📖 英文と日本語訳（一文ずつ）

🎧 通し読み（全文）

リスニング・シャドーイング用の全文です。

OpenAI announced on May 5, 2026, that it has updated the standard ChatGPT model to "GPT-5.5 Instant." Instant is a foundation model used daily by hundreds of millions of users, and its latest update enhances response accuracy, conciseness, and personalization. In terms of factual accuracy, internal evaluations show that false claims (hallucinations) in high-difficulty prompts where precision is critical, such as medicine, law, and finance, have decreased by 52.5% compared to the conventional GPT-5.3 Instant. Furthermore, inaccurate claims reportedly decreased by 37.3% even in particularly challenging conversations flagged by users for factual errors. In a mathematical example released by OpenAI, GPT-5.5 Instant demonstrated behavior where it initially affirmed an incorrect answer but identified an algebraic error during a verification process of substituting back into the original equation, ultimately reaching the correct solution using the quadratic formula. In contrast, GPT-5.3 Instant reached the incorrect conclusion of "no real solutions" during the verification process. Furthermore, accuracy improved from 75.0% to 81.6% on CharXiv-reasoning, which measures scientific chart reasoning, and from 69.2% to 76.0% on MMMU-Pro, which evaluates specialized multimodal reasoning. Response styles have been revised to minimize redundant expressions and excessive formatting while maintaining the richness of the content. In a comparative example regarding advice on dealing with workplace colleagues, GPT-5.5 Instant generated a more concise and practical response, using 30.2% fewer words and 29.2% fewer lines than previous models. Adjustments have been made to refrain from unnecessary follow-up questions and the excessive use of emojis. In terms of personalization, the system can now more effectively reference past chats, files, and information from connected Gmail. This reduces the need for users to repeatedly enter the same information and enables personalized suggestions tailored to specific contexts, such as when handing over ongoing tasks. In an example featured in the article, GPT-5.5 Instant provides tea shop recommendations based on the user's regular habits and personal preferences. Additionally, the "Memory Source" feature will be introduced to all ChatGPT models. This system visualizes information sources used for personalizing responses, such as stored memory and past chats, allowing users to delete or modify outdated or unnecessary information. Memory sources are not visible to others in shared chats, and users can also choose temporary chats that do not use memory. This display identifies the most relevant sources; however, it does not necessarily encompass all factors that influenced the response. GPT-5.5 Instant is rolling out to all ChatGPT users starting today and is available as "chat-latest" via the API. For paid users, the legacy GPT-5.3 Instant will remain available in the model settings for three months before being discontinued. Personalization features leveraging past chats, files, and Gmail will initially be available to Plus and Pro users on the web, with a sequential rollout to mobile and other plans to follow.

📝 学習のヒント

1. 1まず英文を読む — 知らない単語にあたりをつけてから音声へ。
2. 2一文ずつ確認 — 日本語訳と照合し、構文を理解する。
3. 3通し読み Normal で耳を作る — 内容を追いながらリピート。
4. 4Fast でシャドーイング — 口を慣らし、リスニング速度を上げる。
5. 5翌日に復習 — 1日空けて再聴すると長期記憶に定着しやすい。