Reading Books in the LLM Era | Generated by AI
Question: In the LLM era, many people read chatbot answers. Why do we still need to read books? How should we read books now?
Answer:
Books remain valuable — and in many ways more important — in the LLM era, even though chatbots provide fast, convenient answers.
Why read books when LLMs exist?
Here are the main reasons books still matter (and often outperform pure chatbot usage):
-
Deep, coherent, long-chain reasoning and worldview building
A good book usually presents one coherent argument, narrative, or system of thought over 200–500 pages. LLMs tend to give fragmented, context-window-limited, sometimes internally inconsistent answers. Reading the full book lets you internalize a complete mental model. -
Original thinking, nuance, and personality
Books contain the author’s unique voice, subtle reasoning, anecdotes, metaphors, and hard-won insights that LLMs usually average out or dilute. LLM outputs often feel “sanitized” or generic. -
Long-term retention and schema formation
Actively reading (especially with notes, reflection, re-reading hard parts) creates much stronger memory traces and conceptual connections in your brain than skimming chatbot summaries. Neuroscience still shows deep reading activates broader neural networks over days/weeks. -
Trust and source grounding
With a book you know exactly whose knowledge, biases, and evidence you’re absorbing. LLM answers frequently mix high-quality and low-quality sources without clear attribution. -
Emotional, aesthetic, and moral depth
Fiction, philosophy, history, memoirs → these train empathy, taste, ethical intuition, and pattern recognition in ways that pure information retrieval does not. Chatbots are weak at transmitting “felt experience.” -
Escape from optimization & dopamine loops
Book reading is one of the few remaining high-value activities that resists being gamified or turned into short-form content.
In short:
Chatbots → excellent for lookup, quick explanations, surface-level synthesis
Books → superior for deep understanding, original frameworks, character, taste, long-term intellectual growth
How to read books effectively in 2025–2026 (LLM era)
Most people now use a multi-speed, tool-assisted reading strategy rather than the old “read cover-to-cover linearly” approach.
Common effective patterns in 2025:
- Tiered / Just-in-time reading (most popular hybrid style)
- Awareness reading: use LLM to summarize book + reviews + key ideas (5–15 min)
- If valuable → retrieval reading: ask LLM targeted questions (“explain chapter 5 argument”, “criticisms of author’s position X”)
- If still compelling → deep reading: read selected chapters slowly, offline, with pencil/paper or margin notes
- Goal: spend deep time only on 20–40% of books that truly matter to you
- LLM as active reading companion (Andrej Karpathy-style)
- Read chapter offline → close book
- Talk to LLM: “explain this chapter like I’m 15”, “what are the strongest/weakest arguments?”, “give me 5 critical questions I should be asking”, “connect this to [other book/domain]”
- Re-read confusing parts with higher comprehension
- Technical / dense non-fiction workflow
- First pass: fast read + highlight (20–40 min/chapter, offline)
- Second pass: type/run code, reproduce figures, explain concepts to LLM or rubber duck
- Third pass (optional): write your own 1-page summary or teach it back to an LLM
- Pure deep reading (still powerful for fiction/philosophy)
- No devices, no LLM during the session
- Read slowly → journal reactions → let ideas simmer days/weeks
- Use LLM afterward for context, comparisons, or “what has aged poorly?”
Quick decision framework:
| Goal | Best approach in LLM era | Time investment |
|---|---|---|
| Quick knowledge / decide to read? | LLM summary + reviews + table of contents | 5–20 min |
| Learn practical skill fast | LLM + selected chapters + code execution | 2–10 hours |
| Build deep worldview / taste | Full slow read, minimal LLM interference | 10–50 hours |
| Stay current broadly | LLM-powered “awareness” reading of many books | 30–90 min/book |
Bottom line in 2026:
Use LLMs to filter and accelerate → but protect time for real deep reading of the 5–15 books/year that can actually change how you think.
References:
- Why Read Books When You Can Use Chatbots to Talk to Them Instead?
- How Andrej Karpathy Reads Books With LLMs
- Recommendations for Getting the Most Out of a Technical Book
- The Complete AI Reading Guide for 2025