TLDR AI 2024-03-29

Jamba Language Model 🤖, Hume Empathic Voice Interface 🗣️, Long Form Factuality in LLMs 📚

Headlines & Launches

AI21 releases Jamba Language Model (6 minute read)

Mamba is a style of model that is supposed to beat Transformers in efficiency while matching performance. Jamba is a novel variant that includes MoE layers. It can run at 1.6k tokens per second with a context length of 128k tokens. It achieves 67% on the MMLU benchmark. The weights are available.

Hume introduces Empathic Voice Interface (3 minute read)

Hume has released the Empathic Voice Interface (EVI), the first conversational AI with emotional intelligence.

Google starts testing AI overviews from SGE in main Google search interface (2 minute read)

Google is piloting AI-generated overviews in search results for complex queries with a subset of U.S. users without requiring opt-in to its Search Generative Experience. This feedback-driven test aims to enhance user experience and may impact site traffic dynamics due to changes in ad placements and organic search visibility.
Research & Innovation

1 Bit, Post Training, Quantization (12 minute read)

1 Bit language models are exciting. This work shows how to quantize the Linear layers of a language model without sacrificing performance. This can result in a 70B model running on consumer GPUs.

Long Form Factuality in Large Language Models (25 minute read)

You can use language models to generate factual information. Google has released a dataset and benchmarks to show how well each model performs. This report gives insight into how to improve a model’s factuality and shows language models winning against human annotators in a majority of cases.

Improving Model Adaptation (26 minute read)

CoDA is a new approach to Unsupervised Domain Adaptation (UDA). It helps AI models better adapt to unlabeled, challenging environments by learning from differences at both the scene and image levels.
Engineering & Resources

3D Medical Imaging Dataset (GitHub Repo)

CT-RATE is a dataset that marries 3D medical imaging with textual reports. CT-CLIP is a versatile AI framework optimized for these images.

Using AI to Identify Sources of Confusion in Your Content (GitHub Repo)

This approach pinpoints the exact content that triggers user questions online, from lectures to news articles. This method, known as backtracing, aims to help content creators refine their work by identifying and understanding the causes of confusion, curiosity, or emotional reactions.

High-Resolution Large Language-Vision Assistant (GitHub Repo)

Researchers have developed a new method, Mixture-of-Resolution Adaptation (MRA), which improves AI's ability to recognize details in images.

Exponential Growth and AI Escape Velocity (20 minute read)

AI Futurist Ray Kurzweil invented the law of accelerating returns, which states that the rate of progress is increasing at an exponential rate over time. In a recent discussion, Kurzweil explored various topics including prospects that are only getting better, the direction in which the AI economy is headed, human relationships with AIs, longevity escape velocity, and much more.

Plentiful, high-paying jobs in the age of AI (16 minute read)

AI experts are exploring automating human tasks, leading to concerns about job loss and wage declines. However, the view that AI advancement will inevitably result in human job obsolescence may not be correct. Humans may retain employment in an AI-dominated future due to constraints like computing power and opportunity costs, although this scenario is not guaranteed.
Quick Links

Coframe Living Images (Product Launch)

Coframe’s generative A/B testing tool allows you to optimize images on your website in minutes with generative AI.

Meta Is Adding AI To Its Ray-Ban Smart Glasses (1 minute read)

Meta is introducing AI features to its Ray-Ban smart glasses, including translation, and the identification of objects, animals, and monuments, with activation through voice commands.

Google Bringing Gemini Nano To Pixel 8 (1 minute read)

The Pixel 8 will receive Gemini Nano in a developer preview to enhance features like Summarize in Recorder and Gboard Smart Reply.
The most important AI, ML, and data science news in a free daily email.
Join 500,000 readers for