1) learning from experience (although you already mentioned it in the dynamic memory part) 2) physics grounding: llm-agent that can observe and act in an environment, reason spatially, plan, etc 3) integration with robotics
AI
-
New SF Office Opening Signals Expansion and Innovation
By
–
We’re so excited to have everyone join us today in our new SF office. We’re even more excited to see what they build!
-
GPT-4 Regulation Agreement: Definition and Implementation Questions
By
–
I wonder how this will work out… So many questions. – How many people who are actually working on systems that exceed GPT-4 will sign this? – What is a system that exceeds GPT-4 ? How about React / Self-reflective agent build on LLAMA? – Why 6 months? What is it going to
-
Google CoLT5 Processes Long Inputs via Conditional Computation
By
–
Google’s CoLT5 Processes Extremely Long Inputs via Conditional Computation https://
syncedreview.com/2023/03/28/goo
gles-colt5-processes-extremely-long-inputs-via-conditional-computation/
… -
GPT-4 Strengthens Cybersecurity Defense with Formal Verification
By
–
GPT-4 for security, giving a real power boost for defenders. One encouraging (but still tentative) idea that I've heard over the years is that AI may permanently shift the favor to defenders; for example, by enabling formally verified systems.
-
AI Hack Night Demo Event Announcement
By
–
We are ready for the AI Hack Night. Who is coming through to demo?
-
ChatGPT and GPT-4: Parameter Efficiency in Knowledge Distillation
By
–
We don't appreciate enough that ChatGPT & GPT-4, using just a few hundred GB of parameters, can distill immense human knowledge, and display exceptional reasoning & generalization skills. And you? How many GB of parameters are in your brain?
-
LLMs Integrated into Medical Care Plan UI
By
–
🩺 Now that LLMs can pass Medical Licensing Exams, they have been built into a great UI for generating medical care plans.
— AI Breakfast (@AiBreakfast) 28 mars 2023
Try Glass Health AI 2.0: https://t.co/RbcaE26WNZpic.twitter.com/T2yEUZSnMsNow that LLMs can pass Medical Licensing Exams, they have been built into a great UI for generating medical care plans. Try Glass Health AI 2.0: http://
glass.health/ai -
Essential AI Tools: Langchain, Llama Index, Dust for Development
By
–
Thanks! I think it depends on your skill level/use cases but I’d look at Langchain, Llama Index, http://
Dust.tt on the infra side. Also depending on domain, might be worth looking at tools being built (Eg I use http://
elicit.org for science research.) -
Section 4.1 approach and references on arxiv paper
By
–
See section 4.1 for an approach and some references to others: https://
arxiv.org/abs/2109.09774