On the product side, NLP based AI biggest concern to me is that it would drastically decrease traffic to our websites/apps. Which means less new editors ans less donations. So first from a strictly positioning perspective, we have here a major change that needs to be managed. And to be honest, it will come faster than we think. We are perfectionists, I can assure you, most companies would be happy to launch a search product with a 80% confidence in answers quality. From a financial perspective, large industrial investment like this are usually a pool of money you can draw from in x years. You can expect they did not draw all of it yet. Second, GPT 3 and ChatGPT are far from being the most expensive products they have. On top of people you need: * datasets * people to tag the dataset * people to correct the algo * computing power I simplify here, but we already have the capacity to muster some of that, which drastically lowers our costs :) I would not discard the option of the movement doing it so easily. That being said, it would mean a new project with the need of substantial ressources. Sent from my iPhone On Feb 4, 2023, at 9:30 AM, Adam Sobieski &lt;adamsobieski(a)hotmail.com&gt; wrote: ? With respect to cloud computing costs, these being a significant component of the costs to train and operate modern AI systems, as a non-profit organization, the Wikimedia Foundation might be interested in the National Research Cloud (NRC) policy proposal: https://hai.stanford.edu/policy/national-research-cloud . "Artificial intelligence requires vast amounts of computing power, data, and expertise to train and deploy the massive machine learning models behind the most advanced research. But access is increasingly out of reach for most colleges and universities. A National Research Cloud (NRC) would provide academic and *non-profit researchers* with the compute power and government datasets needed for education and research. By democratizing access and equity for all colleges and universities, an NRC has the potential not only to unleash a string of advancements in AI, but to help ensure the U.S. maintains its leadership and competitiveness on the global stage. "Throughout 2020, Stanford HAI led efforts with 22 top computer science universities along with a bipartisan, bicameral group of lawmakers proposing legislation to bring the NRC to fruition. On January 1, 2021, the U.S. Congress authorized the National AI Research Resource Task Force Act as part of the National Defense Authorization Act for Fiscal Year 2021. This law requires that a federal task force be established to study and provide an implementation pathway to create world-class computational resources and robust government datasets for researchers across the country in the form of a National Research Cloud. The task force will issue a final report to the President and Congress next year. "The promise of an NRC is to democratize AI research, education, and innovation, making it accessible to all colleges and universities across the country. Without a National Research Cloud, all but the most elite universities risk losing the ability to conduct meaningful AI research and to adequately educate the next generation of AI researchers." See also: [1][2] [1] https://www.whitehouse.gov/ostp/news-updates/2023/01/24/national-artificial… [2] https://www.ai.gov/wp-content/uploads/2023/01/NAIRR-TF-Final-Report-2023.pdf ------------------------------ *From:* Steven Walling &lt;steven.walling(a)gmail.com&gt; *Sent:* Saturday, February 4, 2023 1:59 AM *To:* Wikimedia Mailing List &lt;wikimedia-l(a)lists.wikimedia.org&gt; *Subject:* [Wikimedia-l] Re: Chat GPT On Fri, Feb 3, 2023 at 9:47 PM Gerg? Tisza &lt;gtisza(a)gmail.com&gt; wrote: Just to give a sense of scale: OpenAI started with a $1 billion donation, got another $1B as investment, and is now getting a larger investment from Microsoft (undisclosed but rumored to be $10B). Assuming they spent most of their previous funding, which seems likely, their operational costs are in the ballpark of $300 million per year. The idea that the WMF could just choose to create conversational software of a similar quality if it wanted seems detached from reality to me. Without spending billions on LLM development to aim for a conversational chatbot trying to pass a Turing test, we could definitely try to catch up to the state of the art in search results. Our search currently does a pretty bad job (in terms of recall especially). Today's featured article in English is the Hot Chip album "Made in the Dark", and if I enter anything but the exact article title the typeahead results are woefully incomplete or wrong. If I ask an actual question, good luck. Google is feeling vulnerable to OpenAI here in part because everyone can see that their results are often full of low quality junk created for SEO, while ChatGPT just gives a concise answer right there. https://en.wikipedia.org/wiki/The_Menu_(2022_film) is one of the top viewed English articles. If I search "The Menu reviews" the Google results are noisy and not so great. ChatGPT actually gives you nothing relevant because it doesn't know anything from 2022. If we could just manage to display the three sentence snippet of our article about the critical response section of the article, it would be awesome. It's too bad that the whole "knowledge engine" debacle poisoned the well when it comes to a Wikipedia search engine, because we could definitely do a lot to learn from what people like about ChatGPT and apply to Wikipedia search. _______________________________________________ Wikimedia-l mailing list -- wikimedia-l(a)lists.wikimedia.org, guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/wiki/Wikimedia-l Public archives at https://lists.wikimedia.org/hyperkitty/list/wikimedia-l@lists.wikimedia.org… To unsubscribe send an email to wikimedia-l-leave(a)lists.wikimedia.org _______________________________________________ Wikimedia-l mailing list -- wikimedia-l(a)lists.wikimedia.org, guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/wiki/Wikimedia-l Public archives at https://lists.wikimedia.org/hyperkitty/list/wikimedia-l@lists.wikimedia.org… To unsubscribe send an email to wikimedia-l-leave(a)lists.wikimedia.org _______________________________________________ Wikimedia-l mailing list -- wikimedia-l(a)lists.wikimedia.org, guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/wiki/Wikimedia-l Public archives at https://lists.wikimedia.org/hyperkitty/list/wikimedia-l@lists.wikimedia.org… To unsubscribe send an email to wikimedia-l-leave(a)lists.wikimedia.org _______________________________________________ Wikimedia-l mailing list -- wikimedia-l(a)lists.wikimedia.org, guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/wiki/Wikimedia-l Public archives at https://lists.wikimedia.org/hyperkitty/list/wikimedia-l@lists.wikimedia.org… To unsubscribe send an email to wikimedia-l-leave(a)lists.wikimedia.org _______________________________________________ Wikimedia-l mailing list -- wikimedia-l(a)lists.wikimedia.org, guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/wiki/Wikimedia-l Public archives at https://lists.wikimedia.org/hyperkitty/list/wikimedia-l@lists.wikimedia.org… To unsubscribe send an email to wikimedia-l-leave(a)lists.wikimedia.org -- Boodarwun Gnangarra 'ngany dabakarn koorliny arn boodjera dardoon ngalang Nyungar koortaboodjar' <http://www.avg.com/email-signature?utm_medium=email&utm_source=link&utm_campaign=sig-email&utm_content=emailclient> Virus-free.www.avg.com <http://www.avg.com/email-signature?utm_medium=email&utm_source=link&utm_campaign=sig-email&utm_content=emailclient> _______________________________________________ Wikimedia-l mailing list -- wikimedia-l(a)lists.wikimedia.org, guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/wiki/Wikimedia-l Public archives at https://lists.wikimedia.org/hyperkitty/list/wikimedia-l@lists.wikimedia.org… To unsubscribe send an email to wikimedia-l-leave(a)lists.wikimedia.org _______________________________________________ Wikimedia-l mailing list -- wikimedia-l(a)lists.wikimedia.org, guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/wiki/Wikimedia-l Public archives at https://lists.wikimedia.org/hyperkitty/list/wikimedia-l@lists.wikimedia.org… To unsubscribe send an email to wikimedia-l-leave(a)lists.wikimedia.org _______________________________________________ Wikimedia-l mailing list -- wikimedia-l(a)lists.wikimedia.org, guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/wiki/Wikimedia-l Public archives at https://lists.wikimedia.org/hyperkitty/list/wikimedia-l@lists.wikimedia.org… To unsubscribe send an email to wikimedia-l-leave(a)lists.wikimedia.org _______________________________________________ Wikimedia-l mailing list -- wikimedia-l(a)lists.wikimedia.org, guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/wiki/Wikimedia-l Public archives at https://lists.wikimedia.org/hyperkitty/list/wikimedia-l@lists.wikimedia.org… To unsubscribe send an email to wikimedia-l-leave(a)lists.wikimedia.org

It would depend on whether it uses the text or the information/data. My guess is that the more it uses its own words, the more drift in meaning there will be, and the less reliable the result, but I have no way to test this hypothesis. Cheers, Peter *From:* Ilario Valdelli [mailto:valdelli@gmail.com] *Sent:* 06 February 2023 09:38 *To:* Wikimedia Mailing List *Subject:* [Wikimedia-l] Re: Chat GPT And this is a problem. If ChatGPT uses open content, there is an infringement of license. Specifically the CC-by-sa if it uses Wikipedia. In this case the attribution must be present. Kind regards On Sun, 5 Feb 2023, 08:12 Peter Southwood, &lt;peter.southwood(a)telkomsa.net&gt; wrote: “Not citing sources is probably a conscious design choice, as citing sources would mean sharing the sources used to train the language models” This may be a choice that comes back to bite them. Without citing their sources, they are unreliable as a source for anything one does not know already. Someone will have a bad consequence from relying on the information and will sue the publisher. It will be interesting to see how they plan to weasel their way out of legal responsibility while retaining any credibility. My guess is there will be a requirement to state that the information is AI generated and of entirely unknown and untested reliability. How soon to the first class action, I wonder. Lots of money for the lawyers. Cheers, Peter. *From:* Subhashish [mailto:psubhashish@gmail.com] *Sent:* 05 February 2023 06:37 *To:* Wikimedia Mailing List *Subject:* [Wikimedia-l] Re: Chat GPT Just to clarify, my point was not about Getty to begin with. Whether Getty would win and whether a big corporation should own such a large amount of visual content are questions outside this particular thread. It would certainly be interesting to see how things roll. But AI/ML is way more than just looking. Training with large models is a very sophisticated and technical process. Data annotation among many other forms of labour are done by real people. the article I had linked earlier tells a lot about the real world consequences of AI. I'm certain AI/ML, especially when we're talking about language models like ChatGPT, are far from innocent looking/reading. For starters, derivative of works, except Public Domain ones, must attribute the authors. Any provision for attribution is deliberately removed from systems like ChatGPT and that only gives corporations like OpenAI a free ride sans accountability. Subhashish On Sat, Feb 4, 2023, 4:41 PM Todd Allen &lt;toddmallen(a)gmail.com&gt; wrote: I'm not so sure Getty's got a case, though. If the images are on the Web, is using them to train an AI something copyright would cover? That to me seems more equivalent to just looking at the images, and there's no copyright problem in going to Getty's site and just looking at a bunch of their pictures. But it will be interesting to see how that one shakes out. Todd On Sat, Feb 4, 2023 at 11:47 AM Subhashish &lt;psubhashish(a)gmail.com&gt; wrote: Not citing sources is probably a conscious design choice, as citing sources would mean sharing the sources used to train the language models. Getty has just sued Stability AI, alleging the use of 12 million photographs without permission or compensation. Imagine if Stability had to purchase from Getty through a legal process. For starters, Getty might not have agreed in the first place. Bulk-scaping publicly visible text in text-based AIs like ChatGPT would mean scraping text with copyright. But even reusing CC BY-SA content would require attribution. None of the AI platforms attributes their sources because they did not acquire content in legal and ethical ways [1]. Large language models won't be large and releases won't happen fast if they actually start acquiring content gradually from trustworthy sources. It took so many years for hundreds and thousands of Wikimedians to take Wikipedias in different languages to where they are for a reason. 1. https://time.com/6247678/openai-chatgpt-kenya-workers/ Subhashish On Sat, Feb 4, 2023 at 1:06 PM Peter Southwood < peter.southwood(a)telkomsa.net&gt; wrote: From what I have seen the AIs are not great on citing sources. If they start citing reliable sources, their contributions can be verified, or not. If they produce verifiable, adequately sourced, well written information, are they a problem or a solution? Cheers, Peter *From:* Gnangarra [mailto:gnangarra@gmail.com] *Sent:* 04 February 2023 17:04 *To:* Wikimedia Mailing List *Subject:* [Wikimedia-l] Re: Chat GPT I see our biggest challenge is going to be detecting these AI tools adding content whether it's media or articles, along with identifying when they are in use by sources. The failing of all new AI is not in its ability but in the lack of transparency with that being able to be identified by the readers. We have seen people impersonating musicians and writing songs in their style. We have also seen pictures that have been created by copying someone else's work yet not acknowledging it as being derivative of any kind. Our big problems will be in ensuring that copyright is respected in legally, and not hosting anything that is even remotely dubious On Sat, 4 Feb 2023 at 22:24, Adam Sobieski &lt;adamsobieski(a)hotmail.com&gt; wrote: Brainstorming on how to drive traffic to Wikimedia content from conversational media, UI/UX designers could provide menu items or buttons on chatbots' applications or webpage components (e.g., to read more about the content, to navigate to cited resources, to edit the content, to discuss the content, to upvote/downvote the content, to share the content or the recent dialogue history on social media, to request review/moderation/curation for the content, etc.). Many of these envisioned menu items or buttons would operate contextually during dialogues, upon the most recent (or otherwise selected) responses provided by the chatbot or upon the recent transcripts. Some of these features could also be made available to end-users via spoken-language commands. At any point during hypertext-based dialogues, end-users would be able to navigate to Wikimedia content. These navigations could utilize either URL query string arguments or HTTP POST. In either case, bulk usage data, e.g., those dialogue contexts navigated from, could be useful. The capability to perform A/B testing across chatbots’ dialogues, over large populations of end-users, could also be useful. In this way, Wikimedia would be better able to: (1) measure end-user engagement and satisfaction, (2) measure the quality of provided content, (3) perform personalization, (4) retain readers and editors. A/B testing could be performed by providing end-users with various feedback buttons (as described above). A/B testing data could also be obtained through data mining, analyzing end-users’ behaviors, response times, responses, and dialogue moves. These data could be provided for the community at special pages and could be made available per article, possibly by enhancing the “Page information” system. One can also envision these kinds of analytics data existing at the granularity of portions of, or selections of, articles. Best regards, Adam ------------------------------ *From:* Victoria Coleman &lt;vstavridoucoleman(a)gmail.com&gt; *Sent:* Saturday, February 4, 2023 8:10 AM *To:* Wikimedia Mailing List &lt;wikimedia-l(a)lists.wikimedia.org&gt; *Subject:* [Wikimedia-l] Re: Chat GPT Hi Christophe, I had not thought about the threat to Wikipedia traffic from Chat GPT but you have a good point. The success of the projects is always one step away from the next big disruption. So the WMF as the tech provider for the mission (because first and foremost in my view that?s what the WMF is - as well as the financial engine of the movement of course) needs to pay attention and experiment to maintain the long term viability of the mission. In fact I think the cluster of our projects offers compelling options. For example to your point below on data sets, we have the amazing Wikidata as well the excellent work on abstract Wikipedia. We have Wikipedia Enterprise which has built some avenues of collaboration with big tech. A bold vision is needed to bring all of it together and build an MVP for the community to experiment with. Best regards, Victoria Coleman On Feb 4, 2023, at 4:14 AM, Christophe Henner &lt;christophe.henner(a)gmail.com&gt; wrote: ?Hi, On the product side, NLP based AI biggest concern to me is that it would drastically decrease traffic to our websites/apps. Which means less new editors ans less donations. So first from a strictly positioning perspective, we have here a major change that needs to be managed. And to be honest, it will come faster than we think. We are perfectionists, I can assure you, most companies would be happy to launch a search product with a 80% confidence in answers quality. From a financial perspective, large industrial investment like this are usually a pool of money you can draw from in x years. You can expect they did not draw all of it yet. Second, GPT 3 and ChatGPT are far from being the most expensive products they have. On top of people you need: * datasets * people to tag the dataset * people to correct the algo * computing power I simplify here, but we already have the capacity to muster some of that, which drastically lowers our costs :) I would not discard the option of the movement doing it so easily. That being said, it would mean a new project with the need of substantial ressources. Sent from my iPhone On Feb 4, 2023, at 9:30 AM, Adam Sobieski &lt;adamsobieski(a)hotmail.com&gt; wrote: ? With respect to cloud computing costs, these being a significant component of the costs to train and operate modern AI systems, as a non-profit organization, the Wikimedia Foundation might be interested in the National Research Cloud (NRC) policy proposal: https://hai.stanford.edu/policy/national-research-cloud . "Artificial intelligence requires vast amounts of computing power, data, and expertise to train and deploy the massive machine learning models behind the most advanced research. But access is increasingly out of reach for most colleges and universities. A National Research Cloud (NRC) would provide academic and *non-profit researchers* with the compute power and government datasets needed for education and research. By democratizing access and equity for all colleges and universities, an NRC has the potential not only to unleash a string of advancements in AI, but to help ensure the U.S. maintains its leadership and competitiveness on the global stage. "Throughout 2020, Stanford HAI led efforts with 22 top computer science universities along with a bipartisan, bicameral group of lawmakers proposing legislation to bring the NRC to fruition. On January 1, 2021, the U.S. Congress authorized the National AI Research Resource Task Force Act as part of the National Defense Authorization Act for Fiscal Year 2021. This law requires that a federal task force be established to study and provide an implementation pathway to create world-class computational resources and robust government datasets for researchers across the country in the form of a National Research Cloud. The task force will issue a final report to the President and Congress next year. "The promise of an NRC is to democratize AI research, education, and innovation, making it accessible to all colleges and universities across the country. Without a National Research Cloud, all but the most elite universities risk losing the ability to conduct meaningful AI research and to adequately educate the next generation of AI researchers." See also: [1][2] [1] https://www.whitehouse.gov/ostp/news-updates/2023/01/24/national-artificial… [2] https://www.ai.gov/wp-content/uploads/2023/01/NAIRR-TF-Final-Report-2023.pdf ------------------------------ *From:* Steven Walling &lt;steven.walling(a)gmail.com&gt; *Sent:* Saturday, February 4, 2023 1:59 AM *To:* Wikimedia Mailing List &lt;wikimedia-l(a)lists.wikimedia.org&gt; *Subject:* [Wikimedia-l] Re: Chat GPT On Fri, Feb 3, 2023 at 9:47 PM Gerg? Tisza &lt;gtisza(a)gmail.com&gt; wrote: Just to give a sense of scale: OpenAI started with a $1 billion donation, got another $1B as investment, and is now getting a larger investment from Microsoft (undisclosed but rumored to be $10B). Assuming they spent most of their previous funding, which seems likely, their operational costs are in the ballpark of $300 million per year. The idea that the WMF could just choose to create conversational software of a similar quality if it wanted seems detached from reality to me. Without spending billions on LLM development to aim for a conversational chatbot trying to pass a Turing test, we could definitely try to catch up to the state of the art in search results. Our search currently does a pretty bad job (in terms of recall especially). Today's featured article in English is the Hot Chip album "Made in the Dark", and if I enter anything but the exact article title the typeahead results are woefully incomplete or wrong. If I ask an actual question, good luck. Google is feeling vulnerable to OpenAI here in part because everyone can see that their results are often full of low quality junk created for SEO, while ChatGPT just gives a concise answer right there. https://en.wikipedia.org/wiki/The_Menu_(2022_film) is one of the top viewed English articles. If I search "The Menu reviews" the Google results are noisy and not so great. ChatGPT actually gives you nothing relevant because it doesn't know anything from 2022. If we could just manage to display the three sentence snippet of our article about the critical response section of the article, it would be awesome. It's too bad that the whole "knowledge engine" debacle poisoned the well when it comes to a Wikipedia search engine, because we could definitely do a lot to learn from what people like about ChatGPT and apply to Wikipedia search. _______________________________________________ Wikimedia-l mailing list -- wikimedia-l(a)lists.wikimedia.org, guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/wiki/Wikimedia-l Public archives at https://lists.wikimedia.org/hyperkitty/list/wikimedia-l@lists.wikimedia.org… To unsubscribe send an email to wikimedia-l-leave(a)lists.wikimedia.org _______________________________________________ Wikimedia-l mailing list -- wikimedia-l(a)lists.wikimedia.org, guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/wiki/Wikimedia-l Public archives at https://lists.wikimedia.org/hyperkitty/list/wikimedia-l@lists.wikimedia.org… To unsubscribe send an email to wikimedia-l-leave(a)lists.wikimedia.org _______________________________________________ Wikimedia-l mailing list -- wikimedia-l(a)lists.wikimedia.org, guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/wiki/Wikimedia-l Public archives at https://lists.wikimedia.org/hyperkitty/list/wikimedia-l@lists.wikimedia.org… To unsubscribe send an email to wikimedia-l-leave(a)lists.wikimedia.org _______________________________________________ Wikimedia-l mailing list -- wikimedia-l(a)lists.wikimedia.org, guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/wiki/Wikimedia-l Public archives at https://lists.wikimedia.org/hyperkitty/list/wikimedia-l@lists.wikimedia.org… To unsubscribe send an email to wikimedia-l-leave(a)lists.wikimedia.org _______________________________________________ Wikimedia-l mailing list -- wikimedia-l(a)lists.wikimedia.org, guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/wiki/Wikimedia-l Public archives at https://lists.wikimedia.org/hyperkitty/list/wikimedia-l@lists.wikimedia.org… To unsubscribe send an email to wikimedia-l-leave(a)lists.wikimedia.org -- Boodarwun Gnangarra 'ngany dabakarn koorliny arn boodjera dardoon ngalang Nyungar koortaboodjar' <http://www.avg.com/email-signature?utm_medium=email&utm_source=link&utm_campaign=sig-email&utm_content=emailclient> Virus-free.www.avg.com <http://www.avg.com/email-signature?utm_medium=email&utm_source=link&utm_campaign=sig-email&utm_content=emailclient> _______________________________________________ Wikimedia-l mailing list -- wikimedia-l(a)lists.wikimedia.org, guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/wiki/Wikimedia-l Public archives at https://lists.wikimedia.org/hyperkitty/list/wikimedia-l@lists.wikimedia.org… To unsubscribe send an email to wikimedia-l-leave(a)lists.wikimedia.org _______________________________________________ Wikimedia-l mailing list -- wikimedia-l(a)lists.wikimedia.org, guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/wiki/Wikimedia-l Public archives at https://lists.wikimedia.org/hyperkitty/list/wikimedia-l@lists.wikimedia.org… To unsubscribe send an email to wikimedia-l-leave(a)lists.wikimedia.org _______________________________________________ Wikimedia-l mailing list -- wikimedia-l(a)lists.wikimedia.org, guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/wiki/Wikimedia-l Public archives at https://lists.wikimedia.org/hyperkitty/list/wikimedia-l@lists.wikimedia.org… To unsubscribe send an email to wikimedia-l-leave(a)lists.wikimedia.org _______________________________________________ Wikimedia-l mailing list -- wikimedia-l(a)lists.wikimedia.org, guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/wiki/Wikimedia-l Public archives at https://lists.wikimedia.org/hyperkitty/list/wikimedia-l@lists.wikimedia.org… To unsubscribe send an email to wikimedia-l-leave(a)lists.wikimedia.org _______________________________________________ Wikimedia-l mailing list -- wikimedia-l(a)lists.wikimedia.org, guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/wiki/Wikimedia-l Public archives at https://lists.wikimedia.org/hyperkitty/list/wikimedia-l@lists.wikimedia.org… To unsubscribe send an email to wikimedia-l-leave(a)lists.wikimedia.org

This is almost definitely the case. On Mon, Feb 6, 2023, 2:39 AM Ilario Valdelli &lt;valdelli(a)gmail.com&gt; wrote: _______________________________________________ Wikimedia-l mailing list -- wikimedia-l(a)lists.wikimedia.org, guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/wiki/Wikimedia-l Public archives at https://lists.wikimedia.org/hyperkitty/list/wikimedia-l@lists.wikimedia.org… To unsubscribe send an email to wikimedia-l-leave(a)lists.wikimedia.org