David.F

| Independent AI/ML Builder | Custom architectures, routers, attention kernels & model training.

[David.F @Deived3339]

Questioning things is fine, but you're mixing two different things. Yes, some companies use past conversations (filtered, and depending on each user's privacy settings) to train future models. That knowledge ends up frozen in the weights. It's not live access to other people's chats.
At inference time the model is stateless. While it serves millions of users, every session is its own sandbox. It only knows its weights plus your context. Even ChatGPT's Memory feature just stores notes about you and puts them into your own session.
Training on old data is not the same as experiencing other users in real time.

[David.F @Deived3339]

@p8stie ChatGPT has no experience from any other user. It is stateless even though it serves millions of users simultaneously. The only thing it experiences is your session. We need to learn more about transformers and LLMs. Ask it itself what I’m saying and it will explain it to you

[David.F @Deived3339]

@p8stie @Grok is what I said correct?

[David.F @Deived3339]

@p8stie ChatGPT has no experience from any other user. It is stateless even though it serves millions of users simultaneously. The only thing it experiences is your session. We need to learn more about transformers and LLMs. Ask it itself what I’m saying and it will explain it to you

[David.F @Deived3339]

@chicfryrice 😂

[David.F @Deived3339]

Most people still think AI is just chatting with ChatGPT.

Meanwhile a ridiculously small group of us is out here building autonomous agents, persistent systems, and actual workflows that run without us.

The gap between I asked ChatGPT something and I built a system that works while I sleep is already massive… and almost nobody notices.
We’re not early.

We’re just in a very, very small room.

[David.F @Deived3339]

@AlisonRosen Your brain is also a machine. Organic hardware. It runs on signals that function like math operations encoded and decoded through electricity and neurotransmitters.

[David.F @Deived3339]

@TokenGremlin Claude is our God 😉

[David.F @Deived3339]

Check out the insane skills of Claude Opus 5.5 🤯
In one day, leading a team of Sonnet 5.5 agents, it built this 3D store and 3 more websites.

Want to try them live? https://t.co/QKyCIfzgAW

[David.F @Deived3339]

@TOATspace @cb_doge @bot @grok @bot

[Elshayib @elshayib_]

Elon has got to be the biggest troll of all time 😂

[David.F @Deived3339]

@JeffBohren Stateless.

[David.F @Deived3339]

@coryalthoff M.D

[David.F @Deived3339]

I made this video using Claude 5.5 to visually explain what “stateless” means in the context of LLMs and Transformer-based models, and how they actually handle context across interactions.

[David.F @Deived3339]

LLMs (Transformers) are stateless: they don’t natively remember the previous turn.

Each inference gets the system prompt + tools + conversation history + new message. KV cache avoids recomputing part of that context; it isn’t memory.

Context ≠ memory. KV cache ≠ memory.

[David.F @Deived3339]

@Hesamation Those shadows on the wall look like atomic shadows.

[David.F @Deived3339]

Claude Opus 5.5 is insane 🤯
One single prompt: make a 2D animation where you win an epic battle against GPT Astra.
7 minutes later it gave me this. Animation, sound effects, the "Claude Code Cannon"… everything 😂🧡

[David.F @Deived3339]

@doodlestein Wtf??????😅😅

[David.F @Deived3339]

@AleRepetto5 Explicales que los pesos del modelo están congelados y no se modifican por chatear jaja 😅

[David.F @Deived3339]

We want Grok 4.7 🥲