Do Large Language Models Dream of AI Agents?

While sleeping, The human brain selected through different memories, united important things when leaving important things. If AI could do the same?

BuiltA company that provides local shopping and restaurant deals for tenants, has recently deployed millions of agents with hopes of doing so.

Using technology from a startup called built Read It allows agents to learn from previous conversations and share memories with each other. Using a process called “Sleeptime Count”, agents decide what information is to be stored in its long -term memory vault and what may be needed for a speedy recovery.

“We can make a single update on one [memory] Bill’s AI Engineer Andrew Fitz said, “Block the behavior of several thousand agents and change their behavior.

Big -language models can only “remember” if the data is usually included in the data Context window. IF you want a chatboat to remember your recent conversation, you need to be stuck in the chat.

Most AI systems can only handle information in the context window before using data phalters, and they are hallucinates or confused. In contrast, the human brain is able to file useful information and remember it later.

“Your brain is constantly improving, adding more information like a sponge,” the Letter CEO Charles Packer says. “With the models of the language, it is just like the opposite you

Packer and his coffowner Sara Uders had developed before MemgopAn Open Source Project that LLM to help decide what information should be stored in short -term vs long -term memory. With the letter, this pair has expanded the agents to learn in the background.

B builder’s cooperation is part of a broad push to preserve and recover AI useful information, which can make the chatboat more smart and less defects for agents. Memory is underdeveloped in the modern AI, which undermines the intelligence and reliability of AI tools, according to the experts I have talked to.

Langchen’s Cafounder and CEO Harrison Chase, who created a method to improve the memory of AI agents, says he sees memories as an important part of context engineering – where a user or engineer decides what information will be given in the context window. Companies provide different types of memory storage for Langchain agents, from long -term information about users to memories of recent experiences. Chase says, “Memories, I will argue, a form of context.” “A large part of an AI engineer’s work is basically getting the model in the correct context [information]The “

Customer AI equipment is slowly becoming less forgotten. This February, Open Declaration This ChatzPT will preserve relevant information to provide more personalized experiences for users – though it does not reveal how it works.

Leata and Langchen create the process of more transparent recreation for engineers to create AI systems.

“I think it’s important to open the models, but also to open the memory systems,” said Kleme Delangu, a CEO of the AI ​​hosting platform and an investor in Lata.

Anxietyly, the Letter CEO Packer indicates that it can also be important to learn what to forget for AI models. “If a user said, ‘The project we were working with delete it from your memory’ should be able to get back to the agent and to re -write each single memory in return.”

The idea of ​​artificial memories and dreams force me to think Androids dream of electric sheep? Philip K. Dick wrote, a man-borne novel that inspired the stylish Dystopian movie Blade runnerThe Big -language models are still not as impressive as it Rebel transcript The story, however, seems to have their memories, Can be just as fragile asThe

---

This is a version Will Nights AI Lab NewsletterThe Read the previous newsletters Here