Issue 72 1st October 2024

Highlights

Hidden traces of humanity: what AI images reveal about our world

When faced with a bit of downtime, many of my friends will turn to the same party game. It’s based on the surrealist game Exquisite Corpse, and involves translating brief written descriptions into rapidly made drawings and back again. One group calls it Telephone Pictionary; another refers to it as Writey-Drawey. The internet tells me it is also called Eat Poop You Cat, a sequence of words surely inspired by one of the game’s results.

Microsoft claims its new tool can correct AI hallucinations, but experts advise caution

Microsoft today revealed Correction, a service that attempts to automatically revise AI-generated text that's factually wrong.

OpenAI’s VP of global affairs claims o1 is ‘virtually perfect’ at correcting bias, but the data doesn’t quite back that up

Departures might be dominating the week’s OpenAI-related headlines. But comments on AI bias from Anna Makanju, the company’s VP of global affairs, also grabbed our attention. Makanju, speaking on a panel at the UN’s Summit of the Future event on Tuesday, suggested that emerging “reasoning” models such as OpenAI’s o1 have the potential to make […]

Research

An exploration of various video summarization techniques

With all the videos available for conference talks and online events, it can be interesting to get a summary of the content in the videos. But where do you start? This paper explores several options to summarize videos using neural networks automatically.

Video

Nanddeep Nachan - Humans in AI

Explore the fascinating world of AI with Nanddeep Nachan, a Microsoft MVP for M365 Apps and Services. Discover how AI acts as a companion in tech, enhancing natural language processing and intelligent applications. Nanddeep shares his excitement about the latest AI models and how they integrate with Microsoft 365 applications. Ready to see AI in action?

Episode 304 - Copilot Studio with Daniel Laskewitz

Join us in this episode where we talk with Daniel Laskewitz, Senior Cloud Advocate at Microsoft about Copilot Studio.

Articles

Will the ‘Goodbye Meta AI’ message protect users’ posts from being used to train AI?

Though the message has been shared by many users, including celebrities, it offers no copyright or privacy protection The “Goodbye Meta AI” message, which purports to protect the user from having the likes of Facebook, Instagram and WhatsApp use their accounts as an AI training camp, has become an increasingly common feature on timelines. It has been shared by actors and sports stars – including James McAvoy, Ashley Tisdale and Tom Brady – as well as hundreds of thousands of others. But why – and what effect, if any, will it have?

Understanding Generative AI and Large Language Models (LLMs)

The field of artificial intelligence has made significant strides in recent years, with generative AI emerging as one of the most exciting areas of research and application. This branch of AI focuses on the creation of content, whether it be text, images, music, or other forms of media, using sophisticated algorithms and models. Among the key players in this domain are Large Language Models (LLMs), which have transformed the way machines understand, generate, and interact with human language.

Stress-testing biomedical vision models with RadEdit: A synthetic data approach for robust model deployment

This paper has been accepted at the 18th European Conference on Computer Vision (ECCV 2024) (opens in new tab), the premier gathering on computer vision and machine learning. Biomedical vision models are computational tools that analyze medical images, like X-rays, MRIs, and CT scans, and are used to predict medical conditions and outcomes. These models […]

Hands-On Optimization Using Genetic Algorithms, with Python

Here’s a full guide on genetic algorithms, what they are, and how to use them.

NotebookLM's automatically generated podcasts are surprisingly effective

Audio Overview is a fun new feature of Google's NotebookLM which is getting a lot of attention right now. It generates a one-off custom podcast against content you provide, where two AI hosts start up a “deep dive” discussion about the collected content. These last around ten minutes and are very podcast, with an astonishingly convincing audio back-and-forth conversation.

Upgrading the Moderation API with our new multimodal moderation model

We’re introducing a new model built on GPT-4o that is more accurate at detecting harmful text and images, enabling developers to build more robust moderation systems.

To Mask or Not to Mask: The Effect of Prompt Tokens on Instruction Tuning

In the last several months I’ve noticed quite a few discussions, here and there, even over here, on the question of whether or not to zero-mask (ignore) prompt tokens when fine-tuning on prompt-completion style data (i.e. instruction-tuning). In this setting, the goal of fine-tuning is to increase (ultimately maximize ) the probability that the LLM will generate the completion when given the prompt as input.

PyTorch Native Architecture Optimization: torchao

We’re happy to officially launch torchao, a PyTorch native library that makes models faster and smaller by leveraging low bit dtypes, quantization and sparsity. torchao is an accessible toolkit of techniques written (mostly) in easy to read PyTorch code spanning both inference and training. This blog will help you pick which techniques matter for your workloads.

Proto-CLIP: Vision-Language Prototypical Network for Few-Shot Learning

We propose a novel framework for few-shot learning by leveraging large-scale vision-language models such as CLIP. Motivated by unimodal prototypical networks for few-shot learning, we introduce Proto-CLIP which utilizes image prototypes and text prototypes for few-shot learning. Specifically, Proto-CLIP adapts the image and text encoder embeddings from CLIP in a joint fashion using few-shot examples. The embeddings from the two encoders are used to compute the respective prototypes of image classes for classification.

Exploring the World of Markov Chains: Unlocking the Power of Probabilistic Transitions

An Introduction to Markov Chains, their applications, and how to use Monte Carlo Simulations in Python

Google will begin flagging AI-generated images in Search later this year

Google says that it plans to roll out changes to Google Search to make clearer which images in results were AI generated — or edited by AI tools. In the next few months, Google will begin to flag AI-generated and -edited images in the “About this image” window on Search, Google Lens, and the Circle […]