Four essential characteristics of human intelligence that current AI systems lack: reasoning, planning, persistent memory, and understanding the physical world.
There is no question that AI will eventually reach and surpass human intelligence in all domains.
But it won't happen next year.
And it won't happen with the kind of Auto-Regressive LLMs currently in fashion (although they may constitute a component of it).
https://t.co/ohg9y6qV37
@bretthuneycutt That’s awesome! Here’s a few stuff I look fwd to:
1. Physical presence for instance bank draft issuance (similar to cash deposite at canada post).
2. Insurance verticals: auto, home, and life insurance.
3. Better mortgage assistance.
4. Ways to invest in Wealthsimple :)
But the data-layer critique seems harder to dismiss than the others. And I haven’t seen it taken seriously in production.
Curious wether this lands folks — convinced, unconvinced, or pointing at a different bottleneck entirely?
(12/12)
Social media feed in general has gotten remarkably good at predicting what I’ll click, and oddly bad at predicting what I’ll value.
Am I biased, or has social media marketing given up on adding value and settled for just being loud? 🧵
(1/12)
Maybe explicit user signal is too noisy and behavior is the least biased input we have.
Maybe richer context inputs introduce more surveillance than the personalization gain is worth.
(11/12)
New work with @egrefen at @GoogleDeepMind:
🚨Interaction Dynamics as a Reward Signal for LLMs🚨
When it comes to interactions, the "how" is just as important as the "what"
There is a signal in how we interact with a model that text analysis misses: hesitation, drift, friction
Introducing Collaborative Reasoner: a framework to improve collaborative reasoning in language models.
Collaborative Reasoner paves the way for developing social agents that can partner with humans and other agents.
Read the research paper and download the code. https://t.co/GgJEPXTH8W
Rethinking Memory in AI
Great overview of memory in AI agents with a more structured and dynamic perspective on research and ideas.
I think looking at memory and its atomic operations (indexing, retrieval, compression,...) can lead to better memory solutions for AI agents.
the best researchers from Meta, Yale, Stanford, Google DeepMind, and Microsoft laid out all we know about Agents in a 264-page paper [book],
here are some of their key findings: