Thanks.
My call for world models and planning goes back a long time.
A good example is my NeurIPS 2016 keynote (slide 31 on). https://
drive.google.com/file/d/0BxKBnD
5y2M8NREZod0tVdW5FLTQ
…
RESEARCH
-
World Models and Planning: A Long-Standing AI Research Call
By
–
-
AI Safety: Engineering Challenge Like Turbojet Safety
By
–
I believe there is as much to worry about AI safety as there is to worry about turbojet safety.
Making turbojets safe and reliable is a very hard engineering problem that is in the capable hands of specialists.
I'm only dismissive of the craziest claims, which should not be taken -
Distil-Whisper: Advanced AI Voice-to-Text Technology Explained
By
–
Distil-Whisper Explained – The most recent AI Voice-to-Text Technology! Learn more in the video: https://
youtu.be/SZtHEKyvuug Thanks to @sanchitgandhi99 and others from @huggingface for this great work, paper, and open-source model. -
Meta-FAIR and DeepMind Employ Neuroscientists Unlike OpenAI
By
–
Both Meta-FAIR and DeepMind have neuroscientitst and cognitive scientists.
OpenAI doesn't, as far as I know. -
Top AI Labs Race to Add Planning Reasoning Capabilities
By
–
No.
Everyone in top AI research labs is working on giving dialog systems the ability to plan & reason.
There are projects along those lines at Meta-FAIR, DeepMind, and OpenAI, with early results (if you follow the literature).
Q* is just one such project among many. The planning -
Q-Learning Graph Networks for SAT Solver Branching Heuristics
By
–
Can Q-Learning with Graph Networks Learn a Generalizable Branching Heuristic for a SAT Solver? Kurin et al.: https://
arxiv.org/abs/1909.11830 #DeepLearning #QLearning #ReinforcementLearning -
AI Systems Learning from Sensory Inputs and Visual Data
By
–
That has been one of my main arguments for years: AI systems need to learn how the world works from sensory inputs (e.g. visual inputs).
-
LLMs Struggle With Logical Equivalence in Prompts
By
–
That statement is easily retrived.
But tests have shown that when faced with a situation in which the answer requires to know that B is the same as A, when the prompt clearly says that A is the same as B, the best LLMs often don't answer correctly. -
Human Genome Storage Capacity vs LLM Requirements Analysis
By
–
I'm not.
There just isn't enough capacity in the genome.
Your entire genome fits in 800MB (uncompressed).
The difference between the human and chimp genomes is 1% of that, or 8MB.
Not enough to encode a significant structure.
For comparison, a small 7B LLM requires 14GB. -
A* Search Without Expansions: Deep Q-Networks Heuristics
By
–
A* Search Without Expansions: Learning Heuristic Functions with Deep Q-Networks Agostinelli et al.: https://
arxiv.org/abs/2102.04518 #ArtificialIntelligence #DeepQNetworks #ReinforcementLearning