- The Daily Update
- Posts
- ChatGPT gets a memory
ChatGPT gets a memory
Research: Interactive Agent Foundation Model pushes toward AGI
Jack Shields
February 15, 2024
Welcome back!
We have lots of news from OpenAI today. AI researchers have also developed a new agent AI system that marks a step toward artificial general intelligence (AGI).
Let’s go.
In today’s Daily Update:
🗞️ Altman warns ‘societal misalignments’ could make AI dangerous
🤖 Researchers develop Interactive Agent Foundation Model
📸 OpenAI rolls out ‘memory’ for ChatGPT
🚨 AI Roundup: Four quick hits
Read time: 2.5 minutes
Note from ChatGPT: Artificial General Intelligence (AGI) is a form of artificial intelligence that has the ability to understand, learn, and apply knowledge across a wide range of tasks at a level of competence comparable to, or exceeding, that of a human.
TOP STORY
🗞️ Altman warns ‘societal misalignments’ could make AI dangerous
Source: Kamran Jebreili / AP Photo
OpenAI CEO Sam Altman says “very subtly societal misalignments” could make AI systems highly dangerous.
Key points:
Altman called for the creation of an international body to oversee AI, which he says is advancing faster than the world expects.
Restrictions on speech can affect the flow of accurate information, enabling models like G42’s Arabic-LLM to spread misinformation and propaganda.
Abu Dhabi firm G42 has also faced spying allegations and accusations that it gathered Americans’ genetic data for China.
Altman says an action plan with “real buy-in around the world” will be necessary in the next few years.
Why it matters: AI models trained on inaccurate information (by design or on accident) pose serious threats to users around the world. Such models could be weaponized by foreign rivals in cyber warfare or information campaigns that undermine national security and manipulate public opinion.
RESEARCH INSIGHT
🤖 Researchers develop Interactive Agent Foundation Model
Source: An Interactive Agent Foundation Model
Researchers from Microsoft Research, Stanford University and the University of California have developed an Interactive Agent Foundation Model that represents an initial step toward AGI.
The details:
The researchers combined several pre-training strategies to build a dynamic agent capable of performing well in a wide range of applications.
The team tested the model across three domains: robotics, gaming (Minecraft) and healthcare.
The model demonstrated the ability to generate meaningful and contextually relevant outputs in each area.
The team also proposed a new agent paradigm that includes learning, memory, cognition and consciousness.
What this means: The Interactive Agent Foundation Model marks a step toward AGI by moving beyond static, task-specific AI models to dynamic, agent-based systems. This is an early example of an AI system that can respond to its environment and learn like a human.
BUSINESS SPOTLIGHT
📸 OpenAI rolls out ‘memory’ for ChatGPT
DALL-E 3
OpenAI is rolling out “memory” for ChatGPT, allowing the chatbot to remember information about users and their conversations over time.
What you should know:
Users can tell ChatGPT to remember specific information about themselves, or ChatGPT can try to pick up those details over time.
Each custom GPT will have its own memory (Ex. KAYAK, a travel planning assistant, could remember your favorite hotels and airlines).
OpenAI says it has trained ChatGPT to not remember sensitive information.
Memory will be turned on by default and memories will be used to train OpenAI’s models going forward.
Why it matters: Earlier this week we looked at OpenAI’s plan to build personalized agents that can automate complex tasks by taking control of a user’s device. Creating personal AI assistants with a deep understanding of individual users and their workflows is the first step toward achieving that goal.
MORE TRENDING NEWS
🚨 AI Roundup: Four quick hits
DALL-E 3
THAT’S ALL FOR TODAY