《Razorfish:2023年AI在市场营销、客户服务和艺术领域的变革性潜力研究报告(英文版)(14页).pdf》由会员分享,可在线阅读,更多相关《Razorfish:2023年AI在市场营销、客户服务和艺术领域的变革性潜力研究报告(英文版)(14页).pdf(14页珍藏版)》请在三个皮匠报告上搜索。
1、WE PROMISEAI DIDNTWRITE THIS:Opportunities and Considerationsof Large-Language Modelsfor MarketersAdam Buhler SVP,Creative Technology DigitasCristina Lawrence EVP,Head of Consumer&Content Experience RazorfishJerry Lawrence SVP,Consumer&Content Experience Strategy RazorfishAndrew McKernan GVP,Consume
2、r&Content Experience Strategy Razorfish Razorfish and Digitas,Part of Publicis Groupe February 2023Generative AI vendors and solutions are changing every day at a rate far outstripping Moores Law.This white paper includes references and integrations up through the time of writing in early February 2
3、023.1 AIS“BLACK SWAN”MOMENT What is the real power behind Internet communities?The not-particularly-secret sauce that drives marketing,the pathway that informs our consumer choices,the variable that makes content relatable and desirable?The answer,the hive mind of the 21st century marketing cognosce
4、nti has determined,is personalization.Successful commerce is all about YOU.When it comes to the personal,marketing is simply taking a cue from the wider culture,and the web3 revolution in particular.Digital natives are increasingly asserting sovereignty over their data,assets,and experiences online.
5、The trend is toward online tools and systems that empower individuals and enable horizontal community building whether leveraging generative AI,web3&blockchain,or rapidly growing communications platforms like Discord.Conscientious marketers are able to tap into these emerging technologies,creating n
6、ew kinds of relationships between brands and their audiences.The closer marketers can get to mind-reading an audience and then speaking directly to their individual preferences and dreams in the context of a single,fleeting moment,the stronger their pitch.Just ask Mark Zuckerberg Meta saw$10 billion
7、 of its market capitalization evaporate in 2022 when Apples new iOS privacy features greatly restricted Facebooks app tracking and profiling abilities.In this document,we address the urgent need for brands and their agencies to understand the new AI wave,with a focus on the breakthroughs in personal
8、ization we believe will define the coming decades of digital.At the end of 2022,the field of Artificial Intelligence drew frenzied attention due to a sequence of astonishing firsts.Now,in 2023,AI is poised for its“iPhone moment.”New use cases,new tools,new impact on society the digital space is alre
9、ady being flooded with AI-related think pieces and amazement at the rapid compression of time between epochal bursts of progress.There are plenty of reasons to be swept up in the breathless excitement of AIs nascent Golden Age:it can now write software code from nothing more than a verbal descriptio
10、n of functionality;it can slash the effort required for creative and business tasks;its now good enough at appearing sentient that it often barely matters whether it really is.All that being said,theres also a simpler,more approachable way of thinking about the value of generative AI:it allows the c
11、rafting of dynamic content,communications,and services that are hyper-personalized to an individual,human level.A doorway to digital experiences that understand you and respond in real time,in a persona tuned precisely to your desire.Generative AI is showing a plausible path to scaled,automated cons
12、umer experiences that are not merely substitutes for human interaction but,under the right conditions,can actually improve upon it.2Now,in 2023,AI is poised for its“iPhone moment.”New use cases,new tools,new impact on society the digital space is already being flooded with AI-related think pieces an
13、d amazement at the rapid compression of time between epochal bursts of progress.1A HOWD WE GET HERE?1B WHATS GETTING DISRUPTED RIGHT NOWIn 2022,the AI space went volcanic,intensifying as the year progressed.Lets chart the eruptions:Web Search Googles prominence as the doorway to the web swallowing u
14、p 92%of all search engine requests as of November 2022 has seemed unshakeable,a basic fact of Internet life.But the Google search experience is not immune to revolution.Google does deploy AI across its search product in several ways:“Extractive summarization”digs into the content of a set of related
15、 articles and comes back with the one paragraph that best encapsulates them all.Google calls these“featured snippets,”and displays them prominently when confidence is high.Auto-complete uses AI to calculate the next-best word when a user types in the search box.Since 2015,Google has deployed RankBra
16、in and other AI tools to supplement its algorithms for understanding natural language,sorting,and ranking.However,Google search has its drawbacks:The ads and paid placements are obstructive.Data is fractured across multiple links.A user must decide which of those links to pursue and spend additional
17、 reading time finding the relevant parts in the linked articles.3May 2020:GPT-3 launch;limited cultural impact due to waitlist requirementsMay 2021:Google LaMDAApril 2022:DALL-E 2November 2022:Lensa allowed everyday social users to train generative AI models to create sci-fi and fantasy versions of
18、themselves,spurring an exponential increase in digital artists claims of intellectual property theft January-February 2023:Google BARD updates and Microsoft Bing integrations of OpenAI January 2021:DALL-E launchJuly 2021:GitHub CoPilotAugust 2022:Stable DiffusionDecember 2022:ChatGPTBreathlessly ant
19、icipated sometime inQ1 2023:GPT-44When OpenAIs ChatGPT was released in late November 2022,rocketing to 100 million users by January(the most viral app in history,according to a UBS study),an urgent buzz arose as its implications unfolded.Could it be,after so many years of unquestioned dominance,that
20、 Google finally faced a real threat?Users found that ChatGPT responded to search and knowledge inquiries with clear,direct,coherent answers no additional cruft,no picking among dozens of links,no hassle.This was a better experience!The search throne may not topple today,but for the first time in dec
21、ades,its possible to imagine a new paradigm.So whats the catch?ChatGPTs AI model was trained on a truly enormous corpus of text(570 gigabytes),text which was current up through 2021.Requests for factual data from after 2021,therefore,will induce ChatGPT to“hallucinate,”and confidently offer up loose
22、 statistical inferences as rock-solid truths.In general,what these large-language AI models are doing is probabilistically determining the next best piece of vocabulary,not the underlying knowledge.The fact that the model is very often correct on the knowledge is remarkable,but as the saying goes,“I
23、f you have a self-driving car that is 99.9999%autonomous,you still dont have a self-driving car.”This is the current Achilles heel for AI-powered conversational search,and some of the best minds in the space are testing approaches that combine conversational interfaces with error-checking and live d
24、ata mining.As Yoav Shoham,founder of AI startup AI21 Labs,declared in December,“The future of generative AI will be in its precision.”In fact,there was much heartburn within Google when OpenAI started receiving accolades for ChatGPT in November.(Its“code red”for the search giant,according to the New
25、 York Times.)Google has similarly powerful AI technology(LaMDA,which set off a firestorm in July when an engineer was fired for insisting the AI was sentient),but had been reluctant to release it because of concerns over accuracy and the potential for sincere-sounding misinformation.Googles hesitanc
26、e at risking their establishment position is the classic innovators dilemma.As of this writing,Microsoft is making moves to outflank Google by incorporating GPT-based tech into their own Bing search engine and Office suite of products.It could launch as early as March,according to industry insiders
27、at The Information.The race is on!Meanwhile,commenters like Ryan Broderick,writing the substack Garbage Day,say reducing the power of AI down to search engine optimization is “the most depressing by-product of all.Imagine building an artificial intelligence that can create images and have a conversa
28、tion with you and using it to automate the most boring and noxious parts of the Internet:Taboola chumboxes and the second and third pages of Google search results.Imagine making real the stuff of science-fiction stories and using it to create an ad network.How lame!”If you have a self-driving car th
29、at is 99.9999%autonomous,you still dont have a self-driving car.Written Content Shockwaves are rippling through the world of academia,as a host of AI-based tools have many people questioning the continued relevance of written essays and long-form writing tests.AI can now create highly convincing doc
30、uments on any subject in any style and voice with vast implications for Internet culture.Recognizing authentic humans online,already a swirling challenge due to bots and bad actors,is about to become much more difficult.On January 4,the New York City education board blocked access to ChatGPT on scho
31、ol networks and devices,citing“its ability to churn out pitch-perfect essay responses to prompts spanning a wide range of subjects.”The Art World In 2022,the art world contended with a new category of AI tools,able to reach into a“collective digital unconscious”and pull gorgeous,fully rendered image
32、ry out of the ether.Immediately controversial among artists,this development raised profound philosophical issues when an AI-generated image won a statewide annual art competition in Colorado in the fall.To focus on just three of the issues being raised by generative AI art:5Copyright and IPBecause
33、these AI systems are trained on millions of human-created visuals,its possible that a generated response may resemble part or all of an artists specific piece.Additionally,users can include the name of a specific artist when prompting the system,resulting in work that exactly matches the artists sty
34、le.This is truly a Pandoras box for visual content.Deprecation of artists laborClipart and stock-image repositories like Shutterstock have announced partnerships with OpenAI and other AI start-ups.Bias“Garbage in,garbage out.”The results outputted by AI systems will only reflect the data on which th
35、eyre trained.Ask one of todays popular AI image generators to render a princess,for example,and the results will likely be specific to a Western cultural ideal.(Some of the platforms are now building in extra layers of technology to induce more diverse palettes.)61C APPLICATIONS YOU CAN EXPLORE RIGH
36、T NOWCustomer Service and Support Over the past decade,Facebook and others spent much time and treasure attempting to harness automated commerce through chatbot technology.These tools were simplistic and clunky,however,and were far more attractive in the minds of brands than they were to the actual
37、consumers who used them.First-wave chatbots were known for constricted conversational paths,brittle user experiences,and a burning feeling of“just let me talkto a human!”An entirely new game is afoot,however,with the flexibility and power of large-language-model AI.For example,the popular“Do Not Pay
38、,”built on IBMs Watson AI platform,uses sophisticated content generation to help users contest parking tickets,obtain refunds on flight tickets,and to help obtain U.S.green cards.Though the service has its critics,Do Not Pay has a current valuation of$210 million,according to its CEO.1.Dynamically g
39、enerated executive summaries.In some fields,such as stock trading and general finance,practitioners sink or swim based on their competitive knowledge edge,and the ability to consume and interpret vast swathes of data quickly.An AI technique called“extractive summarization”solves this challenge by in
40、gesting content from video,PDF,text,and audio online,analyzing it,and then spitting out only the 35 sentences that best represent the collective wisdom therein.10 hours worth of podcast listening can now boil down to a page of focused insight automatically.This knowledge mining was once expensive an
41、d reserved mainly for financial institutions that could afford it,through high-end start-ups like Agolo.With the renaissance of AI in 2023,however,anyonecan use inexpensive variations of these tools(for example,Bearly.AI).2.Hyper-personalized virtual assistants.Imagine an implementation of ChatGPT t
42、hat serves as a personalized shopping buddy,tour guide,cruise line scheduler,flight scheduler,or interpreter of a companys HR documentation.AI has long been deployed for interest targeting,but the next level is an interface that adapts and learns as you go,able to search,recommend,and even purchaseo
43、n your behalf.3.Streamlining of digital production.Image-generation tools can be instructed to dynamically create banner ads,shaped and laid out to fit desired aspect ratios,on demand.More importantly,the content of banners and short videos can be tailored on the fly,hitting each individual viewer r
44、ight in their comfort zone.Generative AI is rapidly evolving from still imagery capabilities to real-time video creation,an area in which we expect to see breathtaking progress in 2023.72 USE CASES/GETTING STARTEDWe would do ourselves,and you,a disservice to attempt to list,in any kind of exhaustive
45、 manner,the full set of AI vendors,solutions,and thought experiments currently in development.In the words of Pioneer Square Labs and PSL Ventures co-founder Ben Gilbert,“Were in an App Store moment.”That is,the vendors and use cases for generative AI are popping up faster than,well faster than Chat
46、GPT can currently respond to your queries.Image:An already outdated list of AI vendors across the landscape,per Antler.co8WHATS CHANGINGCopy and Editorial DevelopmentImage and Music Production Voiceover Production“Transhuman”Experiments in ProductivityOpen AIs conversational ChatGPT product is still
47、 just in beta,but the explosive potential of AI across industries is already creating a“cottage industry”of builders seeking to create platforms for brands to engage their audiences.But its not all positives.Meta took down Galactica,an open-source language model trained on scientific papers,just day
48、s after launching it because it was already“spewing misinformation.”Stability AI,the company behind Stable Diffusion,has raised more than$100M in funding for the production of open-source,machine-generated music and images.Recent updates to Stable Diffusion have improved image outputs,and include at
49、tempts to limit copyright infringement from prompts“in the style of”specific artists.Several vendors,including Microsofts recently announced VALL-E,are showcasing the evolution of voice simulation in terms of“human-like”fidelity.(Users on 4chan are already experimenting with ways to impersonate cele
50、brity hate-speech deepfakes.)Google and iCAD are using AI to provide more robust cancer detection in reading mammograms for malignant growths.At the same time,Google is actively developing an AI imagery deciphering tool to read doctors prescription notes.Though just in prototype,the solution could h
51、ave global implications,particularly for Google in India,where the bot could help decipher prescriptions in more than 100 Indian languages.OUR TAKEAWAYCURRENT USE CASESCopywriters and social media managers can sleep easy for the moment.The raw outputs generated by LLMs still require careful scrutiny
52、 for accuracy,brand safety,and compliance;theres no“set it and forget it”dynamic-content-at-scale solution yet.Several ongoing legal claims,including a class-action lawsuit brought by artists who claim the models are trained on IP,could cool brand interest in engaging.This could offer increased acce
53、ssibility for video content across the web.Looking further out,it may even have ramifications for the way brands and agencies consider negotiations with,and prioritized uses of,human talent.Rather than considering generative AI as an either-or,brands could find opportunities in scaling sales and cus
54、tomer service processes as models help surface FAQ,troubleshooting,and other standardized solutions to client success managers and frontline employees in real time.92A CONSIDERATIONS BEFORE DEPLOYMENTFor now,ChatGPT isnt“hooked up”to the real Internet,and its beta is based on data from 2021 and earl
55、ier.However,theres a potential that as ChatGPT and similarly evolving text-generative AI services become mainstream,people will come to expect immediate,high-touch experiences.This technology could give brands the ability to communicate with customers in highly personalized ways at scale and the bar
56、 for how brands deliver experiences will be raised as a result.Conversely,no human employee has ever followed a training manual to the letter,and these deviations may be an important part of a selling process or mitigation of unconscious bias within outdated training materials.As Meta discovered all
57、 too quickly with Galactica,an open-source language model trained on scientific papers,improper or imprecise data sets,or a lack of sufficient guardrails around user inputs can lead to an AI“spewing misinformation”just days after its launch.At best,this results in a subpar user experience,but,more l
58、ikely,major brand reputational risk or consumer backlash could occur.A second use case:Consider training an image-generating AI on brand image libraries.What guardrails need to exist to protect the brands IP?Is it worth training an AI on years of old makes and models of vehicles for it to inspire ne
59、ar-market drivers with potential concept car designs?Or would image generation be better served for backgrounds that designers can use for in-painting“real”products and services?From prompt generation and refinement to post-production quality control and creative direction,theres still a strong requ
60、irement for human oversight over any generated content.As front-end engagements currently in beta open up,marketers should continually scan the space in order to understand opportunities for streamlining,as well as the risks involved,especially as it pertains to brand safety,copyright,and intellectu
61、al capital.It will be increasingly difficult to identify deepfakes,leading to brand risks,as both marketers and their audiences struggle to know whats real and what isnt.Our POV on detecting such inauthentic visual content,such as photographs of politicians appearing to do things they never actually
62、 did(but in perfect visual fidelity),is this:no tool can conceivably determine whether an image or video has been faked simply by analyzing it.This is an AI arms race that can never be won.As soon as a tool is created that can identify telltale signs of alteration,another tool comes along to defeat
63、it.Consider:Are there existing content libraries or training resources your brand is currently leveraging for coaching human talent?How accessible are these content libraries for ingestion by machine-learning tools?Just as content strategists have argued for search engine optimization and content mo
64、dels that can optimize an enterprises content libraries,auditing and preparing content for AI ingestion now could ameliorate the onboarding of AI consumer-facing front line workers in the near future.10What is to be done to prevent the destruction of consensus reality that AI image generation could
65、usher in?The solution lies at the other end of the process.Adobe and others are introducing new tools that embed asymmetric encryption into hardware like cameras and software like social media feeds.Asymmetric encryption allows an image or video to be“signed”digitally,with the creator able to prove
66、the contents provenance via public and private key technology.Adoption of this approach should slow the civilizational slide into perpetual unknowingness.CONSIDERATIONS BEFORE YOU USEINFORMATION ARCHITECTURE What company-specific materials could influence the“base”large-language model?Where is infor
67、mation housed in the organization?Is it machine-readable?Is it normalized(all in the same format)?What aspects of the brand library(style guides,operating manuals,etc.)require human interpretation for use?Are there clear distinctions between materials that could be consumer-facing and materials that
68、 should remain proprietary?CONTENT PRODUCTION How could generative AI processes supplement existing workstreams?Does anyone have the skills required to build and refine prompts and review generated content outputs?What brand elements can be manipulated and which product,design,or IP elements need to
69、 be protected?How can existing QA/UAT processes accommodate content generation?Does the organization need to develop new guidelines?LEGAL REVIEW How will the company manage compliance and copyright?Are legal teams currently positioned to review content at scale?What background briefings will legal t
70、eams need on production processes in order to approve generated content?What kind of documentation will the organization need to maintain to prove compliance with copyright and intellectual property law?Who or what entity takes on indemnification in case any infringement is found?DISTRIBUTION&GOVERN
71、ANCE How does the generated content get in front of priority audiences?How will the organization operationalize generative content?Who posts or executes a buy,and where?What audits,checkpoints,or qualitative feedback loops exist to ensure generated content is enabling preferred business outcomes?Who
72、 will store and review AI content,outputs,or responses for any issues with compliance?When should generative content expire?Are there existing procedures for taking down content at scale,or sunsetting accounts?113 THE DISRUPTION TO COME Can the categorical disruption that large-scale machine learnin
73、g promises catch up to the hype cycle?Microsoft,with a recent$10B infusion into OpenAI and a pledge to integrate AI services into its entire product suite,seems to think so.But there certainly is a clock ticking against the speculative hype.And the disappointment seems to hit harder the further it i
74、s juxtaposed against a fully extrapolated transhuman future,with everyday citizens merging with personal chatbots to explore the stars together.Within such a context,the hypothetical solar wind in the sails is diminished by a continued scrutiny into redundant,outdated,trivial generated outputs such
75、as the inaccurate outputs of AI-generated articles for CNET.Its this“dystopian”path that Buzzfeed CEO Jonah Peretti also denounces in his call for AI as“the new model for digital media that is more personalized,more creative,more dynamic.”In the immediate future say in Austin,Texas,in March,or in Ca
76、nnes,France,in July we can expect any number of experiments from brands and agencies trying to catch hold of the current AI hype.Its doubtful that any of these early experiments will fully exploit the vast potential for personalized,dynamic content at scale.Its in the next wave of technological adva
77、ncements,including the development of what OpenAI CEO Sam Altman has called a“middle layer”offering of UI and toolkits built on the back of existing large-language models,that disruption will emerge.Consider a recipe generator that could predict the latest food or beverage trend or that could ingest
78、 real-time local grocer sales and coupons.Consider an evolution of the current experiments in robot pets and beyond for meeting emotional and functional needs in eldercare.Consider the impact on white-glove customer service offerings that can deliver truly personalized,bespoke services.Or to take th
79、e glass half-empty:consider recommendation engine,review site,and travel aggregators.Can you truly trust that product review?How many of that influencers followers are not just bought but entirely fake chatbots?Are you sure the beach vacation you just booked actually has beach views like the ones dy
80、namically generated on the reservation detail pages?Are you participating in a community of like-minded peers,or have you entered an echo chamber of Turing Test-proof chatbots who will happily debate you for hours on end in order to drive up your in-app engagement rate?Are you hounded by ads maybe e
81、ven before or around your machine-generated queries that are increasingly personalized,even with the death of cookies,because the machine knows you so well?Successful commerce on the Internet is all about you,and the ways algorithmic prototypes,from social newsfeeds to programmatic buys,have been ab
82、le to curate personalized recommendations for you.The power of these recommendations is in the predictability of consumer habits.But soon,for the first time,you will have access in your hands to the same tools that enterprise marketing solutions have offered for years.So will everybody else.And ther
83、es no telling how digital marketing and the world will transform.12Sam Altman and Reid Hoffman,“AI for the Next Era,”Greymatter podcast,Sept.2022.https:/ Bishop,“Were having an App Store moment:Generative AI and ChatGPT top list of key technologies for 2023,”GeekWire,Dec.17,2022.https:/ Broderick,“T
84、he AI has learned how to tell you a celebritys net worth,”GarbageDay,Jan.4,2023.https:/www.garbageday.email/p/the-ai-has-learned-how-to-tell-youCoalition for Content Provenance and Authenticity,accessed Jan.2023,https:/c2pa.org/Oliver Darcy,“BuzzFeeds CEO says AI could user in a new model for digita
85、l media,but warns against a dystopian path,”CNN,Jan.26,2023.https:/ Edwards,“Microsofts new AI can simulate anyones voice within 3 seconds of audio,”Ars Technica,Jan.9,2023.https:/ Elsen-Rooney,“NYC education department blocks ChatGPT on school devices,networks,”Chalkbeat New York,Jan.3,2023.https:/
86、ny.chalkbeat.org/2023/1/3/23537987/nyc-schools-ban-chatgpt-writing-artificial-intelligence?_amp=trueKatie Engelhart,“What Robots Can and Cant Do for the Old and Lonely,”The New Yorker,May 31,2021.https:/ Foley,“Stable Diffusion has cleaned up its AI art generator but people still arent happy,”Creati
87、ve Bloq,Nov.29,2022.https:/ Forsyth,“Mapping the generative AI landscape,”Antler.co company blog,Dec.20,2022.https:/www.antler.co/blog/generative-aiMatt Growcoot,“AI image generator Stable Diffusion releases controversial new version,”PetaPixel,Nov.28,2022.https:/ Hecht,“Self-driving vehicles:Many c
88、hallenges remain for autonomous navigation,”Laser Focus World,April 14,2020.https:/ Hu,“ChatGPT sets record for fastest-growing user base analyst note,”Reuters,Feb.2,2023.https:/ Kovach,“Mark Zuckerbergs metaverse business lost more than$10B last year,and the losses keep growing,”CNBC,February 2,202
89、2.https:/ Lam,artist Instagram account,accessed January 2023.https:/ Leffer,“CNET is reviewing the accuracy of all its AI-written articles after multiple major corrections,”Gizmodo,Jan.17,2023.https:/ Leffer,“Google takes on doctors terrible handwriting,”Gizmodo,Dec.19,2022.https:/ Morrison,“What Mi
90、crosoft gets from betting billions on the maker of chatGPT,”Vox,Jan.23,2023.https:/ Naresh,“Google to use AI to translate your doctors scribbled prescription,”SamMobile,Dec.20,2022.https:/ Norem,“Nvida CEO calls ChatGPT the iPhone moment for AI,”ExtremeTech,February 14,2023.https:/ Park,“Googles AI
91、will now be used in mammograms,”Time,Nov.28,2022.https:/ Pichai,“An important next step on our AI Journey,”Google company blog,Feb.6,2023.https:/blog.google/technology/ai/bard-google-ai-search-updates/amp/WORKS CITED13WORKS CITED(CONT.)Kevin Roose,“An A.I.-generated picture won an art prize.Artists
92、arent happy,”New York Times,Sept.2,2022.https:/ Ryan,“Meta trained an AI on 48M science papers.It was shut down after 2 days,“CNET,Nov.20,2022.https:/ Schechner,“Microsoft plans to build OpenAI,chatGPT features into all products,”Wall Street Journal,Jan.17,2023.https:/ Jaime Sevilla,Lennart Heim,Ans
93、on Ho,Tamay Besiroglu,Marius Hobbhahn,Pablo Villalobos,“Compute trends across three eras of machine learning,”Arxiv.org,Feb.11,2022.https:/arxiv.org/abs/2202.05924Manish Singh,“Google can now read your doctors bad handwriting,”TechCrunch,Dec.19,2022.https:/ Stokel-Walker,“ChatGPT is on fire,and a co
94、ttage industry of bot builders is exploding alongside it,”The Information,Dec.16,2022.https:/ Tewson,“The worlds first robot lawyer isnt a lawyer,and Im not sure its even a robot,”TechDirt,Jan.24,2023.https:/ Vincent,“4chan users embrace AI voice clone tool to generate celebrity hatespeech,”TheVerge
95、,Jan.31,2023.https:/ James Vincent,“Googles text-to-image AI model Imagen is getting its first(very limited)public outing,”TheVerge,Nov.2,2022.https:/ Vincent,“Microsoft announces new Bing and Edge browser powered by upgraded ChatGPT AI,”TheVerge,Feb.7,2023.https:/ Wiggers,“Stability AI,the startup
96、behind Stable Diffusion,raises$101M,”TechCrunch,Oct.17,2022.https:/ Yokobori,“Hey everyone!I know a lot of people have been posting their Lensa/other AI portraits lately.I would like to encourage you not to do so or,better yet,not to use the service.These AI seem like harmless fun but they are predatory and intend to replace artists.(1/9)”Twitter,Dec.2,2022.https:/ YOU.