《CB Insights:2024生成式人工智能圣經:大模型顛覆行業終極指南(英文版)(112頁).pdf》由會員分享,可在線閱讀,更多相關《CB Insights:2024生成式人工智能圣經:大模型顛覆行業終極指南(英文版)(112頁).pdf(122頁珍藏版)》請在三個皮匠報告上搜索。
1、Generative AI BibleThe ultimate guide to genAI disruptionPablo Xavier via BuzzFeed News2Unleash the generative AI revolutionDive into a treasure trove of generative AI companies,investors,and market insights.Elevate your decisions with exclusive software buyer interviews to understand product pricin
2、g,CSAT,and post-deployment experience.CB Insights Your Gateway to Generative AISign up for access3Generative adversarial networksCommercial deepfakes,AI voice spoofingSynthetic data,self-supervised learningTransformers,multilingual modelsCode generation,deepfake detection,multimodal AICB Insights201
3、9 Trends ReportCB Insights2020 Trends ReportCB Insights2020 Trends ReportCB Insights2021 Trends ReportCB Insights2022 Trends Report12345Generative AI is a theme CB Insights covered before the hype.Weve consistently been ahead of the curve201920202020202120224CB Insights helps the worlds leading comp
4、anies understand everything they need to know about disruptive technologies find out more about why our customers love us here.And were staying ahead today so you can tooUnderstand 25+generative AI marketsUnderstand 25+generative AI marketsLarge language model developersCustomer support operationsTe
5、xt generationProtein&drug designVoice synthesis&cloning+moreDiscover 400+Discover 400+genAIgenAI vendorsvendorsTracked in our analyst-curated Expert CollectionStay on top of the landscape with research&transcriptsStay on top of the landscape with research&transcripts20235ContentsGenerative AI BibleT
6、he generative AI boomAnd now suddenly is accelerating631196376488GraduallyThen suddenly1.Race to dominate genAI infrastructure2.Cross-industry applications face pressure from large players3.Opportunity in vertical genAISo where is generative AI headed?Promising companies to watchHealthcare&life scie
7、ncesFinancial services&insuranceRetail919Hundreds of startups pile into genAIFunding soars as investors flockBig tech is all in and ready to fight384150936The generative AI boom7“How did you go bankrupt?”ERNEST HEMINGWAY,THE SUN ALSO RISES8“Two ways Gradually and then suddenly”ERNEST HEMINGWAY,THE S
8、UN ALSO RISES9GraduallyGenerative AI has been in the works for years201410How did we get here?A recent timeline of select events in the development of generative AI1Generative adversarial networks(GANs)introduced by Ian Goodfellow20142WaveNet and audio generation introduced by DeepMind20164Google AI
9、 releases BERT,a leap in the ability of machines to understand context in language20185OpenAI releases GPT-2,gaining attention for text generation capabilities20198OpenAI releases text-to-image model DALL-E202110OpenAI launches GPT-3.5-based chatbot ChatGPT,unleashing genAI boom2022Text-to-image mod
10、els from Google,Midjourney,Stability AI,and OpenAI proliferate202293New neural network architecture called the“Transformer”introduced by Google researchers2017OpenAI releases GPT-3,accelerating interest in language models62020*Generative AI is artificial intelligence that can generate new content(te
11、xt,code,images,audio,etc.).“Deepfakes”become widely known7202011Image source:Large Scale GAN Training for High Fidelity Natural Image Synthesis*”AI versus AI”:A breakthrough where two neural networks try to outsmart each other,creating and refining synthetic outputs.GANs tap into the idea of“AI vers
12、us AI”advancing image generation dramaticallyImages from 2018 paper where DeepMind researchers trained GANs on a large-scale dataset to create“BigGANs”20141201820202022201612Image source:Google DeepMindWaveNet produces synthetic audio,showcasing the potential of generative models beyond images 20142
13、018202020222016213The Transformer architecture can better understand and generate human language,paving the way for further R&D 6 authors of the seminal research paper have gone on to raise$1.7B across 5 AI companies*175+661%Source:CB Insights 6 authors of a seminal research paper by Google*As of 10
14、/12/202320142018202020222016314Google AI releases BERT,a leap in the ability of machines to understand context in languageImage source:Google201420182020202220164The AI language model predicts a word based on not only the preceding words,but also the succeeding ones(bidirectional understanding of co
15、ntext).BERT is deeply bidirectional,OpenAI GPT is unidirectional,and ELMo is shallowly bidirectional.15OpenAI incorporates the Transformer architecture into its language modelsImage source:OpenAIGPT-1 June 2018GPT-2 February 201920142018202020222016516OpenAIs GPT-3 is a major leap forward,showcasing
16、 ability to generate code,jokes,and moreImage source:NYT,MIT Technology Review201420182020202220166November 2020July 202017Deepfakes go mainstream,highlighting power and pitfalls of video generation Image source:YouTube,TikTok via ABC*Deepfakes refer to synthetic media where a person in an existing
17、image or video is replaced with someone elses likeness using neural networks.20142018202020222016718GPT-3 is the foundation for DALL-E,which can generate images from text descriptionsImage source:OpenAI201420182020202220168192022Then suddenlyGenAI goes from experiment to everywhere20Models get bigge
18、rImage source:The Economist21and better,beating human performance benchmarksImage source:Science22Text-to-image generators take the internet by stormImage source:Imagen(Google),Midjourney,DALL-E 1 vs.DALL-E 2(OpenAI),Stable Diffusion 23Generative AI startups raise major funding to fuel growthDeals w
19、orth$100M+to generative AI startups in 2022Source:CB Insights Advanced Search-Deals24AI coding assistant GitHub Copilot becomes widely availableImage source:GitHub25ChatGPT goes viral,getting to 1M users in 5 days and 100M in 2 months unleashing genAI boomTime to 1M users for select platforms/apps f
20、rom launchSource:Media mentions*App downloads2 years10 months7 months5 months2.5 months*5 days051015202530TwitterFacebookDropboxSpotifyInstagramChatGPT26Source:CB Insights Advanced Search-Earnings transcripts1010000284461,5462,08105001,0001,5002,0002,500Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3202120222023Almost overn
21、ight,exec interest in generative AI skyrockets and companies feel pressured to reactEarnings call mentions of“generative AI”(as of 9/30/2023)27Microsoft invests$10B into ChatGPT-maker OpenAI in Q123,propelling genAI funding to new heightsDisclosed equity funding&deals to generative AI companies(as o
22、f 9/30/2023)$1.7B$0.7B$5.3B$3.2BFunding$17.4B64103154168050100150200250$0.0B$2.0B$4.0B$6.0B$8.0B$10.0B$12.0B$14.0B$16.0B20192020202120222023 YTD$10.0B$2.0B$1.0BFundingDealsfunding64103154168Deals170050100150200250020004000600080001000012000140001600020192020202120222023 YTDSource:CB Insights What ar
23、e customers saying about generative AI startups?28Microsoft debuts AI-powered Bing,running on a new OpenAI large language model(LLM)Image source:Microsoft*A large language model is a deep learning algorithm that analyzes and produces text by learning from extensive language data.29Meta introduces Ll
24、ama,an open-source language modelImage source:Meta30Google sounds alarm bells and releases Bard chatbotSource:CB Insights Advanced Search-Earnings transcripts;Google*Reflects quarter call occurredAlphabet CEO Sundar PichaiQ123 earnings call On the AI side,it is a really exciting time.I think weve be
25、en investing for a while,and its clear that the market is readyObviously,we need to make sure were iterating in public,these models will keep getting better,so the field is fast changing.The serving costs will need to be improved.So I view it as very,very early days,but we are committed toactually b
26、ringing direct LLM experiences in Search,making APIs available for developers and enterprises and learn from there and iterate like weve always done.31GPT-4 becomes OpenAIs most powerful model yet,crushing human examsImage source:OpenAI32Reddit,after providing years of free training data for AI syst
27、ems,plans to charge for access to its contentImage source:Reddit33Chegg blames ChatGPT for declining revenue,sees its share price tank,and,in response,pivots to build LLMsSource:CB Insights Chegg company profile,CheggChegg attributes declining growth to ChatGPT34Nvidia enters the$1T market cap club
28、as demand for GPUs used in genAI sends its revenue soaringUS companies to reach$1T+market cap(as of 10/30/2023)$1.2T$1.1T$1.2T$1.9T$2.0T$2.7T$3.1T$0.6T$0.8T$1.0T$1.4T$1.6T$2.5T$2.7T$0.0T$0.5T$1.0T$1.5T$2.0T$2.5T$3.0T$3.5TAll-time highCurrent market cap*GPUs=graphics processing units,which are used t
29、o run intensive AI applications.35Stack Overflow sees declining traffic and lays off employees amid AI coding boomImage source:SimilarWeb,The Verge36Apple CEO Tim CookQ323 earnings callEven Apple scrambles as it works on its own LLM called Ajax and is on course to spend$1B a year on genAI pushSource
30、:Bloomberg;CB Insights Advanced Search-Earnings transcripts*Reflects quarter call occurredIf you take a step back,we view AI and machine learning as core fundamental technologies that are integral to virtually every product that we buildAnd of course,weve been doing research across a wide range of A
31、I technologies,including generative AI,for years.Were going to continue investing and innovating and responsibly advancing our products with these technologies with the goal of enriching peoples livesAnd as you know,we tend to announce things as they come to market,and thats our MO,and Id like to st
32、ick to that.37And now suddenly is acceleratingAmbitious&flush with cash,young companies and big tech players are all rushing into this next platform shift38Hundreds of startups pile into genAI39Source:CB Insights Generative AI Market MapCommercial genAIapplications are proliferatingGenerative AI Mar
33、ket MapExplore the full map40Source:CB Insights Generative AI Market Map300+vendors have emerged across:GENAI LANDSCAPE LAYERS Cross-industry generative applications(visual media,text generation,code generation,etc.)Industry-specific generative applications(healthcare,finance,etc.)Generative AI infr
34、astructure(foundational models,vector databases,etc.)41Funding soars as investors flock42Source:CB Insights The state of generative AI in 7 charts;Deals StoryAs investors look to ride the generative AI wave,funding soars in 2023Disclosed equity funding&deals(as of 9/30/2023)$1.7B$0.7B$5.3B$3.2BFundi
35、ng$17.4B64103154168Deals170020406080100120140160180$0.0B$2.0B$4.0B$6.0B$8.0B$10.0B$12.0B$14.0B$16.0B$18.0B$20.0B20192020202120222023 YTD43Source:CB Insights The state of generative AI in 7 chartsGenerative AI infrastructure attracts the bulk of funding,due to the capital-intensive nature of developi
36、ng LLMsDisclosed equity funding&deals to generative AI categories,from Q422 to Q323$0.8B$5.5B$11.6B48 deals129 deals30 dealsGenerative AI infrastructureIndustry-specific generative applicationsCross-industry generative applicationsFundingDeals44Source:CB Insights Advanced Search-DealsA record number
37、 of$100M+mega-rounds drive funding surge,with money primarily going to infrastructure layerDisclosed$100M+genAI equity deals(as of 9/30/2023)4198Mega-rounds20051015202500.10.20.30.40.50.60.70.80.9120192020202120222023 YTD45Source:CB Insights Alongside big deals,generative AI is minting unicorns left
38、 and rightNew unicorns($1B+valuation),Q123 Q323CompanyValuationCountry1$4.4BUnited States2$2.2BCanada3$1.5BUnited States4$1.4BIsrael5$1.2BUnited States6$1.0BUnited States6$1.0BUnited States6$1.0BUnited Kingdom6$1.0BUnited States6$1.0BUnited States6$1.0BChina46Source:CB Insights State of AI Q323 Repo
39、rtOut of the 16 new AI unicorns in 2023 so far,11 are genAI companiesNew AI unicorns($1B+valuation)333421241614141426574051015202530Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q32020202120222023AI unicorns2023 genAI unicorns35347Source:CB Insights Advanced Search-Deals Biggest M&A deal of 2023 lands in infrastructu
40、re,but the rest of M&A reflects push at the industry/application levelGenerative AI M&A exits,Q123 Q323CompanyRound Valuation AcquirerGenAI Focus AreaCountry1MosaicML$1.3B DatabricksInfrastructureUnited States2InstaDeep$125M BioNTechHealthcare&life sciencesUnited Kingdom3Casetext$650M Thomson Reuter
41、sLegalUnited States4Light Year AI$234M MeituanInfrastructureChina5Valence$47M RecursionHealthcare&life sciencesCanadaNeevaN/A SnowflakeEnterprise techUnited StatesThankfulN/A GladlyEnterprise techUnited StatesFigN/A Amazon Web ServicesEnterprise techUnited StatesCodigaN/A DatadogEnterprise techUnite
42、d States48Source:CB Insights Advanced Search-DealsThe US is poised to own the genAI boom:2x more deals in the US than the rest of the world combinedDisclosed generative AI equity deals to startups by company HQ,Q422 Q32364143020406080100120140160Rest of worldUnited States4939%13%12%1%19%16%Seed,pre-
43、seed,angel,&convertible noteSeries ASeries B&CSeries D+Not raised outside fundingOtherSource:CB Insights The state of generative AI in 7 charts*Other includes non-equity funding rounds and equity rounds not tied to specific stage of investment.But its still early days for genAI startups 71%are early
44、-stage or havent raised any fundingPercent of companies by latest disclosed round(as of 9/30/2023)17+66%50Big tech is all in and ready to fight51Source:CB Insights*Includes investments from M12 and Google Ventures*AWSGenerative AI is a new battleground for big tech,with overlapping alliances and com
45、mitments into the billionsGenerative AI companies with two or more big tech investors(as of 9/30/2023)1,646+370%Hugging FaceAdeptAI21 LabsAnthropicInflection AIInworld AIOpenAIRunwaySynthesiaTypefaceBig tech investors*CompanyIndicates where big tech invested52Source:CB Insights Advanced Search-Deals
46、*Big tech includes Amazon,Apple,Microsoft,Meta,Google,and NvidiaBig tech backs every top deal in 2023 so far,with other CVCs&corporates joining the action Top generative AI equity deals,Q123 Q3231,646+370%175+661%CompanyRound AmountRoundDateRound ValuationBig Tech InvestorsOther Select InvestorsCoun
47、try1OpenAI$10.0BCorporate Minority-III2023-01-23N/A MicrosoftUnited States2Inflection AI$1.3BSeries B2023-06-29$4.0B Microsoft,Nvidia Gates FrontierUnited States3Anthropic$1.25BCorporate Minority-V2023-09-25N/A AmazonUnited States4Anthropic$450MSeries C2023-05-23$4.1B GoogleMenlo Ventures,SK telecom
48、 ventures,Salesforce Ventures,Zoom VenturesUnited States5Anthropic$400MCorporate Minority2023-02-03$4.1B GoogleUnited States6Adept$350MSeries B2023-03-14$1.0B Microsoft,NvidiaGeneral Catalyst,Spark Capital,Atlassian Ventures,Workday Ventures,Greylock PartnersUnited States7Generate Biomedicines$273MS
49、eries C2023-09-14N/A NVenturesAbu Dhabi Investment Authority,Amgen,Fidelity Investments,Flagship Pioneering,MAPS CapitalUnited States8Cohere$270MSeries B2023-05-02$2.2B NvidiaInovia Capital,Index Ventures,Oracle,Salesforce Ventures,SentinelOneCanada9Hugging Face$235MSeries D2023-08-23$4.5BAmazon,Goo
50、gle Ventures,NVenturesSalesforce Ventures,AMD,IBM Ventures,Intel Capital,Qualcomm VenturesFrance10Imbue$200MSeries B2023-09-07$1.0B NvidiaAstera InstituteUnited States53Source:CB Insights Nvidia company profile investmentsNvidia increases its investment activity dramatically.Equity deals backed by N
51、vidia(as of 10/25/2023)2397Deals1702468101214161800.10.20.30.40.50.60.70.80.9120192020202120222023 YTD54Source:CB Insights Advanced Search-Deals becoming the most active investor in generative AITop generative AI investors by company count,Q123 Q323InvestorCompany Count Investor GroupCountry1Nvidia9
52、CorpUnited States2SV Angel7AngelUnited States3Salesforce Ventures6CVCUnited States3Index Ventures6VCUnited States3Andreessen Horowitz6VCUnited States6GV(Google Ventures)5CVCUnited States7Microsoft4CorpUnited States7Sequoia Capital4VCUnited States7Lightspeed Venture Partners4VCUnited States55Source:I
53、nflection AI;CB Insights Advanced Search-DealsThe dominant chipmaker is cashing in on the genAIcomputing boom,backing startups using its chipsNvidiaNvidia-backed equity deals to generative AI companies(as of 9/30/2023)backed equity deals to generative AI companies(as of 9/30/2023)56Source:CB Insight
54、s Advanced Search Earnings transcripts*Reflects quarter call occurredas demand for Nvidia GPUs far exceeds supplyIf youre trying to do the training of the models,then having the absolute latest,greatest GPU,the H100 from Nvidia right now,theres a lot of constraint in getting those chips.CEO Matthew
55、Prince,Q323 earnings callThere is a significant bottleneck in terms of Nvidias GPU chipsin short,the demand far exceeds the supply in the market.And that is not the situation for us as one company,it is a general situation for the industryWe had originally expected the revenue from AI to be shown in
56、 our financials in the third quarter and that is likely to be delayed due to the shortage of supply of GPU servers through the fourth quarter or even the first quarter of 2024.CEO Tao ZouQ323 earnings call 57Source:CB Insights Advanced Search-Deals*Includes deals backed by Microsofts venture arm M12
57、 Microsoft steps up its genAI investment activity beyond its$13B invested in OpenAIMicrosoftMicrosoft-backed equity deals to generative AI companies(as of 9/30/2023)backed equity deals to generative AI companies(as of 9/30/2023)1,646+370%58.betting that generative AI could tip the competitive scales
58、 in its favor for decades to comeSource:CB Insights Analyzing Microsofts generative AI strategy:How Microsoft is expanding past OpenAI to transform the way we workMicrosoft Investment Thesis Map Generative AI59Source:MicrosoftMicrosofts investments in genAI help reverse slowing Azure growth and cont
59、ribute to 3 percentage point bumpAzure and other cloud services revenue growth(year-over-year)by quarter59%62%59%47%48%50%50%51%50%46%46%40%35%31%27%26%29%0%10%20%30%40%50%60%70%Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1Q2Q3Q4Q1FY 2020FY 2021FY 2022FY 2023FY 202460Source:CB Insights Advanced Search-Deals*Includes i
60、nvestments by Google,Google Ventures,and Gradient VenturesAlongside its Bard chatbot,Google puts billions toward internal AI research&a range of AI startupsGoogle-backed equity deals to generative AI startups(as of 10/27/2023)1,646+370%61Source:CB Insights Anthropic company profile-fundingAmazon and
61、 Google commit billions to LLM developer Anthropic in battle with OpenAI-backer Microsoft 1,646+370%175+661%62Amazon launches$100M generative AI accelerator,looking to feed its cloud computing businessAWS Generative AI Accelerator Investment Thesis MapShare of genAIaccelerator cohortSource:CB Insigh
62、ts Where the AWS Generative AI accelerator is placing its bets across 7 industries63So where is genAI headed?1.Race to dominate genAI infrastructure2.Cross-industry applications face pressure from large players3.Opportunity in vertical genAIGenerative AI will soon be impossible to ignore as disrupti
63、ve applications spread64No clear winner yet in foundational models1.Race to dominate genAI infrastructure65Source:CB Insights The state of generative AI in 7 chartsHighest-valued genAI unicorns compete primarily at the infrastructure layerMost highly valued private generative AI companies(as of 09/3
64、0/2023)$1.4B$1.5B$1.5B$1.8B$2.1B$4.0B$4.1B$4.5B$7.3B$29.0B$0M$5,000M$10,000M$15,000M$20,000M$25,000M$30,000MCompany 10Company 9Company 8Company 7Company 6Company 5Company 4Company 3Company 2Company 166Source:CB Insights Generative AI large language model developers market report4 LLM developers Anth
65、ropic,Cohere,AI21 Labs,and Adept join the unicorn club in 2023Leading LLM developers by valuation(as of 09/30/2023)67Source:CB Insights Generative AI large language model developers market report-scorecardWhile OpenAI has clear lead,vendors are competing on multiple fronts to become the go-to model
66、developer1,646+370%Key KPIs for evaluationSafety&complianceAccuracy&qualityCustomizationPricing&deploymentToken limits68Source:CB Insights Advanced Search-Earnings transcripts*Reflects quarter call occurredExecutive interest in AI has surgedEnterprise demand for AI and accelerated computing is stron
67、g.We are seeing momentum in verticals such as automotive,financial services,healthcare,and telecom,where AI and accelerated computing are quickly becoming integral to customers innovation roadmaps andcompetitive positioning.CFO Colette KressQ223 earnings call69Source:CB Insights Software buyer inter
68、view transcripts and Analyst Briefing dataEnterprises are spending millions with LLM developersAnnual spend ranges displayed$100K-$500K$20K-$2M$25K-$300K$50K-$100K$15K-$800K$40K-$5M70Source:CB Insights MosaicML software buyer interview transcripts but reducing costs&time to train are key prioritiesM
69、osaic ML offers whats called programmatic optimization,which is not so much on the hardware side of things,but rather on the algorithmic side.Can you find ways of optimizing the time it takes to get to a certain performance bar?I think thats really what drove us to evaluate MosaicML.In fact,it was p
70、retty much the main tool out there right now that offers this.I dont think theres any other tool that really has this programmatic optimization layer.Senior Manager,Data Science,$1B+valuation technology companyOne of the things that were really trying to do is reduce the cost for a lot of our large
71、language models and trainingWhat we liked about Mosaic was it is a lot less expensive in terms of the training modelsRight now,I would project,just based on our usage,I think the initial spend was$15,000 per annum.I would expect next year to probably be$200,000,$250,000.Were a very large organizatio
72、n and theres been a lot of interest in MosaicML.Vice President,Innovation,Fortune Global 500 company71Source:CB Insights Cohere software buyer interview transcript;Anthropic software buyer interview transcriptSafety&compliance will be in focus for enterprise customers72Source:CB Insights The respons
73、ible AI market mapConcern around AI risks puts responsible AI solutions in the spotlightThese tools help enterprises build and deploy AI in an ethical and legal manner175+661%73Source:CB Insights Figures represent the latest disclosed revenue(based on company discussion or media sources).All revenue
74、s are full-year 2023 ARR projections,expect for MosaicML(reported ARR at time of acquisition in June 2023).As LLM developers burn through cash,focus will shift to customer adoption and revenueLatest disclosed or whisper revenue(as of 10/18/2023)$1,300M$200M$50M$40M$20M$10MCompany ACompany BCompany C
75、Company DCompany ECompany F74Source:CB Insights Figures represent the latest disclosed valuation divided by revenue(based on company discussion or media sources).All revenues are full-year 2023 ARR projections,except for MosaicML(reported ARR at time of acquisition in June 2023).Companies need to gr
76、ow into big valuations and will come under pressure to build real business modelsRevenue multiples(as of 10/18/2023)113x100 x65x28x22x21xCompany ACompany BCompany CCompany DCompany ECompany F75Source:CB Insights Anthropic,AI21 Labs,OpenAI software buyer transcripts Theres no winner yet in foundation
77、al modelsStrengths,I would say,the ethical considerations of privacy and bias,fairnesstheir model outperformed the other models,including GPT-3 and ChatGPTIn terms of weaknesses,the specificity of the model output and the interestingness of the model output I think that other weakness also was in te
78、rms of speed and efficiency,like latency,and once you ask a question,how long does it take to fully respond.Senior Manager of Data Science,Model-as-a-service platformWe were considering obviously OpenAI,Cohere,Anthropic,and deepsetTo be honest,the reason that we actually chose AI21 is because the in
79、terface is super easy to use for non-tech peopleI actually tested Cohere pretty extensivelyI think that their models are really strong and also for some of the creative work,I thought they were doing slightly even better than even OpenAI when it comes to creative stuff,like ad copy and marketing rel
80、ated tasks.Head of AI,$100M+funded technology startupThe first two things that came to my mind is,of course,OpenAI is still the market leader.That gives us the sense of comfort and we are confident on this market-leading productOn the flip side of the open-source platforms that I just mentioned,the
81、Hugging Face and the Llama 2,we dont have much faith and information about how they are going to deal with our data.Thats the key to the enterprise world.If we are not certain,then I would rather pass my roles to OpenAI.Cloud,Data&AI Lead,Fortune 500 company76Open-source AI movement gains steam1.Rac
82、e to dominate genAI infrastructure77Source:CB Insights Advanced Search-News mentions*The open-source approach to AI development is focused on making source code available for public use and allowing a community of developers to contribute to improving software.Need for AI model transparency and rapi
83、d innovation is fueling the open-source AI movementNews mentions of open-source AI and related terms(as of 9/30/2023)1,646+370%175+661%78Source:CB Insights Advanced Search-Earnings transcripts*Reflects quarter call occurredMeta is leveling the playing field with its open-source LLM,Llama 2175+661%No
84、tably,we recently announced a collaboration with Meta on Llama 2-based AI implementations on flagship smartphones and PCs that will enable developers to create new and exciting genAIapplications using the AI capabilities of Snapdragon platforms beginning in 2024.Qualcomm CEO Cristiano AmonQ323 earni
85、ngs callAzure AI is ushering in newborn-in-the-cloud,AI-first workloads with the best selection of frontier and open models,including Metas recent announcements supporting Llama on Azure and Windows,as well as OpenAI.Microsoft CEO Satya NadellaQ323 earnings callAI is the flavor of the day.And thanks
86、 to ChatGPTs great launch,everyone has discovered this potential.We can expect new changes by the day in the tech world.Just take yesterday,Metas launch of Llama 2,which will be available on Azure free of charge,including for commercial purposes.Publicis CEO Arthur SadounQ323 earnings call79Source:C
87、B Insights Generative AI Large language model developers market reportThe private market is split into open vs.closedDisclosed equity funding to LLM developers(as of 10/27/2023)Closed-source LLMsOpen-source LLMs*Some developers may offer open-source versions of their models but keep their core model
88、s proprietary*Excludes open-source developers that have not raised equity funding80Source:CB Insights OpenAI software buyer interview transcriptsOpenAI customers highlight potential cost-savings,customizability benefits of open-source models though OpenAI has the edge on performance&supportOpenAIs d
89、eveloper API and the developer experience is definitely the best.Its really managed,super clean APIs,well-documented,and has the most integrations.There was a big push toward using open-source models and then fine-tuning them with your own data.Thats still a big thing.Were actually evaluating that f
90、or some of the public data set stuff because its a lot cheaper,especially if you add in more huge amounts of data versus a giant OpenAI model.Partner,Early-stage VC firmMetas invention it may be cheaper than the OpenAI model because its open-source.Then we believe the performance of the Hugging Face
91、 and also the Llama 2 is also comparable to the OpenAI model.Maybe just a little bit weaker than that,but maybe the overallROI is quite a good deal.VP,Machine Learning,Fortune 500 company81Databricks pays$1.3B for MosaicML,which makes AI development tools and has its own open-source model,in June 20
92、23Source:CB Insights Databricks acquired Mosaic ML for$1.3B.How do the valuations of other generative AI companies compare?$20M in FY22 revenue(65x multiple)82Source:CB Insights The open-source AI development market map Growing number of vendors are developing open-source tools to help enterprises b
93、uild and deploy AI projectsOpen-source AI development market map175+661%83LLM infrastructure market grows rapidly1.Race to dominate genAI infrastructure84Source:CB Insights The large language model operations(LLMOps)market map*LLMOps refers to the end-to-end workflow that organizations employ to bui
94、ld,fine-tune,and deploy LLMs into production.Tech vendors supporting LLM operations are gaining traction with enterprisesLLMOps market map175+661%85Source:CB Insights Snorkel AI software buyer interview transcript,Fiddler AI software buyer interview transcriptExecs are buying infrastructure tools fo
95、r better training data,observing performance of models,and moreWe had,basically,messy data and we needed a better way of providing,of doing better training data for generative content.We have a lot of machine learning models that are fueling us here.In both cases,it was the desire to have,basically,
96、a higher level of data hygiene in our training data.Chief Product Officer,IT companyI led a small data science team that created a lot of models in production.As we scaled,our observability of our machine learning models in production was limited,and we felt blind to issues or ways to improve the mo
97、del once in production.We tried to build a solution in-house,which showed the difficulty of the challenge.Senior Manager,$10M+funded data analytics platform86Source:CB Insights LLM application development market reportNew vendors are emerging for LLM fine-tuning&customizationLeading LLM application
98、development vendors by disclosed equity funding(as of 10/30/2023)87Source:CB Insights Vector database market report*Vector databases provide enterprises with an easy way to store,search,and index unstructured data.Vector database startups,which make data more accessible for AI systems,raise record f
99、unding amid LLM boomDisclosed equity funding and deals(as of 9/30/2023)$44M$10M$109MFunding$176M0217Deals501234567$0M$20M$40M$60M$80M$100M$120M$140M$160M$180M$200M20192020202120222023 YTD882.Cross-industry applications face pressure from large players89Source:CB Insights Sourcegraph software buyer i
100、nterview transcript;SlashNext software buyer interview transcriptExecs are demanding their tech vendors keep up with genAI advances and opportunities From a technical perspective,I think this wave of ChatGPT and OpenAI large language models is going to open up a lot of opportunities for Sourcegraphb
101、ecause they already have a lot of the code and can say,Well take a customized model,throw in your code,and give you super good suggestions for your developers.So I think thats an area that is super interesting.And again,they have to compete with GitHub,which already has Copilot.VP,Technology at Publ
102、icly traded e-commerce companySlashNext is working on a generative AI-based solution,which means it will generate its own kind of phishing and malware and it will train the software to automatically be aware of any new kind of threats arriving in the market.So even if tomorrow some human is creating
103、 a new kind of malware or any other software is creating some new kind of phishing or ransomware,because SlashNext is based on AI,it is already aware of these kinds of changes and it will be able to detect them before any other software can do so.That is a differentiator from the technology perspect
104、ive.Senior Design Engineer at Fortune 500 company90Source:CB Insights The state of generative AI in 7 chartsGenerative interfaces,like Anthropics AI assistant Claude,lead in funding among cross-industry toolsDistribution of generative AI funding,Q322 Q2231,646+370%175+661%*Based on an analysis of 21
105、0+generative AI companies building cross-industry solutions;excludes deals to industry-specific companies and model developers such as OpenAI.91Source:CB Insights Mutiny company profile-headcount;Jasper software buyer interview transcriptGrowing competition is a threat to vendors in some cross-indus
106、try markets,like text generation&editingSo in this newly emerging world of generative AI its hard to keep up with all the changes that are going on.Its probably not fair to ask this,but Ill say it:Jasper needs to stay up to date faster to make me a definite yes to renew.We need to look at their pric
107、ing model to make sure,as Im beginning to use it more and more at a higher and higher scale,that it keeps working and the price remains right for me.Im seeing other lower cost options;the price of calls to GPT-3,for example,has gone down to really minimal numbers.It might be harder to say yes to a r
108、enewal when were due next year,so Ill have to really see that our people have picked this up and are finding great value.C-level executive at$10M+funded research platformMutiny and Jasper announce layoffs92Source:CB Insights Generative AI legal case search&summarization;Virtual medical scribes&summa
109、rization toolsWatch for vendors to scramble to build defensible moats in specialized areas93Drive growthImprove customer experienceReduce costs&riskHealthcare&Healthcare&life scienceslife sciences Copilots for doctors automate tedious tasks&improve EHR documentation De-noise radiology scans AI compa
110、nions address well-being&mental health Synthetic patient data protects patient privacy GenAI drug discovery&design reduces time-to-market Biomedical NLP supports clinical decision-makingFinancial Financial services&services&insuranceinsurance GenAI assistants analyze&synthesize financial data at sca
111、le Automated underwriting decisions GenAI chatbots simplify day-to-day financial tasks Personalized interactions in insurance sales process Synthetic training data improves financial models&ensures compliance Pattern identification in unstructured claims filings to minimize lossesRetailRetail LLM-po
112、wered search improves conversion Smarter,more relevant search Personalized avatars GenAI automates product catalogs Synthetic humans save on model costsHow generative AI is going to be used to Industry3.Opportunity in vertical genAI943.Opportunity in vertical genAIHealthcare&life sciences95Health sy
113、stems and pharma players are using genAI to scale everything from drug design to EHR documentationSource:CB Insights 7 applications of generative AI in healthcareHEALTHCARE&LIFE SCIENCES96CB Insights Understanding generative AIs potential in healthcare-webinarAI expertise is a necessity in sectors l
114、ike pharma to reduce time-to-marketSelect generative AI drug discovery&design exits in 2023HEALTHCARE&LIFE SCIENCESAcquired by Recursion Pharma in May 2023Acquired by BioNTech in January 2023Filed for Hong Kong IPO in June 202397Source:CB Insights Virtual scribes&summarization tools market report ES
115、P,Generative AI copilots for doctors have raised more than$240MGenAI copilots for doctors automate tedious tasks like note-takingHEALTHCARE&LIFE SCIENCES98Source:CB Insights Corti Analyst Briefing;Virtual scribes&summarization tools market report Up-and-comer Corti raises$60M Series B in September 2
116、023,taking on Microsofts Nuance HEALTHCARE&LIFE SCIENCES99Source:CB Insights AI companions market reportApplications to enhance well-being and mental health emerge,including AI-generated music,VR landscapes,and companionsTop-funded companies developing AI companions(as of 10/30/2023)1,646+370%175+66
117、1%HEALTHCARE&LIFE SCIENCES100Source:CB Insights OpenAI software buyer interview transcript,Cohere software buyer interview transcriptEHR workflows are ripe for LLM disruption,from document search to summarization to suggested diagnosesI can potentially ask ChatGPT,hey,does this person have out of ne
118、twork coverage and is this person eligible for spine surgery or something like that?Then,we are having to look at multiple documents and you dont know where to look,essentially,and youre essentially just giving the combination of all these documents as an input to ChatGPT.VP,Machine Learning,Fortune
119、 500 companyHEALTHCARE&LIFE SCIENCESWe have tens of thousands,if not hundreds of thousands,of patients on our devices.We have device populations in the millions.For service and operations,theres a high demand in terms of the support that they can give to a patient if theyre in a trial or if theyre j
120、ust going about their day-to-day life on therapy,on one of these devicesSo,why we were investigating these chatbots was to lower their cognitive burden so that 10:1,5:1 ratio could be equalized.Sr.Research Engineer,Fortune 500 company101Source:CB Insights OpenAI software buyer interview transcriptBu
121、ndling genAI tools with existing cloud subscriptions is giving big tech companies an advantage with market reachAdvantage is with,for example,John Snow Labs,its a very sort of clinically trained model.Its not just trained on wiki pages or like general text.In that sense,I think its much betterin ter
122、ms of entity recognition and things like that.But the limitations,I would think,are these models are not getting trained on the volume of data anywhere as close to what ChatGPT is trained on It OpenAI deployment was pretty minimal overhead the goal is to essentially enable the use of ML tools that a
123、re available from the Azure subscription at the Enterprise levelVP,Machine Learning,Fortune 500 companyHEALTHCARE&LIFE SCIENCES102Source:Company research&announcementsVendors will compete to license real-world datasets to build genAI models tailored for the healthcare industryHEALTHCARE&LIFE SCIENCE
124、SGenAI for imaging trained on X-rays and radiology reportsLLM trained on 2M patients clinical notesLLM trained on 10 years of health records and 400K patients clinical notes103Source:CB Insights Cohere software buyer interview transcriptPrivacy and data handling practices will be major issues as the
125、 market matures1,646+370%175+661%HEALTHCARE&LIFE SCIENCES104Source:CB Insights Generative AI Expert CollectionExpect healthcare-focused genAI vendors to lean into data security as a point of differentiation1,646+370%HEALTHCARE&LIFE SCIENCES1053.Opportunity in vertical genAIFinancial services&insuran
126、ce106In financial services,generative AI will automate tasks and transform how organizations use financial dataSource:CB Insights 3 applications of generative AI in financial services FINANCIAL SERVICES&INSURANCE107Insurers are using genAI to drive more personalized customer experiences and internal
127、 automation efforts Source:CB Insights 3 applications of generative AI in insuranceFINANCIAL SERVICES&INSURANCE108Source:CB Insights business relationships and news mentions;company websites and press releasesFinserv incumbents are experimenting with generative AI;prime use cases in document summari
128、zation&extraction 1,646+370%175+661%FINANCIAL SERVICES&INSURANCE109Source:CB Insights Advanced Search-Earnings transcripts*Reflects quarter call occurredGenerative AI will automate&enhance underwritingWe are on the large language models and the potential benefit that that will ultimately bring beyon
129、d algorithmic,particularly in underwriting and claims and the ability to work either replace work that is done or make it more accurate,or work alongside underwriters.Chubb CEO Evan GreenbergQ223 earnings callFINANCIAL SERVICES&INSURANCE110Source:CB Insights news mentions Bloomberg,JP Morgan Chase,I
130、ntuitIncumbents with access to vast financial data&resources will develop LLMs purpose-built for financial servicesFINANCIAL SERVICES&INSURANCEBloomberg is developing BloombergGPT an LLM built for financial tasks like sentiment analysis and news classification.Intuit partnered with OpenAI to develop
131、 GenOS a genAIoperating system that will power customer experiences across its product suite.JP Morgan Chase is developing IndexGPT,an AI chatbot for investment advice and selection.111Source:CB Insights Kasisto Analyst Briefing;Cognaize software buyer interview transcriptB2B fintechs developing gen
132、AI tools across compliance,document processing,and chatbots land big-name customersFINANCIAL SERVICES&INSURANCECognaizes solution had better accuracy,evolution,AI and machine learning capabilities,and user interface than the other vendors we considered Cognaizeis integral to our workflow as they pro
133、vide a tool that our team uses to extract relevant data from public documents and map it to our templates.Senior Vice President,Publicly traded financial management company1123.Opportunity in vertical genAIRetail113GenAI use cases in retail will cut costs and create more engaging contentSource:CB In
134、sights 6 applications of generative AI in retail RETAIL114Source:Company announcementsE-commerce search is getting smarter,with retailers developing search powered by ChatGPTRETAILRecommendations,product attributes,dietary considerationsHelp choosing items based on budget,food constraints,menu ideas
135、115Source:Company announcements;CB Insights AX Semantics software buyer interview transcriptPlatforms will move quickly to revamp text on product pages&automate product catalogs with genAIRETAILShopify Magic generates,revises,and expands product descriptionsScans customer reviews and generates a sum
136、mary We have a custom contract with AX Semantics that contains the usage of the platform itself,as well as extended customer support.Our contract allows us to create automated texts in nine languages for different channels,such as the product description on the web shop or the catalog,search engine
137、optimized texts for our category sites,and also content for social media.Manager,Manufacturing company116Source:CB Insights Generative AI e-commerce search market report;Algolia software buyer interview transcriptSpecialized search vendors are moving fast to catch up,with a focus on relevance to dri
138、ve conversionLeading eLeading e-commerce search vendors by equity funding(as of 9/30/2023)commerce search vendors by equity funding(as of 9/30/2023)RETAILWe ended up rebuilding our entire catalog infrastructure using AlgoliaVisual merchandising and all the rules that come with it was a big important
139、 feature that we really wanted.Until I met Algolia for the first time,I have never seen a search engine of that scale that was not Google or Bing or Yahoo operate that quickly.Head of Digital Product,E-commerce company117Source:CB Insights Generative AI synthetic humans&fashion design market reportL
140、eading brands are working with genAI vendors to create personalized avatars and diverse models at scaleRETAIL118Source:CB Insights The generative AI in retail market mapRetail-specific genAI solutions improve digital shopping operations,engagement,and conversion1,646+370%175+661%RETAIL119Promising c
141、ompanies to watch120Source:CB Insights Generative AI 50Explore the full listWe identified the 50 most promising genAI startupsGenerative AI 50121CB Insights helps the worlds leading companies understand everything they need to know about disruptive technologies find out more about why our customers love us here.Explore the CollectionCB Insights customers can find and track every generative AI company mentioned in this report using our analyst-curated Expert Collection