Skip to content
Brick Moving AI

Blog

ChatGPT Memory and the Bitter Lesson

Shlok Khemani
Overview
Section titled “Overview”

The author of this article reverse-engineered ChatGPT's memory system by directly questioning it, revealing its operational principles and internal structure.

The Four Key Components of ChatGPT's Memory System
Section titled “The Four Key Components of ChatGPT's Memory System”

ChatGPT's memory system primarily consists of four components, all provided to the model during each interaction:

  1. Interaction Metadata:

    • Includes user device information (screen size, browser/OS), usage patterns (topic preferences, message length, activity levels), etc.
    • The model can leverage this data to implicitly infer the user's context (e.g., automatically recognizing iPhone usage), thereby delivering more targeted responses.

  2. Recent Conversation Content:

    • Contains summaries of the user's messages from the last several dozen conversations (excluding AI responses).
    • This helps establish connections across different conversations, allowing the model to better understand context. For instance, after multiple consecutive conversations about travel to Japan, it can infer that "there" refers to Japan.

  3. Model Set Context:

    • Facts explicitly provided by the user, which can be viewed and deleted anytime in the settings—e.g., "I am allergic to shellfish."
    • This is the highest-priority, fully user-controlled "source of truth" that can override information from other memory modules.

  4. User Knowledge Memories:

    • This is the newest and most core component. It consists of highly condensed AI-generated summaries periodically created by OpenAI from the user's extensive conversation history.
    • These memories are invisible and not directly editable by the user. They contain extremely detailed information about the user's profession, interests, projects, technical stack, brand preferences, etc.
    • While incredibly information-dense, they may include outdated or inaccurate content (e.g., a trip the user planned but never took).
Core Mechanism: "The Bitter Lesson"
Section titled “Core Mechanism: "The Bitter Lesson"”

The article points out that ChatGPT's memory system does not use complex techniques like Retrieval-Augmented Generation (RAG) or vector databases to filter relevant memories.

Instead, it adopts a "brute force" yet effective approach: during each interaction, it packs all four types of memory information into the model's context window.

This reflects OpenAI's core bets:

  1. The model is sufficiently intelligent: Powerful models can inherently discern and utilize relevant information within massive contexts while ignoring the irrelevant.
  2. Compute and context windows will become increasingly cheaper: As technology advances, the cost of sending all this information will become negligible.

This reaffirms the lesson articulated by reinforcement learning pioneer Rich Sutton in his 2019 essay "The Bitter Lesson"—rather than building complex engineered solutions, it's more effective to dedicate resources to enhancing the model's inherent capabilities and computational power.

System Analogy
Section titled “System Analogy”

ChatGPT's memory functionality resembles the training process of an LLM: "User Knowledge Memories" act like a large but slow-to-update base model, while the other three components function as steering layers for real-time adjustment and correction (similar to RLHF and in-context learning).

  1. User Knowledge Memories: Act like a pre-trained model, condensing long-term information but prone to becoming outdated.
  2. Model Set Context: Equivalent to the user's RLHF, holding the highest priority.
  3. Recent Conversation Content: Analogous to immediate in-context learning.
  4. Interaction Metadata: Functions like system default parameters, providing environmental signals.
Future Challenges
Section titled “Future Challenges”

Future challenges lie not only in technology (e.g., updating "User Knowledge Memories" more frequently) but also at the product level: how to handle outdated information, how to validate facts, and the privacy and ethical concerns arising from AI building detailed profiles of users.

The Socratic Journal Method

Ed Nite
Overview
Section titled “Overview”

The Socratic Journal Method emphasizes consistent practice over perfection. Through questioning, answering, and tracking, it helps you pause, reflect, and face yourself with honesty.

What - What it is
Section titled “What - What it is”

The Socratic Journal Method is a journaling technique conducted in a self-questioning and self-answering format. Its core inspiration comes from the Socratic philosophical principle that "The unexamined life is not worth living." It transforms journaling from a one-way monologue or log into a dialogic interview with oneself. This method uses carefully designed questions to guide thinking, aiming to achieve self-exploration, emotional release, and thought organization through questioning and free writing.

Why - Why it works
Section titled “Why - Why it works”

The effectiveness of this method is based on psychological evidence and habit science, successfully addressing the common problem of difficulty in maintaining a traditional journaling habit:

  1. Scientific Backing: Research by psychologist James Pennebaker shows that expressive writing can reduce stress, improve mood, and even boost immunity. Carol Dweck's theory of metacognition (thinking about your thinking) is also highly aligned with this method.
  2. Lowering the Barrier to Entry: The "tiny habits" principle by behavior scientist BJ Fogg is applied here—completing one journal entry simply requires answering one question, eliminating the need for long essays and making the habit easier to maintain.
  3. Solving Traditional Pain Points: It perfectly solves common dilemmas of traditional journaling like "not knowing what to write," "feeling like a chore," and "hard to sustain," turning the stressor (the blank page) into a guiding tool (the question).
  4. Self-Cognitive Therapy: Much like the Socratic questioning used by therapists, this method helps individuals challenge and reframe irrational beliefs, uncovering thinking patterns.
How - How to practice
Section titled “How - How to practice”

  1. Core Two-Stage Rhythm:

    • Think Deeply Upfront: Design core questions that reflect your genuine concerns (e.g., "What felt light today? What felt heavy?").
    • Write Freely Afterwards: When answering, discard editing and judgment, let thoughts flow naturally, and focus on honesty rather than length.

  2. Tool Selection (Choose as needed):

    • Paper & Pen: Ideal for deep reflection, distraction-free, but less searchable.
    • Digital Apps (e.g., Obsidian, Notion, simple text files): Efficient, searchable, easy to organize.
    • Audio/Video Recording: Suitable for commutes or busy hands, allows for fast dictation.
    • Key Point: The tool isn't important; consistent use is. A hybrid approach is acceptable.

  3. 5-Minute Starter Guide:

    • Ask One Honest Question (e.g., What's one thing occupying my mind right now?).
    • Answer with Raw Honesty: Don't edit, don't judge, even one sentence is perfect.
    • Note One Thing You're Tracking: e.g., sleep, mood. Just log it; no commentary needed.
    • Keep the Tone Light and Curious: This is a conversation with your future self, not a performance review. Replace "criticism" with "curiosity" (e.g., instead of "Why did I fail?" try "What was the obstacle today?").

  4. Important Reminder:

    • Questions are dynamic and should evolve as your life focus changes.
    • Maintain a conversational mindset, not an interrogative one. If you feel dread, obsess over metrics, or use the journal to punish yourself, it's a sign to adjust your questioning mindset and angle.

The Boring Future of GenAI

The monetization of Generative AI services will likely follow a path similar to that of early search engines: through advertising and sponsored content. The author argues that while people currently use large language models (LLMs) for tasks like creating workout plans or generating recipes, the lack of a breakthrough leading to Artificial General Intelligence (AGI) means that the primary way to make money from these services will be by subtly integrating ads and promoted products into the generated responses.

Why language models hallucinate

OpenAI
Overview
Section titled “Overview”

This article explores the root causes and solutions for “hallucinations” in large language models, which is when they confidently generate false information.

Core Argument
Section titled “Core Argument”

The main reason language models hallucinate isn’t a technical flaw; it’s a problem with their training and evaluation methods. The current standard evaluation system is like “teaching to the test,” rewarding only “accuracy.” This incentivizes models to guess when they’re uncertain instead of admitting, “I don’t know.”

Main Causes Analysis
Section titled “Main Causes Analysis”
  1. Flawed Incentive Mechanism: In evaluations, a model gets zero points for answering “I don’t know,” but a guess has a chance of being correct. To get a higher score on leaderboards, models are trained to be more inclined to guess. While this might increase accuracy, it also significantly raises the risk of hallucinations (incorrect answers).
  2. The Nature of Pre-training: During pre-training, models learn language patterns by predicting the next word. For structured knowledge like grammar and spelling, which have clear patterns, models learn well. But for scattered, low-frequency facts (like someone’s birthday), there’s no fixed pattern, so the model can only make a probabilistic guess. This is the initial source of hallucinations.
Solutions
Section titled “Solutions”

The core solution proposed by the article is to reform the evaluation system:

  • Change Grading Rules: Don’t just focus on accuracy. Instead, severely penalize “confident incorrect answers” while giving partial credit to models that admit uncertainty (e.g., by answering “I don’t know”).
  • Comprehensive Update of Evaluation Standards: This new grading method needs to be applied to all major, core evaluation benchmarks, not just a few specialized “hallucination evaluations.” Only then can a fundamental change in the model’s “behavioral patterns” be achieved.
Conclusion and Clarified Misconceptions
Section titled “Conclusion and Clarified Misconceptions”
  • Hallucinations are not inevitable; models can learn to “be humble.”
  • Solving the hallucination problem doesn’t necessarily require a larger model; sometimes, smaller models are better at knowing the limits of their knowledge.
  • Simply pursuing 100% accuracy can’t eliminate hallucinations, because many real-world problems are inherently unanswerable.
  • The key to solving the problem is to reform all core evaluation metrics so they no longer reward guessing.