News 18 June 2023 - simon-oz/Weekly-AI-news GitHub Wiki

  1. Andrew Ng and Geoff Hinton had an insightful conversation. They want to share (i) It's important that AI scientists reach consensus on risks-similar to climate scientists, who have rough consensus on climate change-to shape good policy. (ii) Do AI models understand the world? We think they do. If we list out and develop a shared view on key technical questions like this, it will help move us toward consensus on risks.
  2. On 12 June 2017, Google published its outstanding paper: “Attention is All You Need” which introduced the transformer structure – an essential element widely used in nearly all large deep learning models, both in NLP and Computer Vision. The paper has been cited over 75K, and of eight authors, only one still working in Google.
  3. A new LLM evaluation leaderboard is released by researchers from UTD Singapore. The proposed model evaluated three features of LLMs: Problem-Solving, Writing, and Alignment (Harmless, Honesty and Helpfulness)
  4. On 13th June, MetaAI announced I-JEPA, the first AI model based on Yann LeCun’s vision for more human-like AI, which is to create machines that can learn internal models of how the world works so that they can learn much more quickly, plan how to accomplish complex tasks, and readily adapt to unfamiliar situations.
  5. On 14th June, OpenAI released new GPT-4 and GPT-3.5 Turbo models with 1) function calling in the API (plugins); 2) 16K context 3.5 Turbo model available to everyone; 3) 75% price reduction on v2 embedding models.
  6. McKinsey released “The economic potential of generative AI: The next productivity frontier”. 1) Generative AI could add $2.6 to $4.4 Trillion in value to the global ecomony; 2) 75% of the value falls in: Customer operations, marketing and sales, software engineering and R&D; 3) Generative AI will have impact across all industry sectors; 4) 50% today’s work activities could be automated between 2030 and 2060; 5) Generative AI is just beginning.
  7. On 14 June, The sequoia published “The New Language Model Stack” which describes how companies are bringing AI applications to life. 1) Nearly every company in the Sequoia network is building LLMs into their products; 2) The new stack centers on LLM APIs, retrieval, and orchestration, but open source usage is also growing; 3) Companies want to customize LLMs to their unique context; 4) LLMs need to become more trustworthy (output quality, data privacy, security) for full adoption
  8. On 15th June, Princeton Uni published a paper “Infinite Photorealistic Worlds using Procedural Generation”. It’s worth noting that Infinigen is entirely procedural: every asset, from shape to texture, is generated from scratch via randomized mathematical rules, using no external source and allowing infinite variation and composition.
  9. On 16th June, Meta published a paper “Voicebox: Text-Guided Multilingual Universal Speech Generation at Scale”. Voicebox is a non-autoregressive flow-matching model trained to infill speech, given audio context and text, trained on over 50K hours of speech that are neither filtered nor enhanced. Voicebox not outperforms the SOT zero-shot TSS model, but also up to 20 times faster
  10. IBM Makes the Best Quantum Computer Open to Public - IBM in collaboration with UC Berkeley researchers announced a recent breakthrough experiment which indicates that quantum computers will soon surpass classical computers in practical tasks.