Trending topics
#
Bonk Eco continues to show strength amid $USELESS rally
#
Pump.fun to raise $1B token sale, traders speculating on airdrop
#
Boop.Fun leading the way with a new launchpad on Solana.

Aakash Gupta
I'm unbelievably excited to launch Aakash's bundle.
I'm making $28,336 of AI tools free to every subscriber.
It's a complete AI PM toolstack:
AI Evals: @arizeai
AI Agents: @relay
AI Prototyping: @reforge
AI Research: @DeepSkyAI
Task Management: @linear
Design Inspiration: @mobbin
AI Video Editing: @DescriptApp
AI Customer Intelligence: @hidovetail
Front-end AI Prototypes: @magicpatterns
Most of these have never done a deal like this, including Arize, Relay, Reforge Build, Dovetail, & DeepSky.
You get it all for just an annual subscription: $150.
(Cheapest in the market)
Here's how to take advantage:
1. Get an annual membership:
2. Or go for a founding plan:
3. Then grab your codes:
📌 Want to get my guide to these 9 tools? Comment 'Aakash bundle guide' + DM me and I'll reply.
I'm adding new brands every month, so this will just keep getting better and better.
Codes are limited so grab them quick.

885
Everyone assumes ChatGPT’s memory is some sophisticated RAG system with vector databases and semantic search.
Manthan reverse engineered it. The actual architecture is almost disappointingly simple: session metadata that expires, explicit facts stored as text, lightweight chat summaries, and a sliding window.
No embeddings. No similarity search. No retrieval at scale.
The interesting part? This explains why it feels so fast. Traditional RAG systems embed every message, run similarity searches on each query, pull full contexts. ChatGPT just injects pre-computed summaries directly. They’re trading detailed historical context for latency.
This is the same lesson that keeps emerging across AI infrastructure: when you control the entire stack, curated simplicity often outperforms sophisticated complexity. OpenAI doesn’t need to build a general retrieval system. They just need one that works for ChatGPT.
The four-layer architecture (session metadata → stored facts → conversation summaries → sliding window) is basically a handcrafted memory hierarchy. Each layer has different persistence and different purposes. Session metadata adapts in real-time. Facts persist forever. Summaries provide continuity. The window maintains coherence.
Anthropic’s memory system uses a similar pattern. The models that feel most personal aren’t the ones with the most sophisticated retrieval. They’re the ones that store the right things and inject them at the right time.

Manthan GuptaDec 10, 12:47
I spent the last few days prompting ChatGPT to understand how its memory system actually works.
Spoiler alert: There is no RAG used

4.53K
Top
Ranking
Favorites

