|
2025年12月17日に発表されたGemini 3 Flashは、最先端のインテリジェンスを搭載し、上位の3 Pro版の4分の1以下のコストで高速処理を実現しており、多くのベンチマークで2.5 Proを上回り、より高速な速度を実現しているだけでなく、一部のベンチマークでは3 Proを凌駕する性能を実現しています。その秘密は、Agentic RL(エージェント強化学習)という革新的な学習手法を採用したことのようです。 このGemini 3 Flashについて、生成AIに深堀りさせました。さらに、報告結果をNotebookLMでインフォグラフィック、スライド資料にさせました。 なお、生成AIによる調査・分析結果は、公開された情報からだけの分析であり、必ずしも実情を示したものではないこと、誤った情報も含まれていることについてはご留意されたうえで、ご参照ください。 Gemini 3 Pro GA版間近?Flashを進化させた「agentic RL」がProに適用で最強モデル爆誕か! https://www.youtube.com/watch?v=gc7u_24lrHo Gemini 3 Flash Surpasses the Flagship Model Gemini 3 Flash, announced on December 17, 2025, is equipped with state-of-the-art intelligence and delivers high-speed performance at less than one-quarter of the cost of the higher-tier Gemini 3 Pro. It outperforms Gemini 2.5 Pro across many benchmarks and achieves faster inference speeds overall. Moreover, in some benchmarks, it even surpasses the performance of Gemini 3 Pro. The key to this breakthrough appears to be the adoption of an innovative training method known as Agentic Reinforcement Learning (Agentic RL). I conducted an in-depth analysis of Gemini 3 Flash using generative AI. Furthermore, the results of this analysis were converted into infographics and slide materials using NotebookLM. Please note that the investigation and analysis conducted by generative AI are based solely on publicly available information and may not fully reflect actual conditions. They may also contain inaccuracies; therefore, please review the content with this understanding in mind. 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.
0 Comments
東京大学 大学院工学系研究科の高原 泉 大学院生(博士後期課程)、同大学 生産技術研究所 溝口 照康 教授、モントリオール大学 Mila - Quebec AI InstituteのBang Liu(バン・リウ) 准教授らの研究グループは、大規模言語モデル(Large Language Model:LLM)の高い推論能力に着目し、目標特性を持つ無機結晶材料を、自然言語による説明を出力しながら自律的に探索・設計する生成AIフレームワーク「MatAgent」を開発しました。 このMatAgentは大規模言語モデル(LLM)を「脳」として活用し、人間のような論理的推論と説明可能なプロセスを通じて自律的に新材料を探索する点に大きな特徴があります。この東京大学生産技術研究所が開発した無機材料設計AIエージェント「MatAgent」について、生成AIに深堀りさせました。さらに、報告結果をNotebookLMでインフォグラフィック、スライド資料にさせました。 なお、生成AIによる調査・分析結果は、公開された情報からだけの分析であり、必ずしも実情を示したものではないこと、誤った情報も含まれていることについてはご留意されたうえで、ご参照ください。 2025.12.18 プレスリリース 【記者発表】大規模言語モデルで専門家のように材料空間を探索――自律性と解釈性を備えた無機材料設計のためのAIエージェントを開発―― https://www.iis.u-tokyo.ac.jp/ja/news/4955/ An Autonomous and Interpretable AI Agent for Inorganic Materials Design, “MatAgent” A research group led by Izumi Takahara, a doctoral candidate (PhD program) at the Graduate School of Engineering, the University of Tokyo; Professor Teruyasu Mizoguchi of the Institute of Industrial Science, the University of Tokyo; and Associate Professor Bang Liu of Mila – Quebec AI Institute, Université de Montréal, has developed a generative AI framework called “MatAgent.” This framework autonomously explores and designs inorganic crystalline materials with target properties while outputting natural-language explanations, leveraging the advanced reasoning capabilities of large language models (LLMs). MatAgent is distinguished by its use of an LLM as a “brain,” enabling it to autonomously search for new materials through human-like logical reasoning and explainable processes. I conducted an in-depth analysis of this inorganic materials design AI agent “MatAgent,” developed by the Institute of Industrial Science at the University of Tokyo, using generative AI. Furthermore, the results of this analysis were converted into infographics and presentation slides using NotebookLM. Please note that the investigations and analyses conducted by the generative AI are based solely on publicly available information and may not fully reflect the actual situation; they may also contain inaccuracies. Please keep this in mind when referring to the content. 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. 「ブリヂストンや荏原が「知財ROIC」経営 出願数競う〝とりあえず特許〟にメス」という記事が出ていました。 従来の「特許出願件数」を競うコストセンターとしての知財管理から脱却し知財を将来のキャッシュフローを生む投資と捉え直すパラダイムシフトが進んでいることを背景に、日本企業における知的財産戦略と投下資本利益率(ROIC)を統合した新たな経営手法について詳述しています。そして、ブリヂストンの「知的財産価値創造性」や荏原製作所の「知財ROICツリー」など、先進企業による具体的な定量化モデルと独自の指標が紹介されています。また、オムロンや日立製作所の事例を通じ、知財活動がいかにして事業の収益性や資本効率の向上に寄与するかという論理的枠組みが示されています。 投資家との共通言語として財務指標を用いることで、無形資産を企業価値へ転換するための戦略的対話の重要性が強調されています。 「知財ROIC」経営について生成AIに深堀りさせました。さらに、報告結果をNotebookLMでインフォグラフィック、スライド資料にさせました。 なお、生成AIによる調査・分析結果は、公開された情報からだけの分析であり、必ずしも実情を示したものではないこと、誤った情報も含まれていることについてはご留意されたうえで、ご参照ください。 ブリヂストンや荏原が「知財ROIC」経営 出願数競う〝とりあえず特許〟にメス 2025.12.17 https://business.nikkei.com/atcl/gen/19/00815/121600010/ Bridgestone and Ebara’s “IP ROIC” Management An article titled “Bridgestone and Ebara Adopt ‘IP ROIC’ Management: Putting an End to ‘Just-in-Case Patents’ and the Numbers Game” was published. The article describes a paradigm shift underway in Japan, moving away from intellectual property management as a cost center focused on competing over the sheer number of patent filings, toward a new perspective that treats IP as an investment capable of generating future cash flows. Against this backdrop, it provides a detailed discussion of a new management approach that integrates intellectual property strategy with Return on Invested Capital (ROIC). The article introduces concrete quantitative models and proprietary metrics developed by leading companies, such as Bridgestone’s concept of “intellectual property value creation capability” and Ebara Corporation’s “IP ROIC tree.” In addition, through case studies of companies such as Omron and Hitachi, it presents a logical framework illustrating how IP activities can contribute to improved business profitability and capital efficiency. By using financial indicators as a common language with investors, the article emphasizes the importance of strategic dialogue aimed at converting intangible assets into corporate value. I asked generative AI to further explore the concept of “IP ROIC” management, and then used NotebookLM to turn the results into infographics and presentation slides. Please note that the investigations and analyses conducted by generative AI are based solely on publicly available information and do not necessarily reflect actual circumstances. They may also contain inaccuracies, so please review them with this in mind. 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. デンカ株式会社は、新規事業の成功率が極めて低いという課題に対し、生成AIを活用してベテランの暗黙知を形式知化する画期的な取り組みを進めています。このプロジェクトでは、若手の知識不足と熟練者の固定観念という相反する障壁を、専門家チームによる多段階のプロンプト設計によって解消し、30年分の経験を短縮するアイデア創出を実現したということです。AIが提案した案の約3割は経験豊富な社員でも思いつかない斬新なものであり、技術継承とイノベーションのジレンマを同時に克服するモデルとして注目されています。 この取り組みを生成AIに深堀りさせました。さらに、報告結果をNotebookLMでインフォグラフィック、スライド資料にさせました。 なお、生成AIによる調査・分析結果は、公開された情報からだけの分析であり、必ずしも実情を示したものではないこと、誤った情報も含まれていることについてはご留意されたうえで、ご参照ください。 デンカが挑む「1,000に3つ」の新規事業成功の壁:生成AIは、若手の「30年の経験」を飛び越えた発想が可能になるか? ~4,000人の熱量が支える、0.3%への挑戦~ https://gomana.ai/project/denka/ Denka Takes on the Barriers to New Business Success with Generative AI Denka Co., Ltd. is pursuing a groundbreaking initiative to address the extremely low success rate of new business development by leveraging generative AI to formalize the tacit knowledge of veteran employees. In this project, two opposing challenges—young employees’ lack of experience and the entrenched assumptions of seasoned experts—are overcome through multi-stage prompt design developed by a team of specialists, enabling idea generation that effectively compresses 30 years of experience. Approximately 30% of the proposals generated by the AI were novel ideas that even highly experienced employees had not conceived, drawing attention to this effort as a model that simultaneously resolves the dilemma between technology transfer and innovation. I asked generative AI to further explore and analyze this initiative. The results were then transformed into infographics and presentation materials using NotebookLM. Please note that the investigation and analysis conducted by generative AI are based solely on publicly available information and may not fully reflect actual conditions. The suggesting may also contain inaccuracies, and should therefore be referenced with due caution. 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. 12月18日に配信された(第267回)知財実務オンライン:「 知財実務における生成AIプロンプトドリブン改革」は、島津製作所 知的財産部 阿久津 好二 部長の話でした。 10月30日に、サマリアのウェビナーもありましたが、どんどん進化していますので、ぜひ、視聴されることをお勧めします。 (第267回)知財実務オンライン:「 知財実務における生成AIプロンプトドリブン改革」(ゲスト:株式会社島津製作所 知的財産部 部長 阿久津 好二) https://www.youtube.com/watch?v=hZCcdHl_19g IP Practice Online: Shimadzu’s Generative AI Prompt-Driven Transformation The IP Practice Online broadcast (Episode 267), delivered on December 18 and titled “Generative AI Prompt-Driven Transformation in Intellectual Property Practice,” featured a talk by Koji Akutsu, General Manager of the Intellectual Property Department at Shimadzu Corporation. There was also a Samaria webinar on October 30, and as developments are progressing rapidly, I highly recommend watching these sessions. 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. キリンHDにおけるAI導入が活発です。キリンのAI導入について生成AIに深堀りさせました。さらに、報告結果をNotebookLMでインフォグラフィック、スライド資料にさせました。 なお、生成AIによる調査・分析結果は、公開された情報からだけの分析であり、必ずしも実情を示したものではないこと、誤った情報も含まれていることについてはご留意されたうえで、ご参照ください。 キリンHDが導入した「AI役員」の舞台裏、4カ月で分かった効果と課題 2025.11.19 https://xtech.nikkei.com/atcl/nxt/column/18/00001/11257/ キリンはAI時代を「データメッシュ」で戦う──独自生成AIの活用拡大で新たに挑むマネジメントの現在地 「今が組織変革の最大のチャンス」AI活用を起点にデータの“当事者意識”向上を狙う 2025/11/25 https://enterprisezine.jp/article/detail/22864 従業員の生成AI利用率は70%。DXの先にある「人間にしかできない仕事、AIと共創する未来」 2025年11月27日 https://note-kirinbrewery.kirin.co.jp/n/nb059f4f338d5 キリンHDの「AI役員」評価上々 1議案に60の意見や論点提示 2025年12月10日 https://www.nikkei.com/article/DGXZQOUC203D00Q5A121C2000000/ ビール造りで独自開発のAI活用 キリン、おいしさ成分見つけ提案 12/12 https://news.yahoo.co.jp/articles/13cc15d7fad69d22734119c8063e37062352c075 ビールの香味成分を特定する嗜好AI「FJWLA」を独自に開発 2025年12月15日 https://www.kirinholdings.com/jp/newsroom/release/2025/1212_01.html キリン、ビールづくりに独自AI「フジワラ」導入 26年3月以降発売の製品から 2025年12月16日 https://www.itmedia.co.jp/aiplus/articles/2512/16/news065.html 2025.12.17 キリンと日立、嗜好データとAIを活用した共同研究を開始 https://www.hitachi.co.jp/New/cnews/month/2025/12/1217b.pdf キリンと日立、嗜好データとAIを活用した共同研究を開始 消費者の飲料選択や飲酒行動の要因を解明し商品開発や健康増進への貢献を目指す 2025年12月17日 https://www.kirinholdings.com/jp/newsroom/release/2025/1217_02.html キリンと日立、嗜好データとAIの活用で消費者の飲料選択や飲酒行動の要因を解明する共同研究を開始 12/18 https://news.yahoo.co.jp/articles/811aee889a9b5a3acf61c51235b9fcf559439db2 世界初!キリンと富士通、創薬DX技術を活用し、AIと実試験でシチコリンの腸脳作用メカニズムを解明 2025年12月17日 https://www.kirinholdings.com/jp/newsroom/release/2025/1217_01.html キリンと富士通、AI創薬DX技術でシチコリンの腸脳作用メカニズムを解明 2025/12/18 https://bizzine.jp/article/detail/12492 AI Adoption at Kirin Holdings AI adoption at Kirin Holdings (Kirin HD) has been accelerating. I asked a generative AI to conduct an in-depth analysis of Kirin’s AI initiatives, and further transformed the findings into infographics and slide materials using NotebookLM. Please note that the investigation and analysis conducted by the generative AI are based solely on publicly available information; therefore, they may not fully reflect the actual situation and may include inaccuracies. I ask that you review the results with this understanding in mind. 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. 2025年1月20日、生成AIを用いて作られた画像を無断で複製したとして、千葉県警は無職の男を著作権法違反(複製権侵害)の疑いで書類送検しました。AIで作られた画像に著作権があるとして、著作権法違反で摘発するのは全国で初めてのようです。 このAI生成画像の著作権の問題を、生成AIに深堀りさせました。さらに、報告結果をNotebookLMでインフォグラフィック、スライド資料にさせました。 なお、生成AIによる調査・分析結果は、公開された情報からだけの分析であり、必ずしも実情を示したものではないこと、誤った情報も含まれていることについてはご留意されたうえで、ご参照ください。 AI生成画像を無断複製、初の摘発か 著作権法違反疑いで男性書類送検 https://www.nikkei.com/article/DGXZQOUD2067T0Q5A121C2000000/ Man Referred to Prosecutors for Allegedly Unauthorized Copying of AI-Generated Images in Violation of Copyright Law On January 20, 2025, the Chiba Prefectural Police referred an unemployed man to prosecutors on suspicion of violating the Copyright Act (infringement of the right of reproduction) for allegedly making unauthorized copies of images created using generative AI. This appears to be the first case nationwide in which enforcement action has been taken for copyright infringement on the basis that AI-generated images are protected by copyright. I asked generative AI to conduct an in-depth analysis of the copyright issues surrounding AI-generated images, and then used NotebookLM to turn the findings into infographics and presentation slides. Please note that the research and analysis conducted by generative AI are based solely on publicly available information and do not necessarily reflect the actual circumstances; they may also contain inaccuracies. Please review the content with this in mind. 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. 多くの企業でAIが定着しないのは、社員が「思ったような回答が出ない=AIは使えない」と早合点してしまうからです。この根本原因は「ナレッジギャップ」にあります。 これは、人間なら当たり前の「暗黙知」や「背景」をAIに伝えきれていない現象です。 しかし、そのギャップを埋める完璧なプロンプトは数千文字に及ぶこともあり、毎回自力で作るのは困難です。 コツは「AIへの指示を、AIに書かせる」ことです。 「何をしたいか」という意図だけを伝え、詳細なプロンプトはAIに生成させる。私自身もこの方法を実践しています。 最新AIのIQは145レベルに達しています。重要なのは小手先のテクニックより「十分なコンテキスト」を与えること。 最初のうちは「ナレフルチャット」のような、プロンプト作成を支援してくれるサービスを活用するのも、定着化への近道です。 【生成AIが期待ハズレになる理由と対策】コンテキスエンジニアリングと生成AI/企業のAI定着化を阻むナレッジギャップの正体/プロンプト作成を書かせるのは〇〇 https://www.youtube.com/watch?v=oT-Y9zD6MOc&list=TLGGA1SCs4a4gvUxNjEyMjAyNQ The Knowledge Gap That Prevents AI from Taking Root in Companies In many organizations, AI fails to become firmly established because employees jump to the conclusion that “if it doesn’t give the answer I expected, AI is useless.” The root cause of this is a knowledge gap. This gap arises when the tacit knowledge and context that humans take for granted are not adequately conveyed to AI. However, a truly complete prompt that fills this gap can run into thousands of characters, making it unrealistic to create from scratch every time. The key is to have AI write the instructions for AI. You simply communicate your intent—what you want to achieve—and let the AI generate the detailed prompt. I personally use this approach myself. The IQ of the latest AI models has reached around 145. What matters is not superficial prompt tricks, but providing sufficient context. In the early stages, using services like “Narefull Chat,” which support prompt creation, can be a shortcut to achieving sustainable AI adoption within an organization. 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. 読売新聞社は2025年12月15日、生成AI(人工知能)の利用環境の整備に向けて、「『信頼できる生成AI』との共生に関する提言」をまとめました。 国産AI開発状況の現状を生成AIにまとめさせ、結果をNotebookLMでインフォグラフィック、スライド資料にさせました。 なお、生成AIによる調査・分析結果は、公開された情報からだけの分析であり、必ずしも実情を示したものではないこと、誤った情報も含まれていることについてはご留意されたうえで、ご参照ください。 生成AI提言 国産開発に高い倫理観生かせ https://www.yomiuri.co.jp/editorial/20251216-GYT1T00003/ Current Status of Domestic AI Development On December 15, 2025, The Yomiuri Shimbun compiled a set of recommendations titled “Proposals on Coexisting with ‘Trustworthy Generative AI’” aimed at improving the environment for the use of generative AI (artificial intelligence). I asked generative AI to summarize the current state of domestic AI development, and then used NotebookLM to transform the results into infographics and presentation slides. Please note that the research and analysis conducted by generative AI are based solely on publicly available information. They do not necessarily reflect actual conditions and may include inaccuracies. I ask that you review the content with this understanding. 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. Preferred Networksの岡野原大輔社長が、国産AIを開発し、その選択肢を持つことの重要性、なぜ国産AIという「選択肢」が必要なのか?を、経済、文化、安全保障の点から指摘しています。そして、日本のAI産業が「お手上げ」と諦めるのは時期尚早であり、他社との協力や、日本ならではの「電力制限のある中での戦い方」を模索し、フロンティアモデルに対抗しながらも、最終的にはユーザーに選ばれる製品・ソリューションを提供していく姿勢を示しています。 内容について生成AIにまとめさせました。さらに、結果をNotebookLMでインフォグラフィック、スライド資料にさせました。 なお、生成AIによる調査・分析結果は、公開された情報からだけの分析であり、必ずしも実情を示したものではないこと、誤った情報も含まれていることについてはご留意されたうえで、ご参照ください。 【ChatGPT・Geminiの学習データ「日本語はたった0.1%」】最先端AIに日本勢「お手上げ」は早すぎる/Preferred Networks岡野原社長「国産AIで勝てる」【1on1 Tech】 TBS CROSS DIG with Bloomberg https://www.youtube.com/watch?v=Vlj0K7r1qkY Why Is the “Option” of Domestic AI Necessary? Daisuke Okanohara, CEO of Preferred Networks, points out the importance of developing domestic AI and having it as an option, explaining why the “option” of domestic AI is necessary from the perspectives of the economy, culture, and national security. He argues that it is premature to give up on Japan’s AI industry as a lost cause, and emphasizes an approach that involves collaboration with other companies and exploring uniquely Japanese ways of competing under constraints such as limited power supply. While competing with frontier models, he presents a stance focused on ultimately delivering products and solutions that are chosen by users. I had a generative AI summarize the content. Furthermore, the results were turned into infographics and slide materials using NotebookLM. Please note that the investigations and analyses conducted by generative AI are based solely on publicly available information and do not necessarily reflect the actual situation; they may also contain inaccuracies. Please keep this in mind when referring to them. 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. 『AI時代の競争力とは「発明力」ではなく「権利化力」である』というリーガルテック株式会社 平井智之 代表取締役CEO のNoteへの投稿を読みました。「そうではないはずだ、そうあっては欲しくない」という気持ちの一方、「そうなってしまうのかもしれない」という感想も持ちました。 『AI時代の競争力とは「発明力」ではなく「権利化力」である』という主張について生成AIに深掘りさせました。さらに、結果をNotebookLMでインフォグラフィック、スライド資料にさせました。 なお、生成AIによる調査・分析結果は、公開された情報からだけの分析であり、必ずしも実情を示したものではないこと、誤った情報も含まれていることについてはご留意されたうえで、ご参照ください。 AI時代の競争力とは「発明力」ではなく「権利化力」である https://note.com/yutori_jd/n/n64f2293db98d In the AI Era, Is Competitiveness About “Patentability Power” Rather Than “Inventive Power”? I read a post on Note by Tomoyuki Hirai, President and CEO of LegalTech Co., Ltd., titled “In the AI era, competitiveness lies not in ‘inventive power’ but in ‘patentability power.’” While part of me felt, “That shouldn’t be the case—I hope it isn’t,” I also found myself thinking, “Perhaps that is what it will come to.” I asked generative AI to take a deeper look into the claim that “In the AI era, competitiveness lies not in ‘inventive power’ but in ‘patentability power.’” Furthermore, the results were turned into infographics and slide materials using NotebookLM. Please note that the research and analysis conducted by generative AI are based solely on publicly available information and therefore may not accurately reflect actual circumstances; they may also contain inaccuracies. Please review the content with this in mind. 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. 最新生成AI であるChatGPT-5.2とGoogle Gemini 3.0の医療分野への応用可能性を生成AI (ChatGPT-5.2とGoogle Gemini 3.0)に深掘りさせました。さらに、結果をNotebookLMでインフォグラフィック、スライド資料にさせました。 なお、生成AIによる調査・分析結果は、公開された情報からだけの分析であり、必ずしも実情を示したものではないこと、誤った情報も含まれていることについてはご留意されたうえで、ご参照ください。 Potential Applications of the Latest Generative AI in the Medical Field I conducted an in-depth exploration, using generative AI itself (ChatGPT-5.2 and Google Gemini 3.0), of the potential applications of the latest generative AI models—ChatGPT-5.2 and Google Gemini 3.0—in the medical field. The results were then converted into infographics and slide materials using NotebookLM. Please note that the research and analysis conducted by generative AI are based solely on publicly available information and may not necessarily reflect real-world conditions. They may also contain inaccuracies; therefore, please review and use the information with these limitations in mind. 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. OpenAIの最新モデルであるGPT-5.2 Thinkingが、AIの経済的価値を測るというOpenAIのAI評価指標「GDPval(Gross Domestic Product valuable tasks)」で専門家に対し70%以上の勝率を記録したことから、専門業務で人間専門家レベルに到達した最先端モデルと評価されています。 この「GDPval」は、従来の学術的テストとは異なり、米国GDPに貢献度の高い9セクター44職種の専門的な知識労働タスク(文書作成や分析など)におけるAIの性能を、人間の専門家とのブラインド比較を通じて測定しています。 この「GDPval」について、生成AIに深掘りさせました。さらに、結果をNotebookLMでインフォグラフィック、スライド資料にさせました。 なお、生成AIによる調査・分析結果は、公開された情報からだけの分析であり、必ずしも実情を示したものではないこと、誤った情報も含まれていることについてはご留意されたうえで、ご参照ください。 “GDPval”: Measuring the Economic Value of AI OpenAI’s latest model, GPT-5.2 Thinking, has achieved a win rate of over 70% against human experts on OpenAI’s AI evaluation metric known as “GDPval (Gross Domestic Product valuable tasks)”, which is designed to measure the economic value of AI. As a result, it is being evaluated as a state-of-the-art model that has reached human expert–level performance in professional knowledge work. Unlike conventional academic benchmarks, GDPval measures AI performance through blind comparisons with human experts on professional knowledge-worker tasks—such as document drafting and analysis—across 44 occupations in 9 sectors that make high contributions to U.S. GDP. I asked generative AI to conduct an in-depth analysis of GDPval, and further had the results transformed into infographics and presentation slides using NotebookLM. Please note that the investigations and analyses conducted by generative AI are based solely on publicly available information and do not necessarily reflect actual conditions; they may also contain inaccuracies. Kindly keep this in mind when referring to the materials. 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. 生成AIが発明や科学的発見を創出する技術的メカニズムと、それが研究開発(R&D)にもたらす戦略的影響について生成AIに深掘りさせました。さらに、結果をNotebookLMでインフォグラフィック、スライド資料にさせました。 なお、生成AIによる調査・分析結果は、公開された情報からだけの分析であり、必ずしも実情を示したものではないこと、誤った情報も含まれていることについてはご留意されたうえで、ご参照ください。 The Mechanisms by Which Generative AI Creates Inventions and Scientific Discoveries I asked generative AI to conduct an in-depth analysis of the technical mechanisms through which generative AI produces inventions and scientific discoveries, as well as the strategic impact these mechanisms have on research and development (R&D). The results were further converted into infographics and slide materials using NotebookLM. Please note that the research and analysis conducted by generative AI are based solely on publicly available information and therefore may not necessarily reflect actual conditions. The content may also contain inaccuracies, so please review and use it with appropriate caution. 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. 著名な免疫学者であるDerya Unutmaz博士は、2025年12月にリリースされたOpenAIのGPT-5.2 Proを、単なるツールではなく「AGI(汎用人工知能)に近いパートナー」と評価し、その創造的な問題解決能力が人間の専門家のそれに匹敵すると論じています。 生成AIに本件について深掘りさせました。さらに、結果をNotebookLMでインフォグラフィック、スライド資料にさせました。 なお、生成AIによる調査・分析結果は、公開された情報からだけの分析であり、必ずしも実情を示したものではないこと、誤った情報も含まれていることについてはご留意されたうえで、ご参照ください。 I can confirm this for GPT-5.2 Pro in biomedical research for sure! “Moreover, we believe GPT-5.2 is the world’s best model for assisting and accelerating scientists.” https://x.com/DeryaTR_/status/1999232994733699190 GPT-5.2 Pro as a “Partner Close to AGI (Artificial General Intelligence)” Renowned immunologist Dr. Derya Unutmaz has described OpenAI’s GPT-5.2 Pro, released in December 2025, not merely as a tool but as a “partner close to AGI (Artificial General Intelligence),” arguing that its creative problem-solving capabilities are comparable to those of human experts. We asked generative AI to conduct an in-depth analysis of this topic. The results were further transformed into infographics and presentation slides using NotebookLM. Please note that the findings derived from generative AI are based solely on publicly available information and do not necessarily reflect the actual situation. They may also contain inaccuracies; readers are advised to consult the material with these limitations in mind. 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. 大規模言語モデル(LLM)は、単なる「次単語予測」マシンだったはずなのに、発明を生み出すまでに急成長しました。なぜ発明を生み出すまでになったのかについては、諸説あるようです。これらの議論を生成AIに調査させました。さらに、結果をNotebookLMでインフォグラフィック、スライド資料にさせました。 なお、生成AIによる調査・分析結果は、公開された情報からだけの分析であり、必ずしも実情を示したものではないこと、誤った情報も含まれていることについてはご留意されたうえで、ご参照ください。 Large Language Models (LLMs) Are Not Merely “Next-Token Prediction” Machines Large language models (LLMs) were originally thought to be nothing more than “next-token prediction” machines, yet they have rapidly evolved to the point of generating inventions. There appear to be various theories as to why they have reached the capability of producing inventions. I asked generative AI to investigate these debates, and further had the results turned into infographics and presentation slides using NotebookLM. Please note that the findings and analyses produced by generative AI are based solely on publicly available information and may not necessarily reflect actual circumstances. They may also contain inaccuracies, so please review them with this in mind. 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. 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年12月4日にGemini 3 Deep Think(Google DeepMind)がリリースされ、2025年12月11日にChatGPT-5.2 Pro(OpenAI)がリリースされました。 2025年12月12日現在の技術水準に基づき、ChatGPT-5.2 Pro(OpenAI)とGemini 3 Deep Think(Google DeepMind)を比較して、画期的な発明を生むのはどちらか、生成AI (Gemini 3 Deep ThinkとChatGPT-5.2 Pro)に聞いてみました。さらに、結果をNotebookLMでインフォグラフィック、スライド資料にさせました。 なお、生成AIによる調査・分析結果は、公開された情報からだけの分析であり、必ずしも実情を示したものではないこと、誤った情報も含まれていることについてはご留意されたうえで、ご参照ください。 Which Is More Likely to Produce Breakthrough Inventions: ChatGPT-5.2 Pro or Gemini 3 Deep Think? Gemini 3 Deep Think (Google DeepMind) was released on December 4, 2025, followed by the release of ChatGPT-5.2 Pro (OpenAI) on December 11, 2025. Based on the state of the technology as of December 12, 2025, I compared ChatGPT-5.2 Pro (OpenAI) and Gemini 3 Deep Think (Google DeepMind) and asked generative AI itself which of the two is more likely to produce breakthrough inventions. I then turned the results into infographics and slide materials using NotebookLM. Please note that the findings and analyses produced by generative AI are based solely on publicly available information and do not necessarily reflect the actual situation. They may also contain inaccuracies, and readers are advised to keep this in mind when referring to them. 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. 2025年12月11日に、ディズニーがOpenAIに10億ドル(約1500億円)を出資し、動画生成AI「Sora」にミッキーマウスやマーベルなどの主要キャラクター200体以上をライセンス供与するという戦略的提携に関する記事が出ました。 ディズニーの狙いは、生成AIを脅威から共存へと戦略転換し、UGC(ユーザー生成コンテンツ)を公式に取り込むことで、Disney+のエンゲージメント向上と制作コスト効率化を図る点にあり、この提携は、GoogleやMidjourneyなど他のAI企業への法的圧力を強めるための「ライセンス市場の確立」という知財戦略の側面も持っているようです。 生成AIに本件について深掘りさせました。さらに、結果をNotebookLMでインフォグラフィック、スライド資料にさせました。 なお、生成AIによる調査・分析結果は、公開された情報からだけの分析であり、必ずしも実情を示したものではないこと、誤った情報も含まれていることについてはご留意されたうえで、ご参照ください。 ディズニー、OpenAIに1500億円出資 動画AI「Sora」にキャラ提供 https://www.nikkei.com/article/DGXZQOGN11CN50R11C25A2000000/ Disney Invests ¥150 Billion in OpenAI, Licenses Characters to Video AI “Sora” On December 11, 2025, an article reported on a strategic partnership in which Disney invested USD 1 billion (approximately ¥150 billion) in OpenAI and licensed more than 200 of its major characters—including Mickey Mouse and Marvel properties—to the video-generation AI “Sora.” Disney’s objective appears to be a strategic shift from viewing generative AI as a threat to pursuing coexistence: by officially incorporating UGC (user-generated content), Disney aims to increase engagement on Disney+ while improving production cost efficiency. At the same time, this partnership also seems to carry an intellectual-property strategy dimension—namely, the establishment of a “licensing market”—designed to strengthen legal pressure on other AI companies such as Google and Midjourney. I asked a generative AI to conduct an in-depth analysis of this development, and the results were further transformed into infographics and slide materials using NotebookLM. Please note that the research and analysis produced by generative AI are based solely on publicly available information and may not fully reflect the actual situation; they may also contain inaccuracies. Readers are advised to review the content with these limitations in mind. 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. OpenAIが2025年12月11日リリースした最新の大規模言語モデル「GPT-5.2」は、長文コンテキスト理解やプロフェッショナルな知識労働タスク、コーディング能力など多方面で性能が大幅に向上したと評価されており、特に企業向けの実用性が強調されています。 ベンチマークでは、ARC-AGI-2 LeaderboardでGemini Deep Thinkを上回る高い評価がある一方、AI performance on a set of Ph.D.-level science questionsではほとんど前モデルと変わらなくGemini 3 proに及ばないという評価もあります。 生成AIに「GPT-5.2」の評価・評判について調査させました。さらに、結果をNotebookLMでインフォグラフィック、スライド資料にさせました。 なお、生成AIによる調査・分析結果は、公開された情報からだけの分析であり、必ずしも実情を示したものではないこと、誤った情報も含まれていることについてはご留意されたうえで、ご参照ください。 2025年12月11日GPT-5.2 が登場 https://openai.com/ja-JP/index/introducing-gpt-5-2/ The Arrival of GPT-5.2 OpenAI released its latest large language model, “GPT-5.2,” on December 11, 2025. GPT-5.2 has been highly evaluated for its significant performance improvements across a wide range of areas, including long-context understanding, professional knowledge-work tasks, and coding capabilities. In particular, its practicality for enterprise use has been strongly emphasized. In benchmark evaluations, GPT-5.2 has received high ratings, surpassing Gemini Deep Think on the ARC-AGI-2 Leaderboard. On the other hand, in assessments of AI performance on a set of Ph.D.-level science questions, some evaluations indicate that its performance is almost unchanged from the previous model and does not reach the level of Gemini 3 Pro. I asked a generative AI to investigate the evaluations and public reception of GPT-5.2, and then used NotebookLM to turn the results into infographics and presentation slides. Please note that the findings produced by the generative AI are based solely on publicly available information and do not necessarily reflect the actual situation. They may also contain inaccuracies, so please review them with this in mind. 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. Your browser does not support viewing this document. Click here to download the document. デロイトは、2025年12月8日、「日本の知財金融の推進に向けた施策の提案 米国と中国の事例を参考にして」を発表し、2026年5月から事業性融資の推進等に関する法律が施行され、知的財産など無形資産を含む新たな資金調達手段である企業価値担保権によって、中小企業やスタートアップの支援を目的とした知財金融の推進が図られている中、知財金融において日本よりも先進している米国・中国の事例をもとに、日本が今後取り組むべき施策の方向性を提示しました。 知財金融とは、企業の知的財産(特許、商標など)に着目して融資を行う取り組みであり、無形資産へと価値の中心が移行する中で中小企業やスタートアップの資金調達手段として重要性が増しています。しかし、日本では知財の評価・管理負担や貸倒リスクの高さが普及の障壁となっており、融資元の負担軽減やリスク分散を目的とした制度整備の必要性が指摘されています。結論として、海外事例を踏まえ、評価・管理サービスへの補助金提供や、貸倒リスクを補填する仕組みの構築が、日本の知財金融普及に有効であると提言されています。 生成AIに、日本の知財金融に関して深掘りさせたうえで、デロイトの提案について評価させました。さらに、結果をNotebookLMでインフォグラフィック、スライド資料にさせました。 なお、生成AIによる調査・分析結果は、公開された情報からだけの分析であり、必ずしも実情を示したものではないこと、誤った情報も含まれていることについてはご留意されたうえで、ご参照ください。 日本の知財金融の推進に向けた施策の提案 米国と中国の事例を参考にして https://www.deloitte.com/jp/ja/services/consulting/perspectives/intellectual-property-finance-japan.html Deloitte’s “Policy Proposals for Promoting IP Finance in Japan” On December 8, 2025, Deloitte released a report titled “Policy Proposals for Promoting IP Finance in Japan: Learning from the U.S. and Chinese Cases.” With the Act on the Promotion of Business-Based Lending scheduled to take effect in May 2026, Japan is aiming to advance IP finance as a means of supporting SMEs and startups through new financing mechanisms such as enterprise-value security interests, which include intangible assets such as intellectual property. In this context, the report outlines policy directions Japan should pursue, drawing on examples from the United States and China, which are more advanced in the field of IP finance. IP finance refers to lending practices that focus on a company’s intellectual property—such as patents and trademarks—as collateral or as a basis for assessing creditworthiness. As corporate value increasingly shifts toward intangible assets, IP finance is becoming a more important funding option for SMEs and startups. However, in Japan, factors such as the burden of IP valuation and management, as well as the high risk of loan defaults, have hindered widespread adoption. The report therefore highlights the need for institutional arrangements that reduce lenders’ operational burden and diversify risk. Based on overseas cases, it concludes that providing subsidies for valuation and management services and establishing mechanisms to compensate for credit losses would be effective measures for expanding IP finance in Japan. Generative AI was used to conduct an in-depth analysis of Japan’s IP finance landscape and to evaluate Deloitte’s proposals. The findings were then transformed into infographics and slide materials via NotebookLM. Please note that the research and analysis performed by generative AI are based solely on publicly available information and may not necessarily reflect the actual circumstances. The results may also include inaccuracies, so please review them with this in mind. 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 |
RSS Feed