Investigates the use of existing LMs (e.g, Flan-U-PaLM 540B) combined with associative read-write memory to simulate the execution of a universal Turing machine. 7 of 11
AI
-
New Multimodal Deep Learning Book Published on ArXiv
By
–
Multimodal deep learning is a new book published on ArXiv. 4 of 11
-
OpenAI Research on Generative LM Disinformation Risks
By
–
OpenAI publishes new work analyzing how generative LMs could potentially be misused for disinformation and how to mitigate these types of risks. 5 of 11
-
DeepMind Tracr: Compiler Converting RASP to Transformer Weights
By
–
DeepMind proposes Tracr, a compiler for converting RASP programs into transformer weights. This way of constructing NNs weights enables the development and evaluation of new interpretability tools. 3 of 11
-
Top Machine Learning Papers of the Week
By
–
Top ML Papers of the Week (Jan 9-15): – DreamerV3
– DeepMatcher
– Multimodal deep learning
– Transformer compiler for RASP
– Potential misuses of LMs and mitigations
– Scaling laws for generative mixed-modal LMs
– Time series forecasting with generative modeling
… 1 of 11 -
DreamerV3 Achieves Minecraft Diamond Collection Without Human Data
By
–
DreamerV3 is a general algorithm to collect diamonds in Minecraft from scratch without human data or curricula, a long-standing challenge in AI.
— DAIR.AI (@dair_ai) 15 janvier 2023
2 of 11https://t.co/pvsmnX3TPODreamerV3 is a general algorithm to collect diamonds in Minecraft from scratch without human data or curricula, a long-standing challenge in AI. 2 of 11
-
Machine Learning Types and Applications Infographic Guide
By
–
Want to know more about #machinelearning? Check out the #infographic for a breakdown of the different types and their applications.
Learn more with @ingliguori #AI #deeplearning #DigitalTransformation #ArtificialIntelligence #Python #datascience #NeuralNetworks #GenerativeAI -
Open Architecture vs Fair Data Compensation in AI
By
–
Seems to me that there are two orthogonal issues:
– should your architecture be open? – where do you get your data from/are contributors justly compensated? Both are tricky. But virtue in one doesn’t logically entail virtue in the other. -
Computer CoPilot Vision Now Tractable Through Text Translation
By
–
Computer CoPilot. Was very much the vision with OpenAI Universe https://
openai.com/blog/universe/ , though it was too early. Now feels tractable if you translate everything to/from text (e.g. like in WebGPT). Could be built e.g. as an extension of natbot https://
github.com/nat/natbot -
LLMs Lack True Compositional Reasoning Capabilities
By
–
Read http://
rebooting.ai; it went to press just before LLMs became popular but lays out many challenges for genuine intelligence. I don’t see how LLMs resolve any of them. It’s not compositional and can’t reason, so it’s just a giant but superficial statistical approximator