AI - AI Dynamics

Autoregressive Sampling: Language Models Generate Hallucinations by Design

By

–

09 December 2023 4h36

When we autoregressively sample from a language model, we are, by definition, seeing its hallucination. Users may think otherwise only when the LLM’s “dream” just so happens be an output that is acceptable to the user.

→ View original post on X — @hardmaru

9 December 2023

Adversarial Training Framework Improves LLM Robustness

By

@googleai

–

09 December 2023 4h28

Drop by the #EMNLP2023 Google booth today at 12:30 PM to learn about an adversarial training framework that uses limited human adversarial examples to generate and scale more useful synthetic examples that improve LLM robustness to unseen attacks.

→ View original post on X — @googleai

9 December 2023

Open-source models will outpace closed-source in the long term

By

@soumithchintala

–

09 December 2023 3h18

the most convincing evidence I've seen so far that open-source will in-due-time accelerate away far beyond closed-source models.

People just want to use these models to best fit their use-case; a single company's closed-source effort rate-limits against people's imaginations and… https://t.co/Vw9Yp131lf
— Soumith Chintala (@soumithchintala) 9 décembre 2023

the most convincing evidence I've seen so far that open-source will in-due-time accelerate away far beyond closed-source models. People just want to use these models to best fit their use-case; a single company's closed-source effort rate-limits against people's imaginations and

→ View original post on X — @soumithchintala

9 December 2023

Pika 1.0 Expand Canvas Transforms 16:9 Videos Into Anamorphic Masterpieces

By

@pika_labs

–

09 December 2023 3h02

🌟Turn your 16:9 videos into anamorphic masterpieces with the Pika 1.0 'Expand Canvas' feature. 🌟

Sign up at https://t.co/nqzjGy82Lx pic.twitter.com/5Oxq2j4arm
— Pika (@pika_labs) 9 décembre 2023

Turn your 16:9 videos into anamorphic masterpieces with the Pika 1.0 'Expand Canvas' feature. Sign up at http://
pika.art

→ View original post on X — @pika_labs

9 December 2023

Video-LLaVA: Unified Visual Representation Learning for Multimodal AI

By

@miketamir

–

09 December 2023 3h01

GitHub – PKU-YuanGroup/Video-LLaVA: Video-LLaVA: Learning United Visual Representation by Alignment Before Projection https://
bit.ly/3Rab0KS #AI #MachineLearning #DeepLearning #LLMs #DataScience

→ View original post on X — @miketamir

9 December 2023

Mistral’s Bold Strategy: Sharing AI Models Without Context

By

@jxmnop

–

09 December 2023 2h08

say what you will about mistral, tweeting exclusively download links to new models with no context is unbelievably cool

→ View original post on X — @jxmnop

9 December 2023

1-Pager: Unified Retrieval and Reasoning with Single Transformer

By

@googleai

–

09 December 2023 1h58

Current retrieval paradigms are complex, pipelined systems. Visit the #EMNLP2023 Google booth today at 10:30 AM to learn about 1-Pager, a system that aligns retrieval and reasoning as a unified task with a single Transformer-based model and decoding pass.

→ View original post on X — @googleai

9 December 2023