@ylecun It's maybe time to get back to good old knowledge rep. Text -> LLM -> Description Logics with unary and binary predicates only.
User query -> LLM -> DL -> Fuzzy Unification -> LLM return user friendly proof.
One model, dynamic knowledge
@GaryMarcus@kareem_carr Been a @GaryMarcus "fan" for some time but man, you are taking this way too far. Why don't you build something... anything. ChatGPT is not perfect but it sure bring some theoretical and practical progress.
Let's be honest. โ๏ธ Large models like #huggingface transformers are expensive to run.
In this FREE webinar, we present our cost-effective and yet scalable solution. https://t.co/2LJX0VkM6D
#Transformers#NLP#MachineLearning#webinar
Generating text with #Cheapity3 can prove to be a great and cheap resource for multilingual data generation.
This simple frontend was built with 3 lines of code using @Gradio.
The model is on the HF hub and Github.
https://t.co/OhRf9DF3KA
https://t.co/l4iiCjGguY
Looking for a cheap open-source multilingual (๐ฉ๐ช๐ซ๐ท๐ฌ๐ง) alternative to " #gpt3 " ? In particular for academic text?
Well, then Merry Christmas. ๐๐. T5 available on @huggingface hub and on GitHub.
https://t.co/l4iiCjGguY
https://t.co/OhRf9DF3KA
#NLP#Transformers#DeepLearning
@GaryMarcus Like humans, machines must CRUD knowledge on the fly. GOFAI reached its peak in the 90s. ML has now reached its peak.
What if scale just helps us acquire knowledge on the fly (intuition) and GOFAI helps us reason on it? end 2 end of course.
@BethCarey12@marktenenholtz@GaryMarcus I have followed Patom Theory for a while now and although interesting at first sight, it still feels a bit sketchy. I highly doubt the progress made there is a result of John Ball's theory or if RRG is what works there - recycling frame semantics.
@psuraj28 @PatrickPlaten Would you use decoding strategies other than beam-search for use cases where correctness of generated output is crucial? ๐ค๐
Weโre releasing the 1.0 version of Opacus, a #PyTorch training library that makes it easier for researchers to adopt differential privacy in #ML. Opacus 1.0 will accelerate differential privacy research in the field. Learn more: https://t.co/rvmDLsaunl
There is a new challenging EXtractive QA dataset. Question operators are based on didactic principles like Bloom's taxonomies.
Want to see how well your QA model performs? Download it at: https://t.co/cre2INLE9y or on the @huggingface hub. #NLP#DeepLearning
๐ฏ The Conceptor is now available on the @huggingface ๐ค hub. Generate concepts or types E.g A dog is an animal.
Ideal for Zero-shot tasks, logical reasoning, NER, QA or even text or intent classification pipelines.
https://t.co/CPDm0XAVT7
#NLP#ArtificialIntelligence