--- title: Lessons from Studying Every Memory System — Shlok... description: A talk breaking down how memory systems evolved across ChatGPT, Claude, and Gemini over three years, drawing on reverse-engineering of each product. ChatGPT... canonical: https://daily.dev/posts/lessons-from-studying-every-memory-system-shlok-khemani-independent-it6cyupj8 twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Lessons from Studying Every Memory System — Shlok Khemani, Independent | daily.dev og:description: A talk breaking down how memory systems evolved across ChatGPT, Claude, and Gemini over three years, drawing on reverse-engineering of each product. ChatGPT... og:url: https://daily.dev/posts/lessons-from-studying-every-memory-system-shlok-khemani-independent-it6cyupj8 og:image: https://api.daily.dev/og/posts/It6CYuPJ8.png og:image:alt: Lessons from Studying Every Memory System — Shlok Khemani, Independent og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Lessons from Studying Every Memory System — Shlok Khemani, Independent **[AI Engineer](https://daily.dev/sources/aidotengineer)** · 19 min read · 0 upvotes · 0 comments ## Summary A talk breaking down how memory systems evolved across ChatGPT, Claude, and Gemini over three years, drawing on reverse-engineering of each product. ChatGPT went from a user-editable fact list (V1, Feb 2024) to an opaque, dense 4,000-token 'running profile' updated every few days (V2, April 2025), later adding conversation search and a visible LLM-generated summary plus explicit edits (mid-2025), deprecating the old fact list. Claude launched memory in August 2025 with on-demand conversation/time-based search tools rather than a profile, then released V2 the same day as the speaker's viral blog post, adding a visible, editable, ~1,000-token profile refreshed every 24 hours. The talk argues memory cannot be outsourced, is fundamentally a function of available compute (trade-offs between update frequency and serving cost), represents a form of continual learning happening outside model weights, and remains bottlenecked by products' inability to reason over rich existing context like email, calendar, and cross-app history. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### How does Claude's memory system work compared to ChatGPT's memory? Claude's first memory version, released August 2025, gave the model tools to search past conversations by keyword or time period rather than maintaining a profile. On the same day a viral blog post noted this contrast, Claude shipped V2, adding a visible, user-editable running profile of about 1,000 tokens that refreshes every 24 hours, using complete sentences rather than dense keywords, unlike ChatGPT's approach. _Track how AI memory architectures diverge and converge by following developer analysis on daily.dev._ ### What is a running profile in ChatGPT's memory system and how big is it? A running profile is ChatGPT's V2 memory feature, released April 2025, where the model periodically reviews conversations and updates a persistent summary of the user, roughly 4,000 tokens across 16 sections, added to context in every new chat. It updates every few days, is dense with keyword-like clues, and was not user-visible until a June update exposed an LLM-generated summary and allowed explicit edits. _Developers designing AI personalization features can follow memory architecture breakdowns on daily.dev._ ### Why did ChatGPT deprecate its original memory fact list feature? ChatGPT deprecated its V1 fact list, where users manually told it things to remember, after a June update made its running profile summary visible and editable, replacing the older mechanism entirely. The fact list suffered from putting memory management burden on users and from storing stale facts, such as travel plans, that no longer applied but stayed in context indefinitely. _Keep pace with how consumer AI products retire old features by following updates like this on daily.dev._ ## Similar posts on daily.dev - [ChatGPT Memory Explained: How AI Finally Started Remembering What Matters.](https://daily.dev/posts/chatgpt-memory-explained-how-ai-finally-started-remembering-what-matters--2bdpez7no) · Medium · 1 upvotes · 0 comments - [ChatGPT now quietly rewrites its memories of you, and that's both good and unsettling](https://daily.dev/posts/chatgpt-now-quietly-rewrites-its-memories-of-you-and-that-s-both-good-and-unsettling-xi6uck4gj) · XDA Developers · 1 upvotes · 0 comments - [ChatGPT finally got the memory transparency feature it needed — it still isn't enough to beat Claude](https://daily.dev/posts/chatgpt-finally-got-the-memory-transparency-feature-it-needed-it-still-isn-t-enough-to-beat-claude-cf3szttku) · XDA Developers · 0 upvotes · 0 comments - [MY TAKE: Transparent vs. opaque — edit Claude’s personalized memory, or trust ChatGPT’s blindly?](https://daily.dev/posts/my-take-transparent-vs-opaque-edit-claude-s-personalized-memory-or-trust-chatgpt-s-blindly--0qrybpznt) · Security Boulevard · 0 upvotes · 0 comments --- Tags: [#llm](https://daily.dev/tags/llm), [#chatgpt](https://daily.dev/tags/chatgpt), [#claude](https://daily.dev/tags/claude) [View this post on daily.dev](https://daily.dev/posts/lessons-from-studying-every-memory-system-shlok-khemani-independent-it6cyupj8) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Lessons from Studying Every Memory System — Shlok Khemani, Independent","url":"https://daily.dev/posts/lessons-from-studying-every-memory-system-shlok-khemani-independent-it6cyupj8","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/lessons-from-studying-every-memory-system-shlok-khemani-independent-it6cyupj8"},"datePublished":"2026-08-12T18:55:08.515Z","dateModified":"2026-08-12T18:55:37.956Z","description":"A talk breaking down how memory systems evolved across ChatGPT, Claude, and Gemini over three years, drawing on reverse-engineering of each product. ChatGPT...","image":"https://i.ytimg.com/vi/5ZGyKWjQDr0/sddefault.jpg","thumbnailUrl":"https://i.ytimg.com/vi/5ZGyKWjQDr0/sddefault.jpg","isAccessibleForFree":true,"articleSection":"AI Engineer","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"AI Engineer","logo":"https://media.daily.dev/image/upload/s--u5PucxNT--/f_auto/v1724338940/logos/aidotengineer","url":"https://daily.dev/sources/aidotengineer"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/lessons-from-studying-every-memory-system-shlok-khemani-independent-it6cyupj8","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"llm,chatgpt,claude","timeRequired":"PT19M","video":{"@type":"VideoObject","name":"Lessons from Studying Every Memory System — Shlok Khemani, Independent","description":"A talk breaking down how memory systems evolved across ChatGPT, Claude, and Gemini over three years, drawing on reverse-engineering of each product. ChatGPT...","thumbnailUrl":"https://i.ytimg.com/vi/5ZGyKWjQDr0/sddefault.jpg","uploadDate":"2026-08-12T18:55:08.515Z","duration":"PT19M","url":"https://api.daily.dev/r/It6CYuPJ8","embedUrl":"https://www.youtube.com/embed/5ZGyKWjQDr0"}} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"AI Engineer","item":"https://daily.dev/sources/aidotengineer"},{"@type":"ListItem","position":3,"name":"Lessons from Studying Every Memory System — Shlok Khemani, Independent"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/lessons-from-studying-every-memory-system-shlok-khemani-independent-it6cyupj8#faq","mainEntity":[{"@type":"Question","name":"How does Claude's memory system work compared to ChatGPT's memory?","acceptedAnswer":{"@type":"Answer","text":"Claude's first memory version, released August 2025, gave the model tools to search past conversations by keyword or time period rather than maintaining a profile. On the same day a viral blog post noted this contrast, Claude shipped V2, adding a visible, user-editable running profile of about 1,000 tokens that refreshes every 24 hours, using complete sentences rather than dense keywords, unlike ChatGPT's approach. Track how AI memory architectures diverge and converge by following developer analysis on daily.dev."}},{"@type":"Question","name":"What is a running profile in ChatGPT's memory system and how big is it?","acceptedAnswer":{"@type":"Answer","text":"A running profile is ChatGPT's V2 memory feature, released April 2025, where the model periodically reviews conversations and updates a persistent summary of the user, roughly 4,000 tokens across 16 sections, added to context in every new chat. It updates every few days, is dense with keyword-like clues, and was not user-visible until a June update exposed an LLM-generated summary and allowed explicit edits. Developers designing AI personalization features can follow memory architecture breakdowns on daily.dev."}},{"@type":"Question","name":"Why did ChatGPT deprecate its original memory fact list feature?","acceptedAnswer":{"@type":"Answer","text":"ChatGPT deprecated its V1 fact list, where users manually told it things to remember, after a June update made its running profile summary visible and editable, replacing the older mechanism entirely. The fact list suffered from putting memory management burden on users and from storing stale facts, such as travel plans, that no longer applied but stayed in context indefinitely. Keep pace with how consumer AI products retire old features by following updates like this on daily.dev."}}]} ```