InoteveraGaidhi rinotevera
Protein Language Models like ESM
Tekinoroji
Nhungamiro yehunyanzvi
Watermarking Mutauro Model Outputs
Watermarking inomisikidza chiratidzo chakavanzika muAI-yakagadzirwa zvinyorwa kuti igozoonekwa seyakanyorwa nemuchina, pasina kuchinja zvinoonekwa nemuverengi wemunhu.
Teerera kugwaro iriAI-yakarondedzerwa · ~2 min
Pa peji ino2 min verenga
Pfupiso
It matters for spotting misinformation, academic dishonesty, and unlabeled AI content at scale.
Kudzika Kwakadzika
Mutauro wemodhi inogadzira mavara echiratidzo chimwe panguva nekuita sampling kubva pakuparadzirwa kwemazwi. A watermark biases iyo sampling muchivande, inoberekana nzira. Mune yakakurumbira Kirchenbauer-style scheme, hashi yeatangire tokens mbeu pseudorandom kupatsanurwa kwemazwi mune yakasvibira runyoro uye runyoro runyoro, wobva wanongedza modhi kuti usarudze girinhi zviratidzo. Mavara emunhu anongo shandisa ma tokens egirinhi nematsvuku zvakaenzana, asi mavara akaiswa watermark ane huwandu husingafungidzike hwakasara hwegirinhi. Detector inoziva kiyi yakavanzika inodzokorodza zvinyorwa uye inoita bvunzo yehuwandu, ichiisa mureza iyo ine green-token count yakawandisa kuti isagone kuitika. Hapana kiyi yakavanzika inochengetwa muzvinyorwa zvacho; chiratidzo chinogara mune sarudzo dzechiratidzo.
Kuitwa muNyika yeChokwadi
Google DeepMind's SynthID-Chinyorwa zvisingaonekwe mawatermark Gemini zvabuda kuitira kuti kambani igone kuzoziva mavara ayo anogadzirwa.
Yunivhesiti inoshandisa watermark detector kuratidza zvinyorwa zvakatumirwa zveAI-yakagadzirwa ndima uku ichichengetedza kuverenga kwevadzidzi.
Chikuva chenhau chinotarisa kana mafashama emashoko akatumirwa anotakura chiratidzo che watermark chinoratidza chizvarwa chebhoti.
Mupi wemuenzaniso anopinza watermark kuti aenderane nemitemo yekubuda pachena iri kubuda pasi pemitemo yakaita seEU AI Act.
Ramangwana reWatermarking Mutauro Model Outputs
Google DeepMind's SynthID-Chinyorwa chakafambisa watermarking mukugadzira, uye vanogadzira mitemo kusanganisira EU AI Act vanowedzera kutarisira masaini ekutanga pane zvakagadzirwa. Tsvagiridzo iri kusunda kune mawatermark akasimba pakutaura nekucheka, semantic watermarks anopona pashanduro, uye zvirongwa zveruzhinji-kiyi kuitira kuti chero munhu aone pasina kubata chakavanzika chinovatendera kuti vafunge. Dambudziko rakavhurika rinoramba riri nhangemutange yemaoko: madhigirii akasimba akatarisana nekuchipa kwekubvisa kurwiswa, uye chokwadi chekuti chero yakavhurika-yehuremu modhi inogona kungodzima watermarking.
Kuzvishandisa
Nei zvichikosha
- Mutengo uye bhajeti. Zvisarudzo zvezvivakwa zvinotyaira kuita uye mutengo wekushandisa kwemakore.
- Sarudzo dzakajeka. Dzidzo yehunyanzvi inobatsira zvikwata kusarudza murwi wakakodzera, kwete iwo mutsva chete.
- Kudzora kwemhando yepamusoro. Sarudzo dzeinjiniya dziri nani dzinoderedza zviitiko zvekuvimbika mukugadzira.
Chenjera
- Kugadzirisa imwe bhenji kunogona kuvanza yakafara system kushaya simba.
- Infrastructure uye mari yekugadzirisa inowanzotarisirwa pasi.
- Chengetedzo uye kucherechedzwa mapundu anogona kukura sezvo masisitimu anowedzera kuoma.
Matanho anotevera
- Tsanangura latency, mhando, uye mutengo zvinangwa usati waitwa.
- Benchmark pasi pechokwadi mutoro uye data mamiriro.
- Chishandiso chekutarisa zvikanganiso, kudonha, uye mushandisi maitiro.
- Gadzirira nzira dzekudzosera kumashure uye dzezviitiko usati wawedzera.
Ramba Uchiongorora
Tsamba yemahara
Tora iyo yemazuva ese AI pfupiso
Nhatu dzakasimbiswa dzeAI nyaya mangwanani ega ega pakati pevhiki, dzakanyorwa neChirungu chakajeka. Yemahara zvachose, hapana ads.
Imwe email zuva rega rega revhiki. Bvisa kunyoresa kamwe chete. Hatimbotengese kana kugovera kero yako.
Zviedze iwe pachako
Take the Watermarking Language Model Outputs quiz
Mhinduro yepakarepo pamhinduro yega yega, uye chitupa chinogovaniswa chine ID inobatika kana wapedza kosi.
Tsigira dzidzo yemahara yeAI. AI Understanding ndeye 501(c)(3) isingabatsiri - hapana ads, hapana paywall, zvachose. Ita chipo
Mibvunzo inowanzo bvunzwa
What is Watermarking Language Model Outputs?
Watermarking inomisikidza chiratidzo chakavanzika muAI-yakagadzirwa zvinyorwa kuti igozoonekwa seyakanyorwa nemuchina, pasina kuchinja zvinoonekwa nemuverengi wemunhu. Izvo zvine basa pakuona manyepo, kusavimbika kwedzidzo, uye zvisina kunyorwa AI zvemukati pachiyero.
Mune yakasvibira-rondedzero/dzvuku-chinyorwa watermark, chiratidzo chakamisikidzwa sei?
Chirongwa ichi chinopatsanura duramazwi kuita runyorwa rwegirinhi uye rutsvuku uchishandisa mhodzi yakavanzika uye unonanzva modhi kuti usarudze zviratidzo zvegirinhi, zvichisiya huwandu hwehuwandu pane chero chiratidzo chinooneka.
Sei mavara e watermark achinetsa kuona kana ari mapfupi?
Huwandu hwekucherekedzwa hunoungana pamusoro pematokeni uye hunokura nehurefu hwekutevedzana, saka ndima mapfupi anoshaya akakwana egirinhi-chiratidzo chakawanda kuti apfuure mukana nechivimbo.
Chii chinowanzo sarudza kana chiratidzo chinogara pane yakasvibira kana tsvuku runyorwa?
A pseudorandom basa rakadyarwa nematangiro ekutanga uye kiyi yakavanzika inopatsanura zvakare kupatsanura mazwi, saka detector inogona kudzokorodza mamwe mazita gare gare.
Ndechipi chiito chingango bvisa kana kunetesa mameseji watermark?
Kududzira uye kududzira kunotsiva kwakawanda kwesarudzo dzakaiswa watermark, kugeza yakanyatsoiswa yakasvibira-chiratidzo yakawandisa iyo detector inovimba nayo.
Chii chakakosha tradeoff kana uchiwedzera kusimba kwegirinhi-token bias?
Kurerekera kwakasimba kunoburitsa chiratidzo chakajeka, chakasimba asi chinomanikidza modhi kubva kune yainoda tokeni, ichikuvadza zvishoma kutsetsenura uye kusiyana.
Ramba uchidzidza
Mazano anoenderana
Mamwe madhairekitori akasarudzirwa nyaya iyi