3 ms·
Indeed. Any meaningful AGI/ASI will have to have a form of memory / continual learning. Sam Altman said last year that this will be the focus for GPT-6. The wh
by bananaflag 4mo ago
Indeed. Any meaningful AGI/ASI will have to have a form of memory / continual learning. Sam Altman said last year that this will be the focus for GPT-6.
The whole "soul.md" stuff today is a poor approximation to that. But I wonder whether it will grow into it, like chain of thought prompting grew into reasoning models.
- Iolaum 4mo agoLLM's as a technology - currently - are stateless. The memory layer is controlled by the agent. I m surprised with the gpt-6 reference unless it has to do with vertical integration between the agent and the layer.
- bananaflag 4mo agoYeah, I am aware of this statelessness. This is what I was referring to: https://www.cnbc.com/2025/08/19/sam-altman-on-gpt-6-people-want-memory.html https://www.cnbc.com/2025/08/19/sam-altman-on-gpt-6-people-w... It's marketing speak, but the goal is clearly there, no idea how achievable.
- fragmede 4mo agoHow it works under the hood is separate from selling the feature.