Reminder: local optima were never a problem in machine learning, where the goal is generalization, not fitting the training data perfectly.
RESEARCH
-
Public Domain Data Availability for AI Model Training
By
–
Sure. But is there a surplus of such data in the public domain? Not totally clear to me. Either way, it feels qualitatively different from ImageNet style settings.
-
Harvard and OpenAI Create Effective AI Output Detector
By
–
Think ChatGPT can write all of your homework for you? Think again, for machine learning is a double-edged sword. Folks from Harvard teamed up with OpenAI to create an "output detector" to scan for AI-plagiarized text… …and it works wonderfully. Check this out
-
Music AI Reaches GPT-2 Moment of Breakthrough
By
–
There’s something there for sure. Seems like the GPT-2 moment for music.
-
Task Arithmetic: Editing AI Models Efficiently
By
–
Editing Models with Task Arithmetic Ilharco et al.: https://
arxiv.org/abs/2212.04089 #ArtificialIntelligence #DeepLearning #MachineLearning -
Intelligence as Accelerated Evolution, AI as Accelerated Intelligence
By
–
Intelligence is sped-up evolution, and AI is sped-up intelligence.
-
Calculator for Words: Reframing ChatGPT’s Core Function
By
–
i don’t quite agree with it, but “a calculator for words” is an interesting framing for ChatGPT
-
Montreal.AI Hosts AGI Debate 3 with 14,936 Registered Attendees
By
–
http://
MONTREAL.AI is hosting "AI DEBATE 3: The AGI Debate", and already 14,936 attendees have registered on LinkedIn. Follow the link for more information: https://
linkedin.com/events/aidebat
e3-thisisthe2021debateth6796651928362221568/
… #AGI #AGIDebate #AIDebate -
All-In Commitment and Optimism: Radical Virtues for Tech Innovation
By
–
all-in commitment is among the most valuable and differentiated virtues in the world right now. optimism is the most powerful (and somehow the most radical) worldview.
-
Model Fine-tuning Across Multiple Genres Discussion
By
–
I think it's only finetuned on some diff genres