Weekly AI recap: OpenAI unveils Sora, ChatGPT now has a ‘memory’

On Thursday, OpenAI introduced Sora, an AI model capable of producing stunningly lifelike video from text-based prompts. Clips from some of these videos – which can be up to one minute in length – were published on the company’s website. One shows what appears at first glance to be human actors wearing some kind of astronaut gear and wandering about on a flat, snow-covered tundra. The prompt for the video: “A movie trailer featuring the adventures of the 30-year-old spaceman wearing a red wool knitted motorcycle helmet, blue sky, salt desert, cinematic style, shot on 35mm film, vivid colors.” Another shows two miniature pirate ships engaged in battle atop a cup of coffee.

The name of the model, which translates to ”sky” in Japanese, was chosen because it suggests expansive creative potential, according to a report from The New York Times.

While the quality of the videos produced by Sora is undeniably impressive, OpenAI acknowledges that the model has its shortcomings. “It may struggle with accurately simulating the physics of a complex scene and may not understand specific instances of cause and effect,” the company wrote on its website. “For example, a person might take a bite out of a cookie, but afterward, the cookie may not have a bite mark.”

Sora is not yet available for public use. OpenAI has shared it with “red teamers” – cybersecurity experts who will try to find vulnerabilities in the model – and also with a group of creative professionals in an effort to learn more about the model’s potential risks and benefits.

“We’re sharing our research progress early to start working with and getting feedback from people outside of OpenAI and to give the public a sense of what AI capabilities are on the horizon,” OpenAI writes.

Given the rate at which AI-generated video is improving, these early Sora videos might one day look as awkward and antiquated as that haunting video of the spaghetti-eating Will Smith looks to us now.

OpenAI gives ChatGPT a ‘memory’

For all of you who say ‘thank you’ to ChatGPT any time the chatbot answers your questions in fear of an inevitable AI uprising, now you have one more reason to be polite. (That’s a joke – kinda.)

OpenAI announced in a company blog post on Tuesday that ChatGPT now has the ability to ’remember’ information from previous conversations, thus allowing for more continuous and personalized dialogues with the chatbot. “Remembering things you discuss across all chats saves you from having to repeat information and makes future conversations more helpful,” the company writes.

For example, the blog post imagines that a user had previously told ChatGPT that they have a toddler who loves jellyfish. When that same user later asks for some creative inspiration for their child’s birthday card, the chatbot will remember the earlier conversation and will accordingly provide some ideas for a jellyfish-themed card.

The new memory feature – which can be turned off in settings – was deployed to a small cohort of ChatGPT Free and Plus users this week, according to the blog post. The company is planning for a broader deployment soon.

Users can tell ChatGPT to ’forget’ specific details from previous conversations, and these can also be viewed and erased in settings. “ChatGPT’s memories evolve with your interactions and aren’t linked to specific conversations,” OpenAI writes. “Deleting a chat doesn’t erase its memories; you must delete the memory itself.“

AI gets some Super Bowl cameos

Despite its ubiquity in pop culture these days, AI was not spotlighted in Super Bowl LVIII ads as much as crypto was during the game last year. Still, it did make a few appearances.

Both Microsoft and Google seized the opportunity to promote new AI-powered products – Copilot and the Pixel 8 smartphone, respectively. In different ways, both ads highlighted the potential for AI to enhance preexisting human abilities.

Then there was a commercial for the sports drink brand Bodyarmor that opened with a Hieronymus Bosch-like portrayal of a sports drink ad that was seemingly created, script and all, with the help of AI, and that was reminiscent of the aforementioned video of Will Smith eating spaghetti. (Given how cartoonish this entire section of the ad was, one gets the feeling that the brand intentionally toyed with the inputs to ratchet up its absurdity and hallucination-like quality.)

Verizon also gave a brief nod to AI in its 90-second ad, which starred Beyoncé. As part of her mission to overwhelm the capacity of the wireless network, the singer at one point unveils ’BeyoncéI’ (pronounced ’Beyons-AI’) – depicted as an Ex Machina-like humanoid robot with the face of Queen Bey herself.

Microsoft says black hat hackers have been using OpenAI

Five groups of state-backed hackers have been leveraging OpenAI’s technology to hone their cyberattack capabilities, the tech companies said in a report published Wednesday. The company did not disclose which specific AI tools the hackers were using.

The report claimed that the hackers had ties to the Russian GRU (the country’s national intelligence agency), Iran’s Islamic Revolutionary Guard Corps and the Chinese and North Korean governments, according to a report from Reuters.

In response to its findings, Microsoft has reportedly rolled out a general ban on state-sponsored hackers using OpenAI’s products; the ban does not include Microsoft’s suite of AI-powered products, including Bing.

According to a report from The New York Times, the ways in which these hacker groups were using OpenAI’s technology were more mundane than US military officials may have feared. One group with ties to the Iranian government, for example, was using AI to learn how to avoid antivirus technology and to compose phishing emails.

“Is it providing something new and novel that is accelerating an adversary beyond what a better search engine might? I haven’t seen any evidence of that,” Bob Rotsted, head of cybersecurity at OpenAI, told the Times.

Sam Altman aims to raise trillions for proposed GPU venture

OpenAI CEO Sam Altman is reportedly seeking as much as $7tn from investors – among which is the government of the United Arab Emirates – for a proposed effort to build graphics processing units, or GPUs, the costly semiconductors that have become the backbone of the emerging generative AI industry.

As The Wall Street Journal pointed out in a report last Thursday, $7tn is more than the combined market value of both Microsoft and Apple, the two wealthiest companies in the world. The sum is exorbitantly high compared with most fundraising efforts in the tech world.

Altman has long floated the idea of starting his own GPU venture amid chip scarcity and cost. The market is currently dominated by US chip manufacturer Nvidia, which is credited with inventing the term ’GPU’ and has seen its market value skyrocket in recent years as demand for its chips has steadily increased.

Altman’s ongoing fundraising efforts appear to be a part of his and OpenAI’s broader vision to expand the global GPU supply chain, reducing the strain on the energy grid and inching closer to the achievement of what OpenAI describes on its website as “safe AGI that benefits all of humanity.”

For more on the latest happenings in AI, web3 and other cutting-edge technologies, sign up for The Emerging Tech Briefing newsletter.