Neural Notes: OpenAI launches text-to-video tool, fails to trademark ‘GPT’

Welcome back to Neural Notes: A weekly column where I condense some of the most interesting and important AI news of the week. In the latest edition, OpenAI launches a text-to-video tool, fails to trademark ‘GPT’ but gets a memory. Meanwhile, Ed Husic establishes an AI expert group and Slack’s AI features finally land.

OpenAI reveals Sora, a new text-to-video tool

OpenAI has just unveiled a text-to-video-generation tool named Sora that it says “can create realistic and imaginative scenes from text instructions”. It claims that Sora is able to craft videos that not only look photorealistic from written prompts but can do so in about a minute.

“Sora is able to generate complex scenes with multiple characters, specific types of motion, and accurate details of the subject and background. The model understands not only what the user has asked for in the prompt, but also how those things exist in the physical world,” OpenAI said in a blog post.

“We’re teaching AI to understand and simulate the physical world in motion, with the goal of training models that help people solve problems that require real-world interaction.”

OpenAI shared a video made with Sola on X (formerly Twitter) that was created from the prompt “Beautiful, snowy Tokyo city is bustling. The camera moves through the bustling city street, following several people enjoying the beautiful snowy weather and shopping at nearby stalls. Gorgeous sakura petals are flying through the wind along with snowflakes”.

More prompt-to-video examples were shared by OpenAI in its blog post, including ‘historical footage of California during the gold rush’ and ‘tour of an art gallery with many beautiful works of art in different styles’.

The results are admittedly quite impressive.

Introducing Sora, our text-to-video model.

Sora can create videos of up to 60 seconds featuring highly detailed scenes, complex camera motion, and multiple characters with vibrant emotions. https://t.co/7j2JN27M3W

Prompt: “Beautiful, snowy… pic.twitter.com/ruTEWn87vf

— OpenAI (@OpenAI) February 15, 2024

However, Sora is not without its quirks. While it can animate a scene based on a single image or even mend and extend the frames of an existing video, it occasionally reveals its artificial nature.

The museum scene, for example, shows some movement on the floor. And another scene following an SUV on a dirty road looks more like video game footage than real life.

“Simulating complex interactions between objects and multiple characters is often challenging for the model, sometimes resulting in humorous generations,” OpenAI admits in the blog post.

“Sora sometimes creates physically implausible motion.”

It’s currently unclear when Sora will be made available to the general public. For now it remains in the hands of a select few “red teamers,” tasked with identifying potential pitfalls and risks, alongside a curated group of visual artists, designers, and filmmakers.

This announcement comes on the heels of OpenAI’s decision to watermark creations from its DALL-E 3 tool, a move reflecting the broader concerns surrounding the generation of photorealistic AI content.

OpenAI doesn’t get to claim ‘GPT’ but does add memory

In other OpenAI news, it just received a denial from the US Patent and Trademark Office over its attempt to trademark the term “GPT”.

“Registration is refused because the applied-for mark merely describes a feature, function, or characteristic of applicant’s goods and services,” the denial document reads. 

As part of its submission, OpenAI said that it popularised “GPT” (generative pre-trained transformer). However, the patent office argued that the term was already used by other companies and in other contexts.

Elsewhere on the OpenAI blog, the company announced the rollout of memory capabilities to select ChatGPT users. This function lets the chatbot remember details from previous chats to better provide contextually charged and personalised responses.

“You’re in control of ChatGPT’s memory. You can explicitly tell it to remember something, ask it what it remembers, and tell it to forget conversationally or through settings. You can also turn it off entirely,” the blog post said.

Slack’s long-anticipated AI features have dropped

This week Salesforce — the owner of Slack — announced the release of Slack AI. First announced back in September, Slack AI is a suite of AI tools for the platform that includes:

• Channel Recaps: These will allow users to instantly gain insights from any channel as well as provide status reports.
• Thread Summaries: With this users can get a grasp on any ongoing thread discussions.
• Search Answers: This feature helps users to quickly extract information – such as messages or files. It’s also able to discern which information is most up-to-date.

“We want to make it as convenient as possible for people to get work done together. So when we figure out ways for people to work together that we can bring into Slack – we’ll try to figure out how to do that,” Slack’s VP of product, Ali Rayl, said on a call with SmartCompany when Slack AI was announced.

This is being offered as a paid add-on for Enterprise plans and is currently available in English for an undisclosed price. There are plans for a larger rollout and additional language support in the future.

Ed still loves AI

The Minister for Industry and Science, Ed Husic, is back with yet another AI announcement. Hot off the heels of the federal government’s interim response to the Safe and Responsible AI in Australia consultation, Minister Husic has announced the establishment of a new Artificial Intelligence Expert Group.

According to a release from the minister, the group first met in early February and will be providing advice to the Department of Industry, Science and Resources regarding “transparency, testing and accountability, including options for AI guardrails in high-risk settings, to help ensure AI systems are safe.”

The members of the group include representations from the CSIRO, the Centre for AI and Digital Ethics, Australia’s national AI standards committee and more.

The group will run until June 30, 2024, with consideration for longer-term arrangements.

What else is going on in AI this week?

• Australian crowdsourced security startup Bugcrowd raised $156.2 million
• Marqo raised $19.4 million for its proprietary AI-powered vector search engine
• Otter introduced Meeting GenAI, a new set of AI tools for meetings
• An Australian mayor has abandoned what would have been a world-first ChatGPT lawsuit
• Chatbots were making up Superbowl stats.