Claude 3.7 Sonnetについては、2月25日の発表時に『Anthropicによれば、Claude 3.7 Sonnetは「市場初のハイブリッド推論モデル」で、ベンチマークではOpenAIのo1やo3-mini、DeepSeek-R1を超える性能とのことです』と書いただけになっていました。 Claude 3.7 Sonnetは、コーディングの分野では群を抜いた存在になっていること、日本語能力が高く自然で親しみやすい文体で長文処理能力が高いことなどが特徴で、スライド作成も可能であり資料作成に便利です。 現時点では、Deep Researchで深く幅広く情報を集め、特許情報や社内情報などをWEBにアップしないNotebookLMにアップロードしバックで動いているGemini 2.0で情報を集約し、OpenAI o1 proで推論を深めた解析を行い、Claude 3.7 Sonnetで最終資料作成するのが最強になりそうです。 特許文書の読解支援AIアシスタント「SUMMARIAサマリア」も、さっそく「拒絶支援ワークフロー機能」で利用しているAIを「Claude3.7Sonnet」にバージョンアップしています。 拡張思考に対応した「Claude 3.7 Sonnet」 25/2/202 https://yorozuipsc.com/blog/claude-37-sonnet 「SUMMARIAサマリア」:サポートデスク(新機能)、拒絶支援機能の強化 https://patent-i.com/summaria/manual/R_20250303 The Remarkable Capabilities of Claude 3.7 Sonnet Regarding Claude 3.7 Sonnet, I had only mentioned in my February 25 announcement that "According to Anthropic, Claude 3.7 Sonnet is the market’s first hybrid reasoning model, outperforming OpenAI’s o1 and o3-mini, as well as DeepSeek-R1 in benchmarks." Claude 3.7 Sonnet has now proven itself to be an outstanding model, particularly excelling in coding. It is also characterized by its high proficiency in Japanese, producing natural and friendly writing, and its strong long-form processing capabilities. Additionally, it can create slides, making it highly convenient for document preparation. At present, the most powerful workflow appears to be leveraging Deep Research for in-depth and broad information gathering, consolidating data using Gemini 2.0, which runs in the background of NotebookLM (a tool that does not upload patent or internal corporate information to the web), conducting deeper inference analysis with OpenAI o1 Pro, and finally using Claude 3.7 Sonnet for final document creation. The patent document analysis AI assistant "SUMMARIA" has already upgraded its AI to Claude 3.7 Sonnet for its rejection support workflow feature. Your browser does not support viewing this document. Click here to download the document. Your browser does not support viewing this document. Click here to download the document. Your browser does not support viewing this document. Click here to download the document. Your browser does not support viewing this document. Click here to download the document. Your browser does not support viewing this document. Click here to download the document.
0 Comments
政府は2月28日、人工知能の開発促進や透明性確保を目指す法案を閣議決定しました。 石破茂首相は昨年12月に「日本が世界のモデルとなるAI制度を構築する」と表明していますが、罰則がない法体系で本当に大丈夫なのか、自民党の会合でも「将来的には罰則規定を盛り込むべきだ」との意見が出たようです。 人工知能関連技術の研究開発及び活用の推進に関する法律案 令和7年2月28日、科学技術・イノベーション推進事務局 https://www.cao.go.jp/houan/217/index.html 概要https://www.cao.go.jp/houan/pdf/217/217gaiyou_2.pdf 要綱https://www.cao.go.jp/houan/pdf/217/217youkou_2.pdf 法律案及び理由https://www.cao.go.jp/houan/pdf/217/217anbun_2.pdf 新旧対照表https://www.cao.go.jp/houan/pdf/217/217shinkyu_2.pdf 参照条文https://www.cao.go.jp/houan/pdf/217/217sansho_2.pdf 日本のAI法案の全文をみてみよう!〜2025/2/28閣議決定「人工知能関連技術の研究開発及び活用の推進に関する法律案」 https://note.com/informationlaw1/n/nea72aa2e6b7e 悪質事業者は公表、AI新法を閣議決定 開発推進めざす https://www.nikkei.com/article/DGXZQOUA253BZ0V20C25A2000000/ AI規制「日本流」探る 国に調査権、新法案を閣議決定 https://www.nikkei.com/article/DGXZQOUA2576L0V20C25A2000000/ AI悪用は事業者名公表、政府が新法策定へ 罰則は見送り https://www.nikkei.com/article/DGXZQOUA313KS0R30C25A1000000/ Draft Bill on Promoting Research, Development, and Utilization of AI-related Technologies On February 28, the government approved a draft bill aimed at promoting the development of artificial intelligence and ensuring transparency. Prime Minister Shigeru Ishiba stated last December that "Japan will establish an AI regulatory framework serving as a global model." However, concerns have arisen regarding the absence of penalties in the proposed legislation. At a Liberal Democratic Party meeting, there were opinions expressed suggesting that "penalty provisions should eventually be incorporated." Your browser does not support viewing this document. Click here to download the document. Your browser does not support viewing this document. Click here to download the document. Your browser does not support viewing this document. Click here to download the document. Your browser does not support viewing this document. Click here to download the document. 2025/02/26 にライブ配信された『サマリア「項目指示機能」で特許情報分析のバリエーションをひろげよう』の動画に続き、ブログ記事が公開されました。
サマリアの分類支援機能は便利です。まだ試していない方は試してはどうでしょうか。 サマリアの「項目指示機能」で特許情報分析のバリエーションをひろげよう https://note.com/yu_py/n/n863aa6a0e1f5?sub_rt=share_b サマリア「項目指示機能」で特許情報分析のバリエーションをひろげよう https://www.youtube.com/watch?v=AdNJawGEyfY&t=2s Summaria's Classification Support Feature Following the live-streamed video "Expand Variations in Patent Information Analysis with Samaria's 'Item Instruction Feature'" on February 26, 2025, a blog article has now been published. Samaria's classification support feature is very useful. If you haven't tried it yet, why not give it a go? 知財実務情報Lab. 管理人の高橋政治弁理士が、不利益がある発明、成功確率が低い発明、従来技術よりも効果が低い発明、商業的に成功した発明であっても特許が取れるのかについて説明されています。
誤解されている方が多いので、確認してみてください。 不利益がある発明、成功確率が低い発明、従来技術よりも効果が低い発明、商業的に成功した発明であっても、特許が取れるのか? 2025.02.25 https://chizai-jj-lab.com/2025/02/25/0222/ Inventions with disadvantages, inventions with a low probability of success, inventions with low effectiveness... Patent attorney Masaharu Takahashi, the administrator of Intellectual Property Practice Information Lab., explains whether patents can be obtained for inventions that have disadvantages, inventions with a low probability of success, inventions that are less effective than prior art, or even commercially successful inventions. Many people have misconceptions about this topic, so be sure to check it out. Can patents be obtained for inventions with disadvantages, inventions with a low probability of success, inventions that are less effective than prior art, or even commercially successful inventions? February 25, 2025 https://chizai-jj-lab.com/2025/02/25/0222/ ChatGPT 4.5(GPT-4.5)は、2025年2月27日にOpenAIがリリースした最新のAIモデルです。従来のGPT-4やGPT-4oから大幅に性能が向上し、より自然な会話、高度な推論、高い共感性(EQ)を実現していますが、モデル自身が回答生成前にチェーン・オブ・ソート(逐次推論)を行う“推論型”ではなく、膨大な事前学習データによる知識強化に特化した「最後の非推論モデル」とされています。 2025年2月25日に発表されたClaude 3.7 Sonnet への対抗馬としての、数か月後に登場することが発表されているChatGPT-5までのつなぎのモデルにも感じられます。 やはり非推論モデルでは、「すもももももももものうち」の「も」の数は正解できないようです。 2025年2月27日 GPT-4.5 が登場 https://openai.com/ja-JP/index/introducing-gpt-4-5/ The Last Non-Reasoning Model: "ChatGPT 4.5 (GPT-4.5)" Emerges ChatGPT 4.5 (GPT-4.5) is the latest AI model released by OpenAI on February 27, 2025. It features a significant performance improvement over its predecessors, GPT-4 and GPT-4o, offering more natural conversations, advanced reasoning, and high emotional intelligence (EQ). However, it is not a "reasoning-type" model that performs chain-of-thought (sequential reasoning) before generating responses. Instead, it is positioned as the "last non-reasoning model," focusing on knowledge enhancement through extensive pre-training data. It also appears to serve as a transitional model leading up to ChatGPT-5, which is scheduled for release a few months later as a competitor to Claude 3.7 Sonnet, announced on February 25, 2025. As expected from a non-reasoning model, it still struggles to correctly count the number of "も" in the Japanese tongue twister "すもももももももものうち". Your browser does not support viewing this document. Click here to download the document. Your browser does not support viewing this document. Click here to download the document. Your browser does not support viewing this document. Click here to download the document. Your browser does not support viewing this document. Click here to download the document. Your browser does not support viewing this document. Click here to download the document. Your browser does not support viewing this document. Click here to download the document. |
著者萬秀憲 アーカイブ
September 2025
カテゴリー |