28 of 42

Chapter 13Questions people ask

Questions people ask

Short answers to the questions that come up most. Open any one; each points to the chapter that explains it.

Does it remember me?

The model itself doesn't. Between conversations, the harness may store notes ("memory") and add them to the context next time. Claude, for example, has memory you can view, edit and delete.¹

More in Chapter 2: How does it work?

Why does it make things up?

It generates plausible text, and training and testing have rewarded confident guesses over "I don't know."² Giving it sources, such as search results or your documents, reduces this.³

More in Chapter 1: What is a language model?

Is it searching the web right now?

Only if the harness has a search tool turned on and the model asks for it.⁴⁵ Otherwise it answers from training, which stops at a cutoff date.⁶

More in Chapter 10: Tools

Is a chatbot the same as an agent?

A chatbot answers. An agent takes steps, deciding which tools to use and in what order until a task is done.⁷

More in Chapter 9: The harness

Can it be tricked?

Yes. Text hidden in a web page or document can try to give it instructions ("prompt injection").⁸ That's one reason harnesses run work in isolated environments and ask before some actions.⁹

More in Chapter 10: Tools

Who decides how it behaves?

Layers of people: the model's maker (training and published principles), the app builder (instructions), and you (your message).¹⁰¹¹

More in Chapter 8: The prompt

Can I just run a free model on my own computer?

Yes, for some jobs. Small downloadable models suit narrow, private tasks, but they know less than the largest models.¹²¹³

More in Chapter 7: Small and free models

Where does my data go?

It depends on the arrangement: a personal account, an organization account, a regulated setup, or a model on your own machine.¹⁴¹⁵ Chapter 12 explains the parts.

More in Chapter 12: Where do my words go?

Does it understand?

People use the word differently. What's measurable: it predicts text very well, can reason through many problems,¹⁶ and can be confidently wrong.² Judge it by checking its work.

"…most of our talk between humans is not so measurable… I don't think it is all that easy to encompass that in a mathematical form."¹⁷
Claude Shannon, 1989

More in Chapter 1: What is a language model?

Notes (17)

1Anthropic, “Use Claude's chat search and memory to build on previous context,” Claude Help Center, https://support.claude.com/en/articles/11817273-use-claude-s-chat-search-and-memory-to-build-on-previous-context (accessed October 26, 2026).
2Adam Tauman Kalai, Ofir Nachum, Santosh S. Vempala, and Edwin Zhang, “Why Language Models Hallucinate” (preprint, arXiv, September 4, 2025), https://doi.org/10.48550/arXiv.2509.04664.
3Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, et al., “Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks” (preprint, arXiv, May 22, 2020), https://doi.org/10.48550/arXiv.2005.11401.
4Anthropic, “How tool use works,” Claude Platform Docs, https://platform.claude.com/docs/en/agents-and-tools/tool-use/how-tool-use-works (accessed October 26, 2026).
5Anthropic, “Enable and use web search,” Claude Help Center, https://support.claude.com/en/articles/10684626-enable-and-use-web-search (accessed October 26, 2026).
6Google, “Gemma 4 model card,” Google AI for Developers, https://ai.google.dev/gemma/docs/core/model_card_4 (accessed October 26, 2026).
7Anthropic, “Building effective agents,” December 19, 2024, https://www.anthropic.com/research/building-effective-agents (accessed October 26, 2026).
8Kai Greshake, Sahar Abdelnabi, Shailesh Mishra, Christoph Endres, Thorsten Holz, and Mario Fritz, “Not what you've signed up for: Compromising Real-World LLM-Integrated Applications with Indirect Prompt Injection,” in Proceedings of the 16th ACM Workshop on Artificial Intelligence and Security (November 30, 2023), 79–90, https://doi.org/10.1145/3605764.3623985.
9Anthropic, “Use Claude Cowork safely,” Claude Help Center, https://support.claude.com/en/articles/13364135-use-claude-cowork-safely (accessed October 26, 2026).
10Anthropic, “Claude's Constitution,” n.d., https://www.anthropic.com/constitution (accessed October 26, 2026).
11OpenAI, “OpenAI Model Spec (2025-10-27 version),” October 27, 2025, https://model-spec.openai.com/2025-10-27 (accessed October 26, 2026).
12Marah Abdin, Jyoti Aneja, Hany Awadalla, Ahmed Awadallah, Ammar Ahmad Awan, Nguyen Bach, Amit Bahree, et al., “Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone” (preprint, arXiv, April 2024 (v4 2024-08-30)), https://doi.org/10.48550/arXiv.2404.14219.
13Peter Belcak, Greg Heinrich, Shizhe Diao, Yonggan Fu, Xin Dong, Saurav Muralidharan, Yingyan Celine Lin, and Pavlo Molchanov, “Small Language Models are the Future of Agentic AI” (preprint, arXiv, June 2, 2025), https://doi.org/10.48550/arXiv.2506.02153.
14Anthropic, “Updates to our consumer terms,” August 28, 2025, https://www.anthropic.com/news/updates-to-our-consumer-terms (accessed October 26, 2026).
15Anthropic, “Is my data used for model training? (commercial),” Anthropic Privacy Center, August 18, 2026, https://privacy.claude.com/en/articles/7996868-is-my-data-used-for-training (accessed October 26, 2026).
16Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Brian Ichter, Fei Xia, Ed Chi, Quoc Le, and Denny Zhou, “Chain-of-Thought Prompting Elicits Reasoning in Large Language Models,” in Advances in Neural Information Processing Systems 35 (2022), 24824–24837, https://doi.org/10.52202/068431-1800.
17John Horgan, “Profile of Claude Shannon, Inventor of Information Theory,” Scientific American (Cross-Check blog), July 26, 2017, https://www.scientificamerican.com/blog/cross-check/profile-of-claude-shannon-inventor-of-information-theory/ (accessed October 26, 2026).

Chapter 13 · page 1 of 1

Turn pages with the ← and → keys.