Explore the most significant talks in the AI world last month.
The Biggest AI Drops of the Month
Probably the most talked about release last month was GPT-4, a large language model developed by OpenAI. The new model has become the next step in OpenAI’s journey to develop AGI — artificial general intelligence. It’s not only able to communicate in a conversational way, help write code, and generate text you can’t tell from human writing, but it also fundamentally outperforms its predecessor, ChatGPT, in many tasks. GPT-4 understands mediums other than text and spreadsheets and can recognize image input as well as generate captions, classifications, and analysis. GPT-4 is currently available for ChatGPT Plus subscribers. In addition, OpenAI has begun supporting plugins for it, allowing developers to create third-party services powered by GPT-4.
Adobe has introduced Adobe Firefly, a suite of generative AI models that focused on visual content creation. After enabling the submission of AI-generated content to its platform a few months ago, the launch of Adobe Firefly is the Adobe’s next logical step in the generative AI rush.
Using generative algorithms, Adobe Firefly includes a toolkit of models that can be used to create various types of visual content – from illustrations, art concepts, and photos to “creative ingredients” such as brushes, color gradients, or video transformations – with the click of a button or the typing of a few words. While the product is currently in beta, there are a few options available, such as generating images and text effects from text prompts. To get access to it, you will need to fill out the form. The new software is expected to be integrated into existing Adobe workflows such as Creative Cloud, Document Cloud, Experience Cloud, and Adobe Express.
The other big news this month came from Midjourney, which unveiled a new model of its algorithm. The results of Midjourney v.5 have improved efficiency, coherence, and higher resolution.
Soon after the release, some fake images went viral: from Donald Trump being arrested to the Pope wearing a stylish puffer jacket. Some of them were interpreted by someone as real. This fact probably led Midjourney to stop the free trials “due to a combination of extraordinary demand and trial abuse,” as the Midjourney CEO, David Holz, explained on Discord. However, as he told The Verge, the latest version of Midjourney was originally available only to paid users, and the reason for suspending free trials was an influx of newcomers and high demand.
Google has opened up early access to its conversational AI service, Bard. The experiment has started for users in the US and UK, while the company promises to expand the list of countries and languages available to interact with the new tool. For now, users can join a waitlist to get access as soon as it is delivered to their location.
Bard, which is based on Google’s large language model LaMDA, is considered the closest competitor to OpenAI’s ChatGPT. For comparison, LaMDA was trained on 137 billion parameters and ChatGPT was trained on 175 billion, which is not a huge difference (although GPT-4 surpassed them all with a dataset of one trillion parameters). And both Google’s Bard and Microsoft’s Bing, powered by GPT-4, have been melting our hearts in recent months with their responses full of human-like emotion. One of them even made the front page of The New York Times, while another went viral on Twitter.
Tech Community Urged to Halt AI Development in Open Letter
A group of more than 1.800 notable signatories, including Elon Musk, Steve Wozniak, and Yuval Noah Harari, have signed an open letter urging all AI labs to pause training AI systems more powerful than GPT-4 for at least six months.
“Powerful AI systems should be developed only once we are confident that their effects will be positive and their risks will be manageable.” — the letter said.
A moratorium would allow time for ethical considerations and discussions about potential risks. By engaging diverse stakeholders in these discussions, the community should develop a set of shared safety protocols for advanced AI design and development. These protocols should ensure transparency in the development and use of AI, and make these systems more accurate, secure, interpretable, and aligned.
At the end of 2022, in our post predicting AI trends for 2023, we expected AI safety to be one of the top topics during this year, with many companies hiring people for AI safety research roles, startups raising investments for AI research efforts, and organizations opening fellowships and education programs aimed at raising awareness of AI, its risks and benefits. However, in a move that is not in line with this trend, Microsoft earlier this month laid off its ethics team, which was focused on guiding the safety of AI innovation. According to Platformer, this team was as large as seven people in October 2022.
Stability AI Acquires Init ML, a ClipDrop App Maker
Stability AI has acquired Init ML, the creators of the ClipDrop app, for an undisclosed amount.
ClipDrop is a platform for creators, that became popular in 2020 thanks to its app, which allows users to capture an image with their smartphone or tablet and then paste it onto their computer screen without the need for a scanner.
With this acquisition, Stability AI aims to integrate its AI technologies into ClipDrop’s ecosystem, which includes a set of tools for image editing, such as removing objects, text, and backgrounds from photos, improving image quality, and cleaning them up by scrubbing away imperfections or adjusting lighting settings.
“This acquisition will provide Clipdrop with the resources it needs to advance the next era of generative AI, marking a new milestone in our shared journey to democratize the power of AI and multimodal foundation models in creative workflows,” — said Emad Mostaque, founder and CEO of Stability AI.
U.S. Copyright Office Issues a New Guidance on AI-assisted Works
The U.S. Copyright Office has issued a statement saying that AI-assisted works may be eligible for copyright protection. This move comes after months of debate and discussion on the issue, as we mentioned in one of these cases in February. In that case, the Copyright Office denied the copyright protection for AI-generated images claimed by Kris Kashtanova in the copyright registration for the book “Zarya of the Dawn”.
The new guidance confirms that works created by humans using AI tools should be evaluated based on whether the human role in the creation of those works was determinative. Thus, if the works were created “without any creative input or intervention from a human author”, they wouldn’t be registered. However, the Copyright Office’s guidance has also sparked a debate about the extent of AI’s role in creative works.
Osmo Brings AI to the World of Smells with the Help from Google Cloud
Osmo, a fragrance tech startup that “gives computers a sense of smell,” has partnered with Google Cloud to explore how AI can be used to “map our sense of smell” by cataloging a wide range of odor molecules and training algorithms to find out patterns and make predictions about them. There are about 40 billion molecules that have an odor, of which only 100 million are known. Osmo’s solution is supposed to discover unknown molecules that no one has ever smelled before.
Osmo says its mission is to build tools that help to create new fragrance experiences and improve human health and well-being. Using AI to better understand the world of smells can also help “to detect diseases earlier, track pandemics faster, grow more food, ward off insects, and much more,” — stated a press-release.