#154 - Google Gemini, Waymo Collision, Smaug-72B, EU AI Act final text, image watermarks
Feb 11, 2024
auto_awesome
Discover the latest advancements in AI with Google's Gemini, a game-changing chatbot and voice assistant. Delve into the complexities of self-driving cars, sparked by a recent Waymo incident. Explore cutting-edge tools like Brilliant Labs’ Frame glasses and Stability AI’s innovative diffusion model for videos. Learn how artists are fighting back against AI appropriation with protective tools. Plus, find out how Meta is ensuring transparency with AI-generated images and the implications of new regulations in the evolving AI landscape.
Google releases Gemini, an AI-driven chatbot and voice assistant
Waymo robotaxi hits a cyclist in San Francisco - incident highlights the challenges of autonomous vehicle technology
Canon plans to disrupt chipmaking with low-cost 'stamp' machine
Meta adds visible and invisible watermarks to AI-generated images to combat misinformation and ensure content authenticity
Deep dives
Allen Institute for AI releases open and transparent AI language models
Allen Institute for AI has released open language models that include the model, training code, and data for training. The models include almost 7B, which outperforms GPT 3.5 and Mistral medium across various benchmarks.
New Quinn 1.5 language model surpasses previous records
Quinn 1.5, developed by Alibaba, is a fine-tuned version of their previously released model. It outperforms previous models on various benchmarks and introduces quantized models along with larger context lengths.
Hugging Face launches open-source AI assistant maker
Hugging Face has introduced an open-source AI assistant maker, allowing users to build their own chat assistants based on open-source language models. This comes as a rival to OpenAI's custom GPTs and offers free access to the models.
MGIE: Apple and UCSB develop AI-assisted image editing model
MGIE, developed by Apple and UCSB researchers, uses multi-model language models to allow users to edit images based on text instructions. This open-source model enables image editing using natural language commands.
Google DeepMind explores universal predictors in theoretical paper
Google DeepMind's theoretical paper discusses learning universal predictors, focusing on Solomonoff induction, the most powerful universal predictor that determines the function generating a set of observed data.
AI-generated scams are a concern for regulators
Researchers have developed tools like Nightshade and Glaze to disrupt AI models from scraping artists' work without consent. Nightshade has received 250,000 downloads in five days, demonstrating artists' concern over unauthorized use of their work. These tools aim to prevent AI models from learning an artist's signature style or using their work for training purposes without permission.
Meta implements image labeling and watermarking to detect AI-generated content
Meta, formerly known as Facebook, is adding visible markers and invisible watermarks to photo-realistic images generated using AI on its platforms. These measures aim to combat misinformation and inauthentic content. Additionally, users will now be able to disclose when they share AI-generated material. Meta is collaborating with the C2PA coalition to develop industry standards for content provenance and authenticity.
AI-generated George Carlin comedy special was actually human-written
Following a lawsuit, it was revealed that the AI-generated George Carlin comedy special was actually human-written. The creators acknowledged that there was significant human involvement in writing the special, despite previously claiming it was AI-generated. This case highlights the novelty and controversy surrounding AI-generated content.
