AI Weekly Summary

Welcome to a new edition of the AI Weekly Summary!

This week has been packed with court rulings, new partnership announcements and new releases, so let’s not waste any more time and dive in.

Early on this week, I shared an article where a U.S. court in Washington, D.C., ruled that artworks created solely by artificial intelligence without human input are not eligible for copyright protection.

U.S. District Judge Beryl Howell upheld the Copyright Office's decision to deny a copyright application for an AI-created artwork titled "A Recent Entrance to Paradise," crafted by Stephen Thaler's DABUS system.

Thaler is a computer scientist and he has already faced similar setbacks with DABUS-generated patents globally, including in the UK, South Africa, Australia, and Saudi Arabia. However, he’s not ready to give up just yet so Thaler and his legal team are determined to appeal.

The main point of the court's decision is the principle that only human-created works qualify for copyright, a notion deeply rooted in legal tradition.

But the same questions still remain unanswered: how much human input is required and how is it quantified? Does prompt engineering count as human input? I guess only time will tell how much longer the opposition will last.

Despite US court rulings against copyrighting AI artwork, the industry is still going full steam ahead with a number of partnerships and funding being raised.

Let’s take the start-up Hugging Face, which has secured $235 million in a Series D funding round, with prominent investors like Salesforce and Nvidia participating.

If you’ve never come across them before, Hugging Face is a tech startup that provides a platform for developers to create, share and train AI models. You can find a myriad of applications, training datasets and tools that are shared and created by a very engaged community with the sole purpose of democratising good machine learning practices.

This new funding has been huge for Hugging Face, putting its valuation at $4.5 billion, doubling its valuation from 2022.

OpenAI has not fallen behind in the partnerships space and has joined forces with Scale to provide support for fine-tuning AI models for enterprises.

This partnership aims to offer companies the ability to customize and enhance the performance of advanced AI models like GPT-3.5 Turbo and the upcoming GPT-4 through secure and proprietary data fine-tuning.

In other words, this means more relevant and accurate results from fine-tuning the models and more security as companies can use their own data sets and not have to share them openly.

Also on the partnership front, Microsoft and Accenture have joined forces to address methane emissions using AI technology. They aim to detect, quantify, and remediate methane emissions in the energy industry, contributing to the fight against climate change.

On a different note, I couldn’t miss Meta’s announcements this week. The first one is about the release of Code Llama, a cutting-edge Large Language Model (LLM) designed specifically for coding.

It can generate code, provide explanations, and assist in debugging across multiple programming languages, including Python, C++, and Java. Although this is not a first as there are already other applications that already do it, the key is that it’s open source and free for personal and commercial use.

Benchmark tests show that Code Llama outperforms other open-source code-specific LLMs and even Llama 2. It scored well on tasks like code completion and writing code based on descriptions. It also underwent safety evaluations to avoid generating malicious code.

The second announcement came from Facebook. Facebook this week introduced SeamlessM4T, a multimodal and advanced AI translation model enabling effortless multilingual communication through speech and text. This means that it can support speech-to-text, speech-to-speech, text-to-text, and text-to-speech translation for various languages.

I’ve had a bit of a play myself and I have to say that the translation from English to Spanish is still a bit sketchy, especially when it comes to idioms and expressions.

However, despite coming late to the party, Meta is really pushing the envelope and catching up with the biggest AI players in the industry… watch this space!

And we couldn’t finish this week’s newsletter without mentioning another research article I shared this week.

AI-powered eye scans could detect Parkinson's disease before symptoms appear, according to a study by London's Moorfields Eye Hospital and UCL Institute of Ophthalmology. The method utilizes retinal markers to identify physical differences in the eyes of people with Parkinson's.

Although the study is still at its inception, it's all very promising. It could mean that treatment can be provided earlier and improve massively the quality of life of those affected.

And that's all for this week!

As always, I'm keen to hear your thoughts, so don’t be shy and share them in the comments.

Have a great week ahead!