{"id":5073,"date":"2024-10-28T13:58:19","date_gmt":"2024-10-28T13:58:19","guid":{"rendered":"https:\/\/howtogeek.blog\/cs\/?p=5073"},"modified":"2024-10-28T13:58:19","modified_gmt":"2024-10-28T13:58:19","slug":"claude-35-sonnet-released-by-anthropic-a-new-and-impressive-ai-model","status":"publish","type":"post","link":"https:\/\/howtogeek.blog\/cs\/claude-35-sonnet-released-by-anthropic-a-new-and-impressive-ai-model\/","title":{"rendered":"Claude 3.5 Sonnet vydan\u00fd spole\u010dnost\u00ed Anthropic: Nov\u00fd a p\u016fsobiv\u00fd model um\u011bl\u00e9 inteligence"},"content":{"rendered":"

Ned\u00e1vno spole\u010dnost Anthropic zavedla vylep\u0161en\u00ed sv\u00e9ho p\u0159edn\u00edho modelu, Claude 3.5 Sonnet, kter\u00fd mohl sklouznout pod radar kv\u016fli dal\u0161\u00edm v\u00fdznamn\u00fdm aktualizac\u00edm, jako je Computer Use a Analysis tool. Aktualizovan\u00fd Claude 3.5 Sonnet v\u0161ak p\u0159in\u00e1\u0161\u00ed n\u011bkolik pozoruhodn\u00fdch vylep\u0161en\u00ed.<\/p>\n

P\u0159esto\u017ee pou\u017e\u00edv\u00e1n\u00ed po\u010d\u00edta\u010de p\u0159edstavuje v\u00fdznamn\u00fd pokrok, jeho p\u0159ijet\u00ed mezi u\u017eivateli je st\u00e1le omezen\u00e9. Je to p\u0159edev\u0161\u00edm proto, \u017ee je v ran\u00fdch f\u00e1z\u00edch v\u00fdvoje a postr\u00e1d\u00e1 plnou spolehlivost. Nav\u00edc je jeho dostupnost omezena na pou\u017eit\u00ed API, tak\u017ee je m\u00e9n\u011b p\u0159\u00edstupn\u00fd pro \u0161ir\u0161\u00ed publikum.<\/p>\n

Naproti tomu nov\u00fd Claude 3.5 Sonnet je k dispozici k okam\u017eit\u00e9mu pou\u017eit\u00ed a nab\u00edz\u00ed v\u00fdrazn\u011b vy\u0161\u0161\u00ed v\u00fdkon. Tak\u017ee, co p\u0159esn\u011b odli\u0161uje tento aktualizovan\u00fd 3.5 Sonet?<\/p>\n

Zpr\u00e1va od Anthropic zd\u016fraz\u0148uje, \u017ee nejnov\u011bj\u0161\u00ed Claude 3.5 Sonnet prok\u00e1zal v\u00fdrazn\u00e1 zlep\u0161en\u00ed oproti p\u0159edchoz\u00edm model\u016fm a mnoha konkurent\u016fm v pr\u016fmyslov\u00fdch m\u011b\u0159\u00edtc\u00edch. Pokroky v agentn\u00edm k\u00f3dov\u00e1n\u00ed a pou\u017eit\u00ed n\u00e1stroj\u016f jsou obzvl\u00e1\u0161t\u011b p\u016fsobiv\u00e9. Nap\u0159\u00edklad zv\u00fd\u0161ila p\u0159esnost agentn\u00edho k\u00f3dov\u00e1n\u00ed z 33,4 % na 49 % p\u0159i ov\u011b\u0159ov\u00e1n\u00ed SWE a pro pou\u017eit\u00ed agentn\u00edch n\u00e1stroj\u016f p\u0159i testu TAU na benchmarku se zv\u00fd\u0161ila z 51,0 % na 69,2 % v maloobchod\u011b a z 22,8 % na 46,0 % v leteck\u00e9m sektoru. Je pozoruhodn\u00e9, \u017ee t\u011bchto vylep\u0161en\u00ed je dosa\u017eeno za stejnou cenu a rychlost jako u jeho p\u0159edch\u016fdce.<\/p>\n

\"Aktualizace
Zdroj: Antropick\u00fd<\/figcaption><\/figure>\n

P\u0159i hodnocen\u00ed Claude 3.5 Sonnet (nov\u00fd), kter\u00fd n\u011bkte\u0159\u00ed u\u017eivatel\u00e9 ozna\u010duj\u00ed jako 3.6 Sonnet kv\u016fli z\u00e1m\u011bn\u011b pojmenov\u00e1n\u00ed, vykazuje pozoruhodn\u00fd pokrok, zejm\u00e9na v k\u00f3dov\u00e1n\u00ed, uva\u017eov\u00e1n\u00ed a schopnostech kreativn\u00edho psan\u00ed.<\/p>\n

Claude, d\u0159\u00edve uzn\u00e1van\u00fd jako jeden z nejv\u00fdkonn\u011bj\u0161\u00edch v k\u00f3dov\u00e1n\u00ed, ud\u011blal s touto aktualizac\u00ed v\u00fdznamn\u00fd pokrok, nyn\u00ed dok\u00e1\u017ee vy\u0159e\u0161it velmi slo\u017eit\u00e9 probl\u00e9my v r\u00e1mci jedin\u00e9 odpov\u011bdi. Zlep\u0161ila tak\u00e9 svou schopnost opravit p\u0159edchoz\u00ed chyby.<\/p>\n

U\u017eivatel\u00e9 zjist\u00ed, \u017ee nov\u00fd Claude 3.5 Sonnet poskytuje odezvy mnohem rychleji a \u0159e\u0161\u00ed frustrace, kter\u00e9 se vyskytly u jeho d\u0159\u00edv\u011bj\u0161\u00ed verze, kde v\u00fdkon znateln\u011b poklesl.<\/p>\n

Claude tak\u00e9 vylep\u0161il sv\u00e9 uva\u017eov\u00e1n\u00ed a p\u0159irovnal jej k modelu OpenAI o1. I kdy\u017e se nemus\u00ed shodovat s o1 ve slo\u017eit\u00fdch \u00faloh\u00e1ch uva\u017eov\u00e1n\u00ed, z\u016fst\u00e1v\u00e1 kompetentn\u00ed pro jednodu\u0161\u0161\u00ed uva\u017eov\u00e1n\u00ed.<\/p>\n

Pro poutav\u00fd test Claudova uva\u017eov\u00e1n\u00ed se ho zkuste zeptat, How many words are in your response to this prompt?<\/code>. Je pozoruhodn\u00e9, \u017ee modely architektury transform\u00e1tor\u016f \u010dasto z\u00e1pas\u00ed s takov\u00fdmi v\u00fdzvami kv\u016fli jejich pot\u0159eb\u011b p\u0159edpov\u00eddat n\u00e1sleduj\u00edc\u00ed slova. Tato v\u00fdzva v\u0161ak povzbuzuje model, aby p\u0159edem vytvo\u0159il strategii.<\/p>\n

Model OpenAI o1 vynik\u00e1 v efektivn\u00edm zpracov\u00e1n\u00ed t\u00e9to v\u00fdzvy. I kdy\u017e nov\u00fd Claude 3.5 Sonnet nedosahuje tak \u00fapln\u011b zdatnosti o1, v\u00fdrazn\u011b p\u0159ekon\u00e1v\u00e1 star\u0161\u00ed verzi Claude. D\u0159\u00edve mohl Claude odhadnout kvalifikovan\u011b, ale nyn\u00ed se pokou\u0161\u00ed slova spo\u010d\u00edtat p\u0159esn\u011b \u2013 i kdy\u017e bez toho, aby je v\u0161echny zva\u017eoval stejn\u011b. Zda bude Claudeova odpov\u011b\u010f pova\u017eov\u00e1na za spr\u00e1vnou, bude nakonec z\u00e1viset na osobn\u00ed interpretaci.<\/p>\n

Let me count the words in my response to your question:
\nThis (1) is (2) my (3) first (4) response (5) to (6) your (7) prompt (8).
\nThere are 8 words in my response up to this point. Note that I did not count the words after the period since those were explaining my counting process rather than being part of my direct response to your question.<\/code><\/code><\/p>\n

\"Claude\u016fv<\/figure>\n

Podle poskytnut\u00e9 instance by n\u011bkte\u0159\u00ed u\u017eivatel\u00e9 mohli nam\u00edtnout, \u017ee Claude p\u0159esn\u011b po\u010d\u00edtal slova, ve kter\u00fdch zva\u017eoval svou odpov\u011b\u010f. Objas\u0148uje, \u017ee \u017e\u00e1dn\u00e1 n\u00e1sleduj\u00edc\u00ed slova netvo\u0159\u00ed sou\u010d\u00e1st jeho odpov\u011bdi, proto\u017ee jednodu\u0161e rozv\u00e1d\u011bj\u00ed proces uva\u017eov\u00e1n\u00ed. Zat\u00edmco n\u011bkte\u0159\u00ed mohou p\u0159ehl\u00e9dnout vynech\u00e1n\u00ed \u00favodn\u00ed v\u011bty \u201eNechte m\u011b spo\u010d\u00edtat slova v m\u00e9 odpov\u011bdi na va\u0161i ot\u00e1zku\u201c, jin\u00ed to mohou vid\u011bt jinak. Nakonec se to st\u00e1v\u00e1 v\u011bc\u00ed osobn\u00edho \u00fasudku.<\/p>\n

Nicm\u00e9n\u011b tento p\u0159\u00edklad ilustruje, \u017ee nov\u00fd Claude 3.5 Sonnet skute\u010dn\u011b zlep\u0161il sv\u00e9 schopnosti uva\u017eov\u00e1n\u00ed. Nyn\u00ed m\u016f\u017ee artikulovat sv\u00e9 my\u0161lenkov\u00e9 pochody nebo \u0159\u00edci: \u201eP\u0159em\u00fd\u0161l\u00edm o tom, vydr\u017e\u201c, kdy\u017e to vy\u017eaduje dal\u0161\u00ed \u010das na zv\u00e1\u017een\u00ed \u2013 n\u011bco podobn\u00e9ho jako o1. Ob\u010das se to m\u016f\u017ee dokonce opravit samo, kdy\u017e \u0159ekne: “Nech m\u011b to znovu promyslet.”<\/p>\n

Tato schopnost progresivn\u00edho my\u0161len\u00ed tak\u00e9 pos\u00edlila Claudeovy tv\u016fr\u010d\u00ed schopnosti psan\u00ed. D\u00edky schopnosti myslet dop\u0159edu dok\u00e1\u017ee vytvo\u0159it roz\u0161\u00ed\u0159en\u00e9 p\u0159\u00edb\u011bhy, kter\u00e9 obsahuj\u00ed koherentn\u00ed oblouky, p\u0159edv\u00eddav\u00e9 prvky a podmaniv\u00e9 postavy.<\/p>\n

Krom\u011b toho bylo dosa\u017eeno v\u00fdznamn\u00e9ho pokroku v jeho analytick\u00e9m v\u00fdkonu. Claude nyn\u00ed v oblasti anal\u00fdzy \u00fazce konkuruje nejrozs\u00e1hlej\u0161\u00edmu modelu Anthropic, Claude 3 Opus, a tak\u00e9 OpenAI o1 mini.<\/p>\n

Stru\u010dn\u011b \u0159e\u010deno, nejnov\u011bj\u0161\u00ed aktualizace p\u0159inesly pozoruhodn\u00fd pokrok v mnoha dimenz\u00edch. Nov\u00e9 funkce k\u00f3dov\u00e1n\u00ed si z\u00edskaly velkou pozornost. Sou\u010dasn\u00fdm omezen\u00edm pro Clauda jsou v\u0161ak limity pou\u017eit\u00ed, kter\u00e9 jsou podstatn\u011b p\u0159\u00edsn\u011bj\u0161\u00ed i pro u\u017eivatele Pro ve srovn\u00e1n\u00ed s ChatGPT.<\/p>\n

Zdroj<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"

Ned\u00e1vno spole\u010dnost Anthropic zavedla vylep\u0161en\u00ed sv\u00e9ho p\u0159edn\u00edho modelu, Claude 3.5 Sonnet, kter\u00fd mohl sklouznout pod radar kv\u016fli dal\u0161\u00edm v\u00fdznamn\u00fdm aktualizac\u00edm, jako je Computer Use a Analysis tool. Aktualizovan\u00fd Claude 3.5 Sonnet v\u0161ak p\u0159in\u00e1\u0161\u00ed n\u011bkolik pozoruhodn\u00fdch vylep\u0161en\u00ed. P\u0159esto\u017ee pou\u017e\u00edv\u00e1n\u00ed po\u010d\u00edta\u010de p\u0159edstavuje v\u00fdznamn\u00fd pokrok, jeho p\u0159ijet\u00ed mezi u\u017eivateli je st\u00e1le omezen\u00e9. Je to p\u0159edev\u0161\u00edm proto, \u017ee je […]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[179,242,126],"class_list":["post-5073","post","type-post","status-publish","format-standard","hentry","category-how-to","tag-artificial-intelligence","tag-claude","tag-microsoft"],"acf":[],"_links":{"self":[{"href":"https:\/\/howtogeek.blog\/cs\/wp-json\/wp\/v2\/posts\/5073","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/howtogeek.blog\/cs\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/howtogeek.blog\/cs\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/howtogeek.blog\/cs\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/howtogeek.blog\/cs\/wp-json\/wp\/v2\/comments?post=5073"}],"version-history":[{"count":1,"href":"https:\/\/howtogeek.blog\/cs\/wp-json\/wp\/v2\/posts\/5073\/revisions"}],"predecessor-version":[{"id":5074,"href":"https:\/\/howtogeek.blog\/cs\/wp-json\/wp\/v2\/posts\/5073\/revisions\/5074"}],"wp:attachment":[{"href":"https:\/\/howtogeek.blog\/cs\/wp-json\/wp\/v2\/media?parent=5073"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/howtogeek.blog\/cs\/wp-json\/wp\/v2\/categories?post=5073"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/howtogeek.blog\/cs\/wp-json\/wp\/v2\/tags?post=5073"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}