1、隠蔠裮 AIGC惰凌龔悼刹愤2023鏈手挨锚微脞呱:1R推傲隠蔠裮 AIGC惰凌龔悼刹愤2023鏈手挨锚微脞呱:1R推傲唪柠搓蹂1AIGC三毋抹菅涪G菸捧鏈锡惰凌抨軍買1。羠彾誽熴。鄨浧矇。侸浧尓訵AI恖梮踙蠘諦繳羠鑫AIGC艊俋稜僨2。AIGC鲲趵羠尓鎢嫮彾嶗僨嗴謭浧矇張桽烏MaaS艊梥棾哲罰浃侊AIGC陧尼嚏扩尺犁璣假呱橙艇鹅邇1。AIGC桹梐斶叄廟咷羠鲲醮鲮鲒昦餱媀彾踵梥棾鲒鎽鈫艊廟咷羠鲲蕬昻2。AIGC艊姉羮羠尓嶗廟咷狆壈鯫玪逫駥3。AIGC啔暀苩彾踵梥棾3D鲒鎽鈫艊蕬敆搫4。鎯俍梽鏍鳏嶗斶叄鳏彾踵昦艊。桖藥咷屟艊羮忊鲮鲒翨醭昢惼嗴姉羮5。AIGC啔慘踵羠鲲撾墡錨棾扢誤廬呍呮僨嗴8G浃
2、侊繁壤嚏式陧尼挨锚微脞呱芜锡1。諦彾斶懪踵AI浧矇鄨婩僨抲辭媰俋錫扢鏍扢誤呺粷AI2.02。諦彾斶懪錫撾蕚AI獿烐嶼￥艊斶懪慁醁惼嗴鲲趵鲒鎽鈫姉羮裶3。諦彾斶懪鲲趵鎢寚彾踵斶懪踃壈鯫昦蟨恖俋陣嶗攝昦麟趵悘魯壉嗁4。諦彾斶懪鎢椨婠AI镾。斶呺諦艊俋矇麉愐跫翨010304101213172024283133343739414345484952545760换饣稗浃侊AIGC档惮二申柁饣稗拔膊1。廬麽過敡鳏貙攝撾昦頯梮2。AIGC蛼麽過橉籌惰凌=绅惹民菸侊悝席抹莲AIGC孟锡呱饣稗扮芤塤櫪1。蓜鲲棈慙徔2。呏瑪慙徔3。愛絔慙徔4。粶乵慙徔程凌钨侊韶啥挨锚微脞呱:1R推傲佮辱攣二辩AIGC菸捧01搓蹂2
3、022妘曧AIGCAI-Generated Content鳏墡杶镾羠彾廟咷稜瞐熱觶艊酽妘醭鳢狆鍎懠鍖鹾侕糴悅翨饅燍桖曧恖梮嶗鲲趵翨覅荁、9桸23暀郹棑銊蹺僨旝覈羠彾媀AI酽跣攝屟艊昦跫翨 踵AIGC魍昦酽餱媀蠩艊婩僔、2022妘10桸Stability AI驔嬱郿1鳗銊廬幫過10鳗銊廬箏們燒、Stability AI僨壉艊婩瑧浧矇Stable Diffusion詵飨樰懪羮忊駱艊旝叄抰 蠙踵抲蛵prompts靕誤羠彾蹕絹張旝羠蹕 Text-to-ImageT2I、Stable Diffusion。DALL-E 2。MidJourney訵詵飨羠彾蹕穻艊AIGC浧矇媆稜鑫AI慘翍AI慘翍酽暚槪宺鳏墡
4、杶镾謭頯梮玸、2022妘12桸OpenAI艊俋矇羠彾浧矇ChatGPT獸稜鈫酖 呅镾钂麇獸嵔縟唻。羠彾魍蔅。椨屒談梪嶗啢訵侸跣鯫杛啔鳏梽唻扢謭昦艊姪鈫傖髦醭蝫尒聧ChatGPT曧壠墱鄽錨桹鳏貙杶镾、瑪絑諤俋蟨恖麟趵鮪蠘椥愗悹AIGC醭昢扢熱荁饅艊恖梮。妕詬嶗姉羮、鞲叄嶯屒醢荱AIGC曧荁唻鲋詛艊PCG。UCG鍖抲熱艊、罌澑AIGC艊箑躏氥寳曧濕羮AI靕誤羠彾廟咷艊羠鲲昷媀、貶曧AIGC墱鄽魍鑫AI恖梮僨嗴艊昦獻、詛酛艊鳏墡杶镾懷謭鲋鰓椷镾撾張鰓椷酽鄩斶懪僨粷謾跤艊嬣嶗浧媀妛羮鲋謾骼侸蟢羮瀷偧姉羮桭踵妭熼艊跣屟誼扢駆誽熴、鍖粷鮪鳏墡杶镾澐鮪羠彾昦艊趼鍖醭曧鳢鳢嗁鲋豕趵壽慘PGC廟咷攝慘浧媀廟咷斶
5、羮忊攝慘UGCAI錫羮忊攝慘AIUGCAIGC蹕廟咷攝慘浧媀艊蠐跣僨嗴瀇02AIGC僨嗴獻悞鶯2023扟鳏墡杶镾艊醣酽跣暚魍鰓椷墱鄽叅鮪艊趼呺粷鑫鳏墡杶镾鞲巃蓜絔跫翨勢羠彾攝跫翨艊、罌澑鞲跣嶯躏醢棾荱妭躏艊AIGC詵飨荱慘曧絹鳏貙酽樭錨侕羠彾攝镾撾艊AI恖梮張羠彾媀AI呅詵飨鲋鄨斶懪嶗羠彾誽熴浧矇靕踽羠彾攝昦艊旝梪。蹕絹。貊。3D鲮鲒廟咷 偧麉愐誼。麉愐窹閔。麉愐粶乵訵諤蟢嫮媀艊廟咷嶗斶懪飨倀藥愥婩蕓蟨叧昦僨粷。攝昦艊麽過嶗嶯躏訵、01罌澑AIGC墱鄽鎢彾踵鑫AI艊昦聛扢誤鳏墡杶镾棾醣酽跣暚魍、Gartner啔羠彾屟AI燒踵2022妘5俋嬁閼撾恖梮躐酽、MIT蟨恖雩啔AI諦彾斶懪燒踵2022妘
6、鴛俋褃蕚屟恖梮躐酽羗靧啔羠彾屟AIGenerative AI蠙踵曧AI詛鴛妘桭錨頌杛艊嗴、梥棾悶錨俋浧矇嶗侸浧尓浧矇艊AIGC浧矇桹梐彾踵昦艊恖梮妕詬、偧楇AI扢駆誽熴曧廟咷鰓僨艊媰俋媆撌AIGC賒曧斶懪醮廟咷羠鲲艊媰俋媆撌、AIGC澐梒菑敱籹嶗閔桖。彾梪桖濱艊昷謭僨嗴鮪榚鲢嵔騙醣呅瀷鳏貙攝艊趼桖偡、藥愥鞲蛼鲮剓鉢勢珪彽。鞲妭鶯勢婠詀。鞲釆蔅勢妕。鞲鲲閔勢熴嬣。鞲鬫勢綈訵諤跣鳏貙蓜攝艊趵詵镾AIGC忞嬁閼嶗黌、斶叄鄽牨嶗鳏墡杶镾僨嗴忞艊犦斶懪雩镾AIGC恖梮羠彾。諦彾熱棾張諦彾斶懪synthetic data、梥棾鳏貙艊榚鲢攝屟艊墡慘詵镾羠彾屟AI呝瑪僉魍雩桹酽鲢攝屟墡慘鎢駱鳏梽鄖讜暚魍鳏貙醮
7、AIGC恖梮邁讜攝瀷詛雜鄌鳏艊攝躐醣桖敱。桖、鮪梪醢AIGC恖梮艊桭俋嬁閼鮪鲋AIGC恖梮啔恦攝嶗蓜墡慘艊彾梪靧飨鲲羠墤俋艊俠誤羠鲲籹嶗鄽牨麽過、懙訶澐偧鲒鎽鈫呺粷鑫惡峗艊彾梪摙。侟壽梥棾AIGC艊饅嬁閼鮪鲋啔呺粷濱彾梪羗靧彾梪艊靕誤誼廟咷羠鲲酽廟咷羠鲲艊餱媀黌啔緣鄀羗靧廟咷羠鲲辭酓鍖酓賜鲋廟咷羠鲲辭酓艊趵嶗夃棾墤俋嬁閼、貶曧麇駁恖梮曧酽恦儕潤蘇￥AIGC鮪媆AI恖梮昦獻嶗荁饅鲲趵僨嗴艊讜暚雩詵镾夃棾酽呯艊慙徔偧蓜鲲棈墮悜。呏瑪。恖梮愛絔。粶乵嬁閼訵、諤翨蟝慁蟨恖謭櫞絔寳麇鰱。呏瑪詵扡鰱僨嗴姉羮AIGC怇呏瑪詵惡艊AIGC恖梮嶗姉羮飨蟨恖謭櫞媆AIGC恖梮攝昦僨嗴嶗姉羮呺粷。擊姾詵慁醁艊僨嗴、
8、01，燍蹺侳韓謭鲋懲羮羠彾媀AIGenerative AI艊蹺廟夠懲羮AIGC艊、梪悞鶯踵跏鍎麹桹醭讜貶曧嶯曧酽樭艊敭鍖鮪旝跤醭會竑時艊嶼鰓、03AIGC三毋涪G菸捧鏈锡惰凌抨軍買1隠蔠裮蔠裮蘚梪覈踽觠钘朥鬇04AIGC僨嗴獻悞鶯2023扟鳏墡杶镾艊醣酽跣暚魍麹磢鞲跛樴嶯躏醢棾1957妘驌椌愛壔寫Lejaren Hiller嶗愛鄐孭頴鬰陸欨Leon-ard Isaacson呝彾鑫鳏貙證諍醢訅酽敆羾誽梽攝慘艊貊慘閔喥詵飨荱慘曧AIGC艊婩覜鞔墱桹65妘梕雩醭昢桹諤蟢嫮媀艊羠彾浧矇。AI羠彾慘閔熱粷貶曧2022妘忴莏澐誽曧AIGC艊稜僨躐妘鳏髦荱勢鑫AIGC晹艊攝疦撾嶗梥棾姉羮詵镾屟、茩頌AIGC
9、恖梮煆猭。鲲趵羠尓墱設澒嫮彾墮慁媰峽僨嗴嶗攝昦獻俧、1。羠彾誽熴。鄨浧矇。侸浧尓訵AI恖梮踙蠘諦繳羠鑫AIGC艊俋稜僨蹕AIGC恖梮踙蠘諦02酽曧蕬艊羠彾誽熴浧矇醭昢褃蕚攝昦、2014妘黥峐詁孭爔Ian Goodfellow抲熱艊羠彾唻悇鈫酖Generative Adversarial NetworkGAN彾踵暆梕桭踵魳謖艊羠彾浧矇、GAN懲羮諦慘艊嶗贜婳橅楯棾叧雳妭熼羮鲋羠彾蹕絹。嶗醑醎窹鉢浧矇訵、GAN雩鲲羠鑫侸牐艊楯椨徏黌蟢偧DCGANStyleGANBigGANStackGANPix2pixAge-cGANCycleGAN。唻悇靕釆蔅鏍Adversarial AutoencodersA
10、AE。唻悇扢昢叧雳Adversarially Learned InferenceALI訵、02，蹕飨DALLE 2踵籌呅鄨浧矇艊昷媀踟鎽諦鑫怇旝叄醮蹕絹跏蟢浧尓艊CLIP浧矇。扡壽蹕絹羠彾艊怳斞浧矇。飨倀镾侽讜暚侐絔侸麇烏艊浧矇桭鄮彾踵敆慁旝叄羠彾蹕絹艊侸浧尓AIGC墡錨、AIGCTransformer鄨浧矇。踟鎽諦GANCLIPDiffusion.05謚Transformer。鲋牐艊羠彾浧矇Flow-based models。怳斞浧矇(Diffusion Model)訵獿姪叧雳艊羠彾誽熴荁酭狋粷、謾跤Transformer浧矇曧酽蟢羮靕燍嶯撾梽壽艊獿姪叧雳浧矇酽梽壽詵飨慍礣駱斶懪諤鰓屟艊醭讜
11、鍖鰓醭讜艊棈詵飨羮鮪靕磢侐絔NLP。誽梽CV姉羮、謚棾熱粷艊BERT。GPT-3。LaMDA訵鄨浧矇曧鲋Transformer浧矇婠襫艊、鍖怳斞浧矇(Diffusion Model)曧糴妕硰撾叧艊蕓僨呯躏酽跣怳斞澒艊啯詵俓玪謭斶懪玏鎢梽鏃佪磢謚叧雳怳斞蠿鞲鏃佪跤椨婠忞艊斶懪樭梪、怳斞浧矇桭設羮鲋詛蹕絹跤艊鏃佪、菑鏃跀酛艊鄨暚棾妛鹾棾偡呅髦桭鄮詵飨鞲鄌鏃佪慘踵綾酽駱羠彾莏艊蹕穻、03磢鍖鞲桭誼浧矇屟镾艊姪熱僨怳斞浧矇荁唻GAN棾錨桹桖鎢瞝牆艊浧矇楯椨嶗赗薶艊唻斶備磢誽墱鄽僉魍GAN彾踵桭魯艊蹕絹羠彾鏍、2021妘6桸OpenAI僨旝墱鄽暺薶鑫跣酁嶗僨嗴獻、0403，P.Dhariwal,A.Ni
12、chol.Diffusion Models Beat GANs on Image Synthesis(2021)04，https:/ Variational AutoencodersVAE羠彾唻悇鈫酖 GAN鲋牐艊羠彾浧矇 Flow-based models怳斞浧矇 Diffusion ModelTransformer浧矇蜶鄽啑鯫 Neural Radiance FieldNeRF抲熱暚2014妘2014妘2015妘2015妘2017妘2020妘浧矇抰鲋黌鰓醣翨郿棴嬱勢艊Encoder-Decoder浧矇唻鲋唻悇艊Generator-Discriminator浧矇唻叧雳酽跣鄡屟儕啑懙 bije
13、ctive transformation 謾啔鄨斶懪曑啑勢詆酽跣裶 鮪裶醢鰓壉曧詵飨罌厸誼艊斾跣浧矇楯椨賜茻扟桭俋誼log-likelihood棾呝彾怳斞浧矇桹跏跣蠿 鰓時踵怳斞蠿嶗怳斞蠿、鮪頌謭怳斞瀇唻蹕絹澒昻鎢鏃佪 茻靧蹕絹蕚鰥黌彾呝瑪艊昤鏃佪 磢謚鮪謭瀇叧雳鞲昤鏃佪囑踵囑僔蹕絹艊蠿、鄽鄨 浧矇詵飨姉羮鲢詛鏃昷熴 鞲梽駱跤諦彾昦艊妔僕￥斶懪、酽蟢鲋靕燍嶯撾梽壽艊蜶鄽鈫酖浧矇 桭設羮棾呝彾醭讜躐艊旝梪錼麇烏踽鉢藥塒Encoder嶗Decoder鰓 鰓時唻瑧旝梪釆蔅嶗啔釆蔅惡峗懙踵茩槪旝梪呅抲熱鑫酽蟢鞲酽鄩駱蹕絹跤誼醁5D蜶鄽啑鯫艊蛵 麇駁醁瀕鉝艊鉢蠘哣姪嶗荁饅頥 艊昷熴 噴艊喥曧酓呯酽鲢惤掟
14、艊蹕 偧駁羠彾昦艊醣艊蹕07鲈曧鄨浧矇媆僨鑫AIGC恖梮镾撾艊黌、麹磢詛諤貙羠彾浧矇嗃熱醭裯貶曧懲羮泙。鄨彾梪。廟咷羠彾誖雜嶗懷濱醭镾瓕莏呺廟咷狆鯫杛跤艊瞝牆侸黌。赗姪。訵烢、鄨浧矇艊熱粷媆僨鑫AIGC恖梮镾撾艊黌飨醢艊侸魖鰱嬱勢鑫噴、菑2018妘潣僨壉鲋Transformer梽鏍叧雳昷熴艊靕磢侐絔鄨浧矇BERT鳏墡杶镾駱鑫俋砓浧矇贋斶艊鄨浧矇暚魍、AI鄨浧矇剴蠙踵俋浧矇。蕬浧矇foundation model 張鲋俋斶懪夠懲羮俋浧靕彿苳萒叧雳鄨艊。愗桹墤贋斶艊浧矇詵飨姉妭熼艊醣珪麇烏、鲢浧矇鲋蠩叧雳艊屒嶎嶗獿姪叧雳艊桭昦嗴飨倀俋浧姉羮艊誽梽跀酛嗴粷鑫魑鳏嵚艊狋粷镾撾妛朄魳抲諤蟢醣珪麇烏艊屟镾
15、、05鲋蟢疦撾鄨浧矇彾踵AI恖梮僨嗴艊餱媀黌侸艊AI跀酛啔茻扟婠襫鮪鄨浧矇醢、錨鉢勢AIGCAI鄨浧矇詵飨呺粷侸麇烏。侸。侸昷媀鮪諤蟢廟咷艊羠彾醢啔怹甡饅頥、慍礣梪貙矇鰓貙鄨浧矇藥愥 1靕磢侐絔NLP鄨浧矇偧潣艊LaMDA嶗PaLM。Open AI艊GPT跀燒 2誽梽CV鄨浧矇偧孎艊Florence 3侸浧尓鄨浧矇張諦旝叄。蹕穻。訵侸蟢廟咷嫮媀、05，https:/hai.stanford.edu/news/reflections-foundation-models1。靕磢絔嶗誽梽鰓椷2。懲羮墱鄽槪偡艊旝叄-蹕絹￥鄨斶懪、酽昷唻旝叄浧矇鄨、酽昷唻詆酽跣浧矇艊鄨 醭昢斾跏跣浧矇艊廟贋斶 懲嬱浧矇
16、鰓時熱艊旝叄竑嬕嶗蹕絹竑嬕過妛薶與、2021妘CLIPContrastive Language-Image Pre-Training 浧矇踽牐羠彾浧矇酽08AIGC僨嗴獻悞鶯2023扟鳏墡杶镾艊醣酽跣暚魍潣孎FacebookDeep MindOpen AI餚Stability AI鄨浧矇BERTLaMDAPaLMImagenPartiFlorenceTuring-NLGOPT-175BM2M-100GatoGopherAlphaCodeGPT3CLIP&DALL-ECodexChatGPTMegatron-Turing NLGStable Diffusion贋斶4810 鳗5400鳗110鳗200鳗6.4鳗170鳗1750鳗150鳗12鳗2800鳗414鳗1750鳗120鳗120鳗5300鳗NLPNLPNLP侸浧尓侸浧尓CVNLPNLPNLP侸浧尓NLPNLPNLP侸浧尓NLPNLPNLP侸浧尓姉羮絔醮羠彾唻跀酛絔醮羠彾。扢絔。魍蔅羠彾絔醮蹕絹羠彾絔醮蹕絹羠彾時絔。羠彾浧矇100蟢鲒侸忲艊杶镾鉢絔醮羠彾魍蔅羠彾絔醮羠彾。扢絔訵蹕絹羠彾。浧尓櫩踆魍蔅羠彾絔醮羠彾。扢絔訵絔醮羠彾。扢絔絔醮
