|
OpenAIのサム・アルトマン最高経営責任者(CEO)はインドで開催された「India AI Impact Summit 2026」(2026年2月16日 – 2026年2月20日)で講演し、2028年末までにAIが人類の知的能力を上回る超知能に到達するとの見通しを示しました。また、AI技術が特定の国家や企業に独占されることによる全体主義のリスクを警告し、技術の恩恵を分散するAIの民主化と国際的な枠組みによるガバナンスの必要性を訴えました。 India AI Impact Summit 2026でのサム・アルトマンCEOの講演について、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。 OpenAIのサム・アルトマンCEO「2028年までにAIが人類の知能を超える超知性に到達」 AIが特定の国家や企業に独占されることによる全体主義のリスクを警告 https://www.sbbit.jp/article/cont1/181280#image230099%E3%80%8F OpenAIのサム・アルトマンCEO「2028年までにAIが人類の知能を超える超知性に到達」 2/22 https://news.yahoo.co.jp/articles/b7a98232dacfd61f59ac2de1bdf0258be72528fe AI to Reach Superintelligence Surpassing Human Intelligence by 2028 Sam Altman, CEO of OpenAI, delivered a speech at the India AI Impact Summit 2026 (February 16–20, 2026), held in India, where he expressed his outlook that by the end of 2028, AI will reach superintelligence surpassing human intellectual capabilities. He also warned of the risk of totalitarianism arising from the concentration of AI technology in the hands of specific nations or corporations, and emphasized the need for democratizing AI to distribute its benefits broadly, as well as establishing international governance frameworks. I asked generative AI to provide an in-depth analysis of CEO Sam Altman’s speech at the India AI Impact Summit 2026. Furthermore, I used NotebookLM to convert the results into infographics and presentation slides. Please note that the investigation and analysis conducted by generative AI are based solely on publicly available information and do not necessarily reflect the full reality of the situation. They may also contain inaccuracies. I ask that you review the materials 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. 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.
0 Comments
xAIが2026年2月17日にリリースした「Grok 4.20」は、4つの専門AIエージェントが同時並行で同一の問題に取り組み、複数の視点から解答を統合するアーキテクチャとなっています。この仕組みにより、一つの視点だけでは気づかない問題点を拾い上げられる精度が上がります。 Grok 4.20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。 “Grok 4.20”: Four Expert AIs Deliver a Unified Answer “Grok 4.20,” released on February 17, 2026 by xAI (the AI company led by Elon Musk), adopts an architecture in which four specialized AI agents simultaneously work on the same problem in parallel and integrate their answers from multiple perspectives. This mechanism improves accuracy by identifying issues that might be overlooked from a single viewpoint. I asked a generative AI to conduct an in-depth analysis of Grok 4.20. Furthermore, the results were turned into an infographic and slide materials using NotebookLM. Please note that the research and analysis conducted by generative AI are based solely on publicly available information and do not necessarily reflect actual conditions. They may also contain inaccuracies. Kindly review the materials 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. 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は2026年2月13日、「GPT‑5.2、理論物理学における新たな結果を導き出す:新しいプレプリントによると、GPT‑5.2 がグルーオン振幅の公式を提案し、その後、内部のOpenAI モデルによって証明されたことが執筆陣によって確認されました。」と発表しました。それが「GPT5.2が理論物理学の定説を覆し、シンプルな新公式を発見」、「OpenAIのGPT-5.2が物理学の新たな式を作り出すことに成功」と報道されています。 この内容を生成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。 GPT‑5.2、理論物理学における新たな結果を導き出す https://openai.com/ja-JP/index/new-result-theoretical-physics/ GPT5.2が理論物理学の定説を覆し、シンプルな新公式を発見 https://www.sbbit.jp/article/cont1/180888 OpenAIのGPT-5.2が物理学の新たな式を作り出すことに成功 https://gigazine.net/news/20260216-openai-gpt-5-2-theoretical-physics/ GPT-5.2 Overturns a Prevailing View in Theoretical Physics On February 13, 2026, OpenAI announced: “GPT-5.2 Derives a New Result in Theoretical Physics: According to a new preprint, GPT-5.2 proposed a formula for gluon amplitudes, which was subsequently proven by an internal OpenAI model, as confirmed by the authors.” This development has been widely reported under headlines such as “GPT-5.2 Overturns Conventional Wisdom in Theoretical Physics and Discovers a Simple New Formula” and “OpenAI’s GPT-5.2 Successfully Generates a New Equation in Physics.” I asked generative AI to conduct a deeper analysis of this development. The results were further compiled into infographics and presentation slides using NotebookLM. Please note that the research and analytical results generated by AI are based solely on publicly available information. They do not necessarily reflect the full reality of the situation and may contain inaccuracies. Kindly review the materials 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. 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. AIの「流動的知能」(未知の状況への適応能力)を測定するベンチマークである「ARC-AGI-2」におけるLLMモデル間の性能差は極めて大きくなっています。ARC-AGI-2が測定する「抽象推論」「新規パターンへの適応」「一般化能力」は、特許文書の本質的理解、先行技術調査の精度、発明の新規性・進歩性判断において重要です。 「ARC-AGI-2」におけるLLMモデル間の性能差が知的財産業務へどう影響するのか、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。 How Performance Gaps in “ARC-AGI-2” Impact Intellectual Property Practice The performance gap among LLM models on “ARC-AGI-2,” a benchmark designed to measure AI’s “fluid intelligence” (the ability to adapt to novel situations), has become extremely pronounced. The capabilities assessed by ARC-AGI-2—such as abstract reasoning, adaptation to new patterns, and generalization—are critical for fundamentally understanding patent documents, improving the accuracy of prior art searches, and evaluating the novelty and inventive step of inventions. I asked generative AI to conduct an in-depth analysis of how performance differences among LLM models on “ARC-AGI-2” may affect intellectual property practice. In addition, the results were transformed into infographics and presentation slides using NotebookLM. Please note that the investigation and analysis conducted by generative AI are based solely on publicly available information and may not necessarily reflect actual circumstances. The results may also contain inaccuracies. Please 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. 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3月にARC Prize Foundationが公開したARC-AGI-2は、AIの「流動的知能」(未知の状況への適応能力)を測定するベンチマークです。 2026年2月現在、DeepSeekやQwen等の中国製LLMは、SWE-Bench、AIME、GPQA-Diamondなど従来型ベンチマークで欧米フロンティアモデルに匹敵するスコアを叩き出していますが、ARC-AGI-2においては大幅に低いスコアにとどまっています。 この中国製LLMがARC-AGl-2で苦戦している状況を生成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。 Chinese LLMs Struggling on ARC-AGI-2 ARC-AGI-2, released by the ARC Prize Foundation in March 2025, is a benchmark designed to measure AI’s “fluid intelligence” — its ability to adapt to novel situations. As of February 2026, Chinese-developed LLMs such as DeepSeek and Qwen have achieved scores comparable to Western frontier models on conventional benchmarks like SWE-Bench, AIME, and GPQA-Diamond. However, on ARC-AGI-2, their scores remain significantly lower. I asked a generative AI system to conduct an in-depth analysis of why Chinese LLMs are struggling on ARC-AGI-2. The results were then 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 do not necessarily reflect actual conditions. They may also contain inaccuracies. I ask that you keep this in mind when reviewing 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. 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. Googleが2026年2月19日にリリースした「Gemini 3.1 Pro」によって、企業の知的財産業務はどう変わるか、知的財産業務にもたらすインパクトについて、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 Impact of Gemini 3.1 Pro on Intellectual Property Operations With the release of Gemini 3.1 Pro by Google on February 19, 2026, I explored how this new model could transform corporate intellectual property (IP) operations and examined the potential impact it may have on IP practices. I asked generative AI to conduct an in-depth analysis of these implications. Furthermore, the results were converted into infographics and slide materials using NotebookLM. Please note that the research and analysis generated by AI are based solely on publicly available information and do not necessarily reflect actual circumstances. They may also contain inaccuracies, so please review the materials 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. 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. Googleは、2026年2月19日にGemini 3.1 Proを発表しました。本モデルは、未知の課題への適応力を示す「ARC-AGI-2」ベンチマークで77.1%という高得点を記録し、競合他社と比較した際の圧倒的なコストパフォーマンスを示しています。 2026年2月12日に発表された特殊推論モード「Gemini 3 Deep Think」(「ARC-AGI-2」ベンチマークで84.6%、高価格帯ユーザーしか使えない)とは、異なるアプローチに基づくシステムで、Gemini 3.1 Proは前モデルであるGemini 3.0 Proと同一の価格設定を維持しながらその実質的な能力は世代交代レベルの飛躍を遂げているということです。 Gemini 3.1 Pro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。 グーグル「Gemini 3.1 Pro」発表、“AGIへの登竜門”で高得点 2/20(金) https://news.yahoo.co.jp/articles/09a9a78c3e217f18d221159d6861efd5775de105 Gemini 3.1 Pro の概要 https://note.com/npaka/n/nc859d1b06094 【徹底解説】Gemini 3.1 Pro 登場。実力を4モデル比較で検証 https://chatgpt-lab.com/n/n64d9f440c178 February 20 – Gemini 3.1 Pro Achieves 77.1% on “ARC-AGI-2” On February 19, 2026, Google announced Gemini 3.1 Pro. The model achieved a high score of 77.1% on the “ARC-AGI-2” benchmark, which measures adaptability to novel tasks, demonstrating outstanding cost performance compared with competing models. Unlike the specialized reasoning mode “Gemini 3 Deep Think,” announced on February 12, 2026 (which scored 84.6% on ARC-AGI-2 but is available only to high-priced-tier users), Gemini 3.1 Pro is based on a different architectural approach. Notably, while maintaining the same pricing as its predecessor, Gemini 3.0 Pro, Gemini 3.1 Pro is said to deliver a generational leap in practical capability. I asked generative AI to conduct an in-depth analysis of Gemini 3.1 Pro. Furthermore, the results were transformed 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 actual conditions. The results may also contain inaccuracies. I kindly ask that you review the material 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. 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. 発明抽出・特許調査・発明提案書作成を生成AIで “一気通貫”させることが出来るようになってきました。 発明の着想から特許調査、そして提案書の作成までを一つのシステムで完結させる「一気通貫」の知的財産ツールについて、生成AIに調査させ、2026年時点の最新動向を整理・比較した調査資料としてまとめさせました。 それぞれのAIがどうとらえどう調べどうまとめたか、ご確認ください。 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。 End-to-End Invention Extraction, Patent Search, and Invention Proposal Drafting with Generative AI It has now become possible to use generative AI to achieve a fully integrated, “end-to-end” workflow covering invention extraction, patent search, and invention proposal drafting. I asked generative AI to investigate intellectual property tools that complete the entire process—from the initial conception of an invention, through patent searching, to the preparation of an invention proposal document—within a single system. The findings were compiled into a research report that organizes and compares the latest trends as of 2026. Please review how each AI interpreted the task, conducted its research, and structured its conclusions. Please note that the research and analysis results generated by AI are based solely on publicly available information and may not necessarily reflect actual conditions. They may also contain inaccuracies. Kindly refer to the material with these considerations 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. 2026年2月16日に開催された「AIネットワーク社会推進会議 AIガバナンス検討会(第29回)」において、AI事業者ガイドラインの改定案が示され、議論が進められています。 配布された資料「AI事業者ガイドラインの令和7年度更新内容(案)」によると、普及が進むAIエージェントや、ロボットなどを制御するフィジカルAIに関する定義や便益、リスク、対策が新たに盛り込まれ、2026年3月末に第1.2版を公開する予定としています。 生成AIに、AI事業者ガイドライン改定案について深堀させました。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ネットワーク社会推進会議 AIガバナンス検討会(第29回) https://www.soumu.go.jp/main_sosiki/kenkyu/ai_network/02tsushin06_04000136.html AIエージェントやロボAI「人の判断必須の仕組みを」 政府指針に明記 https://www.nikkei.com/article/DGXZQOUA136YP0T10C26A2000000/ 2026/2/17 [TUE] 政府、AI事業者ガイドライン改定案でAIエージェントとフィジカルAIを追加──「人間の判断必須の仕組み」明記、Xで議論広がる https://ledge.ai/articles/government_ai_guideline_revision_human_judgment_required_x_debate Proposed Revision to the AI Business Operator Guidelines At the 29th meeting of the AI Governance Study Group under the AI Network Society Promotion Council, held on February 16, 2026, a proposed revision to the AI Business Operator Guidelines was presented and discussions are currently underway. According to the distributed document titled “Proposed Updates to the AI Business Operator Guidelines for FY2025 (Reiwa 7)”, new sections have been added addressing definitions, benefits, risks, and countermeasures related to increasingly widespread AI agents and physical AI systems that control robots and other hardware. The government plans to publish Version 1.2 of the revised guidelines by the end of March 2026. I asked generative AI to conduct an in-depth analysis of the proposed revision to the AI Business Operator Guidelines. Furthermore, the results were transformed 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 actual circumstances. The results may also contain inaccuracies. We ask that you review the materials 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. 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. 「最新の生成AIモデルを用いて研究開発の現場でどんな革新が行われているか」を生成AIに調べさせました。プロンプトがあいまいなため、非常に面白い結果が返ってきました。 それぞれのAIがどうとらえどう調べどうまとめたか、ご確認ください。 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。 Innovation in R&D Sites Driven by the Latest Generative AI I asked generative AI to investigate “what kinds of innovations are being carried out in research and development settings using the latest generative AI models.” Because the prompt was somewhat ambiguous, the results returned were extremely interesting. Please take a look at how each AI interpreted the task, how it conducted its research, and how it structured and summarized its findings. Please note that the research and analysis results generated by AI are based solely on publicly available information. They do not necessarily reflect the actual situation and may contain inaccuracies. Kindly review them 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. 生成AIを用いたFTO(Freedom to Operate:侵害予防・技術動向)調査が知財実務において急速に導入が進んでおり、従来の調査手法を大きく変革しています。 一般的なFTO調査は、①製品技術要素の分解、②母集団の検索・形成、③スクリーニング、④重要特許のクレーム対比、⑤レポート化という流れで行われます。製品の技術要素を細かくリストアップし、分類+キーワードの検索式で母集団を作成し、そこから請求項レベルで自社構成と対比してオールエレメントルールに基づき侵害可能性を検討する、というのが基本です。 FTO調査プロセスの中流〜下流に強みがあるSummaria (サマリア)と FTO調査プロセスの上流〜下流全体をカバーする設計のPatsnap Eurekaについて、生成AIにそれぞれのFTO調査アプローチを比較させました。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。 FTO Investigations Using Generative AI FTO (Freedom to Operate: infringement prevention and technology trend) investigations utilizing generative AI are rapidly being adopted in intellectual property practice, significantly transforming conventional research methodologies. A typical FTO investigation follows this process: (1) decomposition of product technical elements; (2) search and construction of the patent corpus; (3) screening; (4) claim chart comparison of key patents; and (5) report preparation. In principle, the product’s technical elements are broken down into detailed components, a patent set is constructed using classification codes and keyword-based search queries, and the claims are then compared against the company’s product configuration at the claim level. Based on the all-elements rule, the possibility of infringement is examined. I asked generative AI to compare the FTO investigation approaches of Summaria—which demonstrates strengths in the midstream to downstream stages of the FTO process—and Patsnap Eureka, which is designed to cover the entire FTO workflow from upstream to downstream. Furthermore, the results were transformed into infographics and slide materials using NotebookLM. Please note that the research and analytical results generated by AI are based solely on publicly available information and do not necessarily reflect actual conditions. They may also contain inaccuracies. Kindly review them 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. 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. 2026年2月11日、Google DeepMindは、開発したAI研究エージェント「Aletheia(アレテイア)」と、その基盤技術である「Gemini 3 Deep Think」による科学的成果を発表しました。 Aletheiaは、生成・検証・修正(GVR)という循環型の思考ループを用いることで、未解決のエルデシュ問題を含む高度な数学的難問に対して自律的な解決策を提示しました。特筆すべきは、AIが単なる計算補助に留まらず、創薬エンジンIsoDDEによる分子設計や物理学、経済学などの多領域で自律的な研究者として機能し始めている点です。 一方で、AIの貢献を客観的に評価するために、DeepMindは自動運転になぞらえた数学的研究の自律性レベルという新たな指標を提唱しており、人間とAIが協力して未知の真理を探究する新たな科学的方法論の幕開けを象徴する内容となっています。 この「Aletheia(アレテイア)」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。 February 11, 2026 Accelerating Mathematical and Scientific Discovery with Gemini Deep Think Gemini Deep Thinkで数学と科学の発見を加速 https://deepmind.google/blog/accelerating-mathematical-and-scientific-discovery-with-gemini-deep-think/ AIが「科学者」になる日:Google DeepMindの『Aletheia』が数学の歴史を塗り替えた理由 https://zenn.dev/nakano_teppei/articles/9aecb89df91f1c Google DeepMind’s “Aletheia” Rewrites the History of Mathematics On February 11, 2026, Google DeepMind announced scientific achievements enabled by its AI research agent “Aletheia” and its underlying technology, “Gemini 3 Deep Think.” Aletheia employs a cyclical reasoning loop known as Generate–Verify–Revise (GVR), allowing it to autonomously propose solutions to advanced mathematical challenges, including previously unsolved Erdős problems. Notably, the AI has moved beyond serving merely as a computational assistant. It is beginning to function as an autonomous researcher across multiple domains—ranging from molecular design through the drug discovery engine IsoDDE, to physics and economics. At the same time, in order to objectively assess AI’s contributions, DeepMind has introduced a new metric inspired by levels of autonomous driving: levels of autonomy in mathematical research. This framework symbolizes the dawn of a new scientific methodology in which humans and AI collaborate to explore previously unknown truths. I asked generative AI to conduct an in-depth analysis of this “Aletheia” development. Furthermore, the results were transformed 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. They may not fully reflect actual circumstances and may contain inaccuracies. Readers are advised to keep this in mind when referring to the material. 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. サマリアに関する2026年2月12日付ニュースリリースによれば、2025年末にリリースされたサマリアの「調査支援ツール」の調査能力を客観的に検証するため、2024年度 特許検索競技大会の実問題を用いた検証を実施した結果、人手による特別な操作なしに、【電気分野】【機械分野】【化学・医薬分野】のすべての技術分野において、集合上位100件について再現率85%以上(100件中に正解特許の85%以上を含めることができた)という極めて高品質な検索集合作成ができることが確認できたということです。 このニュースリリースの内容について、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。 特許検索競技大会の問題に「サマリア」が挑戦。集合上位100件について85%以上の再現率を達成!/ 導入実績360社超を誇る中小企業向け知財管理システム「root ip クラウド」とサービス連携を開始 「調査支援ツール」の精度を検証〜電気・機械・化学・医薬の全分野において再現率85%以上(集合上位100件)の検索集合作成精度を確認!サービス連携第2弾。「考える」業務と「管理する」業務をシームレスに https://prtimes.jp/main/html/rd/p/000000018.000086119.html Summaria’s “Research Support Tool” Achieves 85% Recall According to a news release dated February 12, 2026, concerning Summaria, an objective verification test was conducted to evaluate the investigative capabilities of Summaria’s “Research Support Tool,” which was released at the end of 2025. The test utilized actual questions from the 2024 Patent Search Competition. The results confirmed that, without any special manual intervention, the tool was able to generate extremely high-quality search result sets across all technical fields—Electrical, Mechanical, and Chemical/Pharmaceutical. Specifically, for the top 100 retrieved documents, the recall rate exceeded 85% (i.e., more than 85% of the correct patents were included within the top 100 results). I asked a generative AI system to conduct an in-depth analysis of the contents of this news release. Furthermore, the results were transformed into infographics and presentation slides using NotebookLM. Please note that the investigation and analysis conducted by the generative AI are based solely on publicly available information. They do not necessarily reflect the full reality of the situation and may contain inaccuracies. Kindly review the materials with this consideration 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. 近年、PatsnapやAI.Tokkyoなどが提唱する「検索式を組まずに特許調査」を行うAIエージェントが注目を集めています。 生成AIに、これらの新しい特許調査手法の信頼性を評価させ、その現状と将来性について、公開情報、専門家の評価、実務導入事例に基づき多角的に分析させました。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 Reliability of AI-Driven “Patent Searches Without Search Queries” In recent years, AI agents that conduct patent searches “without constructing search queries,” as advocated by platforms such as Patsnap and AI.Tokkyo, have been attracting growing attention. I asked a generative AI system to evaluate the reliability of these new patent search methodologies and to analyze their current status and future prospects from multiple perspectives, based on publicly available information, expert evaluations, and practical implementation cases. Furthermore, the results were transformed into infographics and presentation slides using NotebookLM. Please note that the research and analysis conducted by the generative AI are based solely on publicly available information and may not necessarily reflect the actual situation. The analysis may also contain inaccuracies. Kindly review the content with these considerations 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. 2026年2月12日にOpenAIがリリースしたGPT-5.3-Codex-Spark は「リアルタイムでのコーディング体験(体感レイテンシ)」を最優先に設計された、Codex向けの超低遅延モデルです。位置づけとしては GPT-5.3-Codex の“小型版”で、高速な反復(小さな編集、UI調整、ロジック微修正、即時のフィードバック)に強みがあります。 このGPT-5.3-Codex-Spark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。 GPT‑5.3‑Codex‑Spark のご紹介 Codex でのリアルタイムコーディング向けに設計された高速モデルです。 https://openai.com/ja-JP/index/introducing-gpt-5-3-codex-spark/ GPT-5.3-Codex-Spark Excels at Real-Time Coding Released by OpenAI on February 12, 2026, GPT-5.3-Codex-Spark is an ultra-low-latency model for Codex, designed with top priority given to the “real-time coding experience” (perceived latency). Positioned as a smaller variant of GPT-5.3-Codex, it is particularly well-suited for rapid iteration—such as small edits, UI adjustments, minor logic refinements, and immediate feedback. I asked a generative AI system to conduct an in-depth analysis of GPT-5.3-Codex-Spark. Furthermore, I used NotebookLM to transform the results into an infographic and slide presentation. Please note that the research and analysis generated by AI are based solely on publicly available information and do not necessarily reflect the actual situation. They may also contain inaccuracies. I encourage you to review the material 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. Googleは2026年2月12日、「Gemini 3 Deep Think」のメジャーアップデートを公開しました。この新バージョンのGemini 3 Deep Thinkは、「人間にとっては容易いが、AIにとっては難しい課題」を解かせるベンチマークテスト「ARC-AGI-2」で、OpenAI「GPT-5.2」(52.9%)やAnthropic「Claude Opus 4.6」(68.8%)を上回る84.6%を達成するなど、科学分野を中心に、現行最高クラスの性能を実現しています。 この「Gemini 3 Deep Think」のメジャーアップデートについて、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。 Google、推論AI「Gemini 3 Deep Think」劇的に刷新 科学分野で金メダル級性能を達成 https://news.yahoo.co.jp/articles/fad88cfd86c4f36e40fa89297a35e4e1c667111c Major Update to Gemini 3 Deep Think, Specialized for Scientific Research On February 12, 2026, Google released a major update to “Gemini 3 Deep Think.” This new version achieved 84.6% on the benchmark test “ARC-AGI-2,” which is designed to present problems that are easy for humans but difficult for AI. In doing so, it outperformed OpenAI’s “GPT-5.2” (52.9%) and Anthropic’s “Claude Opus 4.6” (68.8%). With this result, Gemini 3 Deep Think has reached state-of-the-art performance, particularly in scientific domains. I asked generative AI to conduct an in-depth analysis of this major update to “Gemini 3 Deep Think.” Furthermore, I used NotebookLM to generate infographics and presentation slides based on the results. Please note that the research and analysis conducted by generative AI are based solely on publicly available information. They do not necessarily reflect the full reality of the situation and may contain inaccuracies. I ask that you 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. 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. 2023年頃の生成AI黎明期には、詳細な手順を記した長文プロンプトが必要とされ「プロンプトエンジニアリング」というまさに職人技が重要でした。しかし、2025年以降の技術的進展は、これを根本から覆し、単なる「指示の書き手」としての「プロンプト職人」は終焉を迎えています。 この「プロンプト職人の終焉」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。 プロンプトエンジニアリングの終焉!「プロンプト職人」からのパラダイムシフト「DSPy」「APE」「OPRO」を徹底解説! https://note.com/mbbs/n/n40b492e01bf6 The End of the Prompt Artisan In the early days of generative AI around 2023, long and highly detailed prompts outlining step-by-step instructions were considered essential, and “prompt engineering” was regarded as a true artisan skill. However, technological advances since 2025 have fundamentally overturned this premise, bringing an end to the era of the “prompt artisan” as merely a writer of instructions. I asked generative AI to conduct an in-depth analysis of this “end of the prompt artisan.” Furthermore, we used NotebookLM to transform the results into infographics and presentation slides. Please note that the research and analysis generated by AI are based solely on publicly available information. They may not necessarily reflect actual circumstances and may contain inaccuracies. I ask that you review the materials 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. フランスのAIスタートアップ企業Mistral(ミストラル)は、米国のハイテク企業に代わるサービスを求める欧州の企業や政府からの需要の高まりに乗って、過去1年間で収益が20倍に増加したと発表しました。 このMistralに関する記事について、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。 欧州がAlの独立を推進する中、ミストラルの収益は1年で20倍に増加 https://www.ft.com/content/664249e7-e8d5-4425-b397-ad3ed590b305 French AI Company Mistral’s Revenue Increases 20-Fold in One Year French AI startup Mistral announced that its revenue has grown twentyfold over the past year, driven by rising demand from European companies and governments seeking alternatives to U.S. high-tech firms. I asked generative AI to provide an in-depth analysis of articles related to Mistral. Furthermore, we used NotebookLM to transform the results into an infographic and slide presentation. Please note that the research and analysis conducted by generative AI are based solely on publicly available information and may not necessarily reflect the full reality. The findings may also contain inaccuracies. I ask that you review the materials 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. 2026年2月12日、中国のスタートアップ企業である智譜AI(Zhipu AI)が発表した最新AIモデル「GLM-5」は、複数のベンチマークでGemini 3.0 ProやGPT-5.2を超えるスコアを記録しており、エージェントとしてWord文書作成などの事務処理を実行することもでき、無料でダウンロード可能なオープンモデルとして公開されています。 米国の輸出規制下でNVIDIA製チップを使わず、ファーウェイ製のハードウェアと独自フレームワークのみで開発され、MoE(混合専門家)アーキテクチャの採用により、巨大なパラメータ数を持ちながら低コストでの推論を実現したということです。 このモデルは、従来の単純なテキスト生成から、AIが自律的に計画・実行を行う「エージェント・エンジニアリング」への転換を象徴していると言えます。 この「GLM-5」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。 2026年02月12日 Gemini 3.0 ProやGPT-5.2に匹敵する中国製AI「GLM-5」が登場、Word文書作成などの事務作業も可能でモデルを自由にダウンロード可能 https://gigazine.net/news/20260212-z-ai-glm-5/ Feb 12 2026 中国Zhipu AI、GPT-5に挑む745億パラメータ「GLM-5」を発表 https://tech-noisy.com/2026/02/12/zhipu-glm-5/ Gemini 3.0 Pro–Level Performance, Rivaling GPT-5.2: China’s AI “GLM-5” On February 12, 2026, Zhipu AI, a Chinese startup, announced its latest AI model, GLM-5. The model reportedly achieved scores surpassing Gemini 3.0 Pro and GPT-5.2 on multiple benchmarks. In addition to advanced language capabilities, it can function as an agent capable of performing administrative tasks such as creating Word documents. It has been released as a downloadable open model available free of charge. Developed under U.S. export restrictions without the use of NVIDIA chips, GLM-5 was built using Huawei hardware and Zhipu AI’s proprietary framework. By adopting a Mixture-of-Experts (MoE) architecture, the model achieves low-cost inference despite its massive parameter count. This model symbolizes a shift from conventional text generation toward “agent engineering,” in which AI autonomously plans and executes tasks. I asked generative AI to conduct an in-depth analysis of GLM-5. The results were then turned into infographics and presentation slides using NotebookLM. Please note that the investigation and analysis conducted by generative AI are based solely on publicly available information and may not necessarily reflect actual conditions. They may also contain inaccuracies. I ask that you review the materials 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. Your browser does not support viewing this document. Click here to download the document. 北海道大学サマーセミナー2026の日程が公開されました。今年は8月22日~25日です。
レベルの高い講義で、ハイブリッド開催(対面及びオンライン)です。 サマーセミナー2026の日程 https://www.juris.hokudai.ac.jp/riilp/news/seminar2026.html Hokkaido University Summer Seminar 2026 The schedule for the Hokkaido University Summer Seminar 2026 has been announced. This year, it will be held from August 22 to August 25. The seminar features high-level lectures and will be conducted in a hybrid format (both in-person and online). |
著者萬秀憲 アーカイブ
January 2026
カテゴリー |
RSS Feed