Issue 50 30th April 2024

Highlights

OpenAI’s commitment to child safety: adopting safety by design principles

We’re joining Thorn, All Tech Is Human, and other leading companies in an effort to prevent the misuse of generative AI to perpetrate, proliferate, and further sexual harms against children.

Towards infinite LLM context windows

From 512 to 1M+ tokens in 5 years — LLMs have rapidly expanded their context windows. Where’s the limit?

Microsoft’s heavy bet on AI pays off as it beats expectations in latest quarter

World’s largest public company reports $61.86bn revenue after investing billions into artificial intelligence Profits at Microsoft beat Wall Street’s expectations as its heavy bet on artificial intelligence continued to bear fruit in the latest quarter. The technology giant has invested billions of dollars into AI in a bid to turbocharge its growth, particularly of its cloud computing services. Its cloud computing revenue surged by more than 20%.

Copilot Workspace is GitHub’s take on AI-powered software engineering

GitHub's new Copilot Workspace brings AI-powered assistive software engineering tools front and center.

Research

Constructing knowledge graphs using LLMs

RAG systems are cool, but sometimes can't come up with a good answer. This is where you may need to structure knowledge to provide the LLM with higher quality knowledge. This week's paper covers research on how to use LLMs to create knowledge graphs. It also covers some interesting use-cases once you have a knowledge graph.

Video

AI Show | A Tour through RAI with a dash of Phi (3 that is)

This week on The AI Show, join us to learn more about the pragmatic aspects of Responsible AI, how to think about it, and how to write validation code. We wi...

Articles

BBC presenter’s likeness used in advert after firm tricked by AI-generated voice

Exclusive: Liz Bonnin’s face used on insect repellant ads after faked message ostensibly granted permission There was something strange about her voice, they thought. It was familiar but, after a while, it started to go all over the place. Science presenter Liz Bonnin’s accent, as regular BBC viewers know, is Irish. But this voice message, ostensibly granting permission to use her likeness in an ad campaign, seemed to place her on the other side of the world.

Llama-2 vs. Llama-3: a Tic-Tac-Toe Battle Between Models

Making a non-scientific benchmark with Python and Llama-CPP

Introducing more enterprise-grade features for API customers

Increasing enterprise support with more security features and controls, updates to our Assistants API, and tools to better manage costs.

How do you accidentally run for President of Iceland?

Anna Andersen writes about a spectacular user interface design case-study from this year's Icelandic presidential election. Running for President requires 1,500 endorsements. This year, those endorsements can be filed online through a government website.

Mapping the brain pathways of visual memorability

For the first time, researchers use a combination of MEG and fMRI to map the spatio-temporal human brain dynamics of a visual image being recognized.

‘Eugenics on steroids’: the toxic and contested legacy of Oxford’s Future of Humanity Institute

Two weeks ago it was quietly announced that the Future of Humanity Institute, the renowned multidisciplinary research centre in Oxford, no longer had a future. It shut down without warning on 16 April. Initially there was just a brief statement on its website stating it had closed and that its research may continue elsewhere within and outside the university.

Google’s new ‘Speaking practice’ feature uses AI to help users improve their English skills

Google is testing a new “Speaking practice” feature in Search that helps users improve their conversational English skills. The company told TechCrunch that the feature is available to English learners in Argentina, Colombia, India, Indonesia, Mexico, and Venezuela who have joined Search Labs, its program for users to experiment with early-stage Google Search experiences.

Fast Conformer with Linearly Scalable Attention for Efficient Speech Recognition

Conformer-based models have become the dominant end-to-end architecture for speech processing tasks. With the objective of enhancing the conformer architecture for efficient training and inference, we carefully redesigned Conformer with a novel downsampling schema.

Ideas: Exploring AI frontiers with Rafah Hosn

Energized by disruption, partner group product manager Rafah Hosn is helping to drive scientific advancement in AI for Microsoft. She talks about the mindset needed to work at the frontiers of AI and how the research-to-product pipeline is changing in the GenAI era.

Hugging Face releases a benchmark for testing generative AI on health tasks

As I wrote recently, generative AI models are increasingly being brought to healthcare settings — in some cases prematurely, perhaps. Early adopters believe that they’ll unlock increased efficiency while revealing insights that’d otherwise be missed. Critics, meanwhile, point out that these models have flaws and biases that could contribute to worse health outcomes.

Upcoming events

AI Community Day - 14 May 2024

Welcome to AI Community Day, where all the AI developers come together to geek out and share what they know about artificial intelligence. It's a chance for us to connect, learn from each other, and get inspired by the latest and greatest in AI. Whether you're a pro at AI or just getting started, this event is all about building our community and shaping the future of AI development.

Code

GitHub - guidance-ai/guidance: A guidance language for controlling large language models.

A guidance language for controlling large language models. - guidance-ai/guidance