OpenAI's new Sora text-to-video generation tool won't be publicly available until later this year, but in the meantime it's serving up some tantalizing glimpses of what it can do – including a mind-bending new video (below) showing what TED Talks might look like in 40 years.
To create the FPV drone-style video, TED Talks worked with OpenAI and the filmmaker Paul Trillo, who's been using Sora since February. The result is an impressive, if slightly bewildering, fly-through of futuristic conference talks, weird laboratories and underwater tunnels.
The video again shows both the incredible potential of OpenAI Sora and its limitations. The FPV drone-style effect has become a popular one for hard-hitting social media videos, but it traditionally requires advanced drone piloting skills and expensive kit that goes way beyond the new DJI Avata 2.
Sora's new video shows that these kind of effects could be opened up to new creators, potentially at a vastly lower cost – although that comes with the caveat that we don't yet know how much OpenAI's new tool itself will cost and who it'll be available to.
What will TED look like in 40 years? For #TED2024, we worked with artist @PaulTrillo and @OpenAI to create this exclusive video using Sora, their unreleased text-to-video model. Stay tuned for more groundbreaking AI — coming soon to https://t.co/YLcO5Ju923! pic.twitter.com/lTHhcUm4FiApril 19, 2024
See more
But the video (above) also shows that Sora is still quite far short of being a reliable tool for full-blown movies. The people in the shots are on-screen for only a couple of seconds and there's plenty of uncanny valley nightmare fuel in the background.
The result is an experience that's exhilarating, while also leaving you feeling strangely off-kilter – like touching down again after a sky dive. Still, I'm definitely keen to see more samples as we hurtle towards Sora's public launch later in 2024.
How was the video made?
(Image credit: OpenAI / TED Talks)
OpenAI and TED Talks didn't go into detail about how this specific video was made, but its creator Paul Trillo recently talked more broadly about his experiences of being one of Sora's alpha tester.
Trillo told Business Insider about the kinds of prompts he uses, including “a cocktail of words that I use to make sure that it feels less like a video game and something more filmic”. Apparently these include prompts like “35 millimeter”, “anamorphic lens”, and “depth of field lens vignette”, which are needed or else Sora will “kind of default to this very digital-looking output”.
Right now, every prompt has to go through OpenAI so it can be run through its strict safeguards around issues like copyright. One of Trillo's most interesting observations is that Sora is currently “like a slot machine where you ask for something, and it jumbles ideas together, and it doesn't have a real physics engine to it”.
This means that it's still a long way way off from being truly consistent with people and object states, something that OpenAI admitted in an earlier blog post. OpenAI said that Sora “currently exhibits numerous limitations as a simulator”, including the fact that “it does not accurately model the physics of many basic interactions, like glass shattering”.
These incoherencies will likely limit Sora to being a short-form video tool for some time, but it's still one I can't wait to try out.
OpenAI's Sora gave us a glimpse earlier this year of how generative AI is going to change video editing – and now Adobe has shown off how that's going to play out by previewing of some fascinating new Premiere Pro tools.
The new AI-powered features, powered by Adobe Firefly, effectively bring the kinds of tricks we've seen from Google's photo-focused Magic Editor – erasing unwanted objects, adding objects and extending scenes – to video. And while it isn't the first piece of software to do that, seeing these tools in an industry standard app that's used by professionals is significant.
For a glimpse of what's coming “this year” to Premiere Pro and other video editing apps, check out the video below. In a new Generative panel, there's a new 'add object' option that lets you type in an object you want to add to the scene. This appears to be for static objects, rather than things like a galloping horse, but it looks handy for b-roll and backgrounds.
Arguably even more helpful is 'object removal', which uses Firefly's AI-based smart masking to help you quickly select an object to remove then make it vanish with a click. Alternatively, you can just combine the two tools to, for example, swap the watch that someone's wearing for a non-branded alternative.
One of the most powerful new AI-powered features in photo editing is extending backgrounds – called Generative Fill in Photoshop – and Premiere Pro will soon have a similar feature for video. Rather than extending the frame's size, Generative Extend will let you add frames to a video to help you, for example, pause on your character's face for a little longer.
While Adobe hasn't given these tools a firm release date, only revealing that they're coming “later this year”, it certainly looks like they'll change Premiere Pro workflows in a several major ways. But the bigger AI video change could be yet to come…
Will Adobe really plug into OpenAI's Sora?
(Image credit: Adobe)
The biggest Premiere Pro announcement, and also the most nebulous one, was Adobe's preview of third-party models for the editing app. In short, Adobe is planning to let you plug generative AI video tools including OpenAI's Sora, Runway and Pika Labs into Premiere Pro to sprinkle your videos with their effects.
In theory, that sounds great. Adobe showed an example of OpenAI's Sora generating b-roll with a text-to-video prompt, and Pika powering Generative Extend. But these “early examples” of Adobe's “research exploration” with its “friends” from the likes of OpenAI are still clouded in uncertainty.
Firstly, Adobe hasn't committed to launching the third-party plug-ins in the same way as its own Firefly-powered tools. That shows it's really only testing the waters with this part of the Premiere Pro preview. Also, the integration sits a little uneasily with Adobe's current stance on generative AI tools.
(Image credit: Adobe)
Adobe has sought to set itself apart from the likes of Midjourney and Stable Diffusion by highlighting that Adobe Firefly is only trained on Adobe Stock image library, which is apparently free of commercial, branded and trademark imagery. “We’re using hundreds of millions of assets, all trained and moderated to have no IP,” Adobe's VP of Generative AI, Alexandru Costin, told us earlier this year.
Yet a new report from Bloomberg claims that Firefly was partially trained on images generated by Midjourney (with Adobe suggesting that could account for 5% of Firefly's training data). And these previews of new alliances with generative video AI models, which are similarly opaque when it comes to their training data, again sits uneasily with Adobe's stance.
Adobe's potential get-out here is Content Credentials, a kind of nutrition label that's also coming to Premiere Pro and will add watermarks to clarify when AI was used in a video and with which model. Whether or not this is enough for Adobe to balance making a commercially-friendly pro video editor with keeping up in the AI race remains to be seen.
OpenAI recently published a music video for the song Worldweight by August Kamp made entirely by their text-to-video engine, Sora. You can check out the whole thing on the company’s official YouTube channel and it’s pretty trippy, to say the least. Worldweight consists of a series of short clips in a wide 8:3 aspect ratio featuring fuzzy shots of various environments.
You see a cloudy day at the beach, a shrine in the middle of a forest, and what looks like pieces of alien technology. The ambient track coupled with the footage results in a uniquely ethereal experience. It’s half pleasant and half unsettling.
It’s unknown what text prompts were used on Sora; Kamp didn’t share that information. But she did explain the inspiration behind them in the description. She states that whenever she created the track, she imagined what a video representing Worldweight would look like. However, she lacked a way to share her thoughts. Thanks to Sora, this is no longer an issue as the footage displays what she had always envisioned. It's “how the song has always ‘looked’” from her perspective.
Embracing Sora
If you pay attention throughout the entire runtime, you’ll notice hallucinations. Leaves turn into fish, bushes materialize out of nowhere, and flowers have cameras instead of petals. But because of the music’s ethereal nature, it all fits together. Nothing feels out of place or nightmare-inducing. If anything, the video embraces the nightmares.
We should mention August Kamp isn’t the only person harnessing Sora for content creation. Media production company Shy Kids recently published a short film on YouTube called “Air Head” which was also made on the AI engine. It plays like a movie trailer about a man who has a balloon for a head.
Analysis: Lofty goals
It's hard to say if Sora will see widespread adoption judging by this content. Granted, things are in the early stages, but ready or not, that hasn't stopped OpenAI from pitching its tech to major Hollywood studios. Studio executives are apparently excited at the prospects of AI saving time and money on production.
August Kamp herself is a proponent of the technology stating, “Being able to build and iterate on cinematic visuals intuitively has opened up categorically new lanes of artistry for me”. She looks forward to seeing “what other forms of storytelling” will appear as artificial intelligence continues to grow.
In our opinion, tools such Sora will most likely enjoy a niche adoption among independent creators. Both Kamp and Shy Kids appear to understand what the generative AI can and cannot do. They embrace the weirdness, using it to great effect in their storytelling. Sora may be great at bringing strange visuals to life, but in terms of making “normal-looking content”, that remains to be seen.
People still talk about how weird or nightmare-inducing content made by generative AI is. Unless OpenAI can surmount this hurdle, Sora may not amount to much beyond niche usage.
It’s still unknown when Sora will be made publicly available. OpenAI is holding off on a launch, citing potential interference in global elections as one of its reasons. Although, there are plans to release the AI by the end of 2024.
OpenAI has been rapidly developing its ChatGPT generative AI chatbot and Sora AI video creator over the last year, and it's now got a new artificial intelligence tool to show off: Voice Generation, which can create synthetic voices from just 15 seconds of audio.
In a blog post (via The Verge), OpenAI says it's been running “a small-scale preview” of Voice Engine, which has been in development since late 2022. It's actually already being used in the Read Aloud feature in the ChatGPT app, which (as the name suggests) reads out answers to you.
Once you've trained the voice from a 15-second sample, you can then get it to read out any text you like, in an “emotive and realistic” way. OpenAI says it could be used for educational purposes, for translating podcasts into new languages, for reaching remote communities, and for supporting people who are non-verbal.
This isn't something everyone can use right now, but you can go and listen to the samples created by Voice Engine. The clips OpenAI has published sound pretty impressive, though there is a slight robotic and stilted edge to them.
Safety first
Voice Engine is already used in ChatGPT’s Read Aloud feature (Image credit: OpenAI)
Worries about misuse are the main reason Voice Engine is only in a limited preview for now: OpenAI says it wants to do more research into how it can protect tools like this from being used to spread misinformation and copy voices without consent.
“We hope to start a dialogue on the responsible deployment of synthetic voices, and how society can adapt to these new capabilities,” says OpenAI. “Based on these conversations and the results of these small scale tests, we will make a more informed decision about whether and how to deploy this technology at scale.”
With major elections due in both the US and UK this year, and generative AI tools getting more advanced all the time, it's a concern across every type of AI content – audio, text, and video – and it's getting increasingly difficult to know what to trust.
As OpenAI itself points out, this has the potential to cause problems with voice authentication measures, and scams where you might not know who you're talking to over the phone, or who's left you a voicemail. These aren't easy issues to solve – but we're going to have to find ways to deal with them.
OpenAI’s Chief Technology Officer Mira Murati recently sat down with The Wall Street Journal to reveal interesting details about their upcoming text-to-video generator Sora.
The interview covers a wide array of topics from the type of content the AI engine will produce to the security measures being put into place. Combating misinformation is a sticking point for the company. Murati states Sora will have multiple safety guardrails to ensure the technology isn’t misused. She says the team wouldn’t feel comfortable releasing something that “might affect global elections”. According to the article, Sora will follow the same prompt policies as Dall-E meaning it’ll refuse to create “images of public figures” such as the President of the United States.
Watermarks are going to be added too. A transparent OpenAI logo can be found in the lower right-hand corner indicating that it's AI footage. Murati adds that they may also adopt content provenance as another indicator. This uses metadata to give information on the origins of digital media. That's all well and good, but it may not be enough. Last year, a group of researchers managed to break “current image watermarking protections”, including those belonging to OpenAI. Hopefully, they come up with something tougher.
Generative features
Things get interesting when they begin to talk about Sora's future. First off, the developers have plans to “eventually” add sound to videos to make them more realistic. Editing tools are on the itinerary as well, giving online creators a way to fix the AI’s many mistakes.
As advanced as Sora is, it makes a lot of errors. One of the prominent examples in the piece revolves around a video prompt asking the engine to generate a video where a robot steals a woman’s camera. Instead, the clip shows the woman partially becoming a robot. Murati admits there is room for improvement stating the AI is “quite good at continuity, [but] it’s not perfect”.
Nudity is not off the table. Murati says OpenAI is working with “artists… to figure out” what kind of nude content will be allowed. It seems the team would be okay with allowing “artistic” nudity while banning things like non-consensual deep fakes. Naturally, OpenAI would like to avoid being the center of a potential controversy although they want their product to be seen as a platform fostering creativity.
Ongoing tests
When asked about the data used to train Sora, Murati was a little evasive.
She started off by claiming she didn’t know what was used to teach the AI other than it was either “publically available or license data”. What’s more, Murati wasn’t sure if videos from YouTube, Facebook, or Instagram were a part of the training. However she later admitted that media from Shutterstock was indeed used. The two companies, if you’re not aware, have a partnership which could explain why Murati was willing to confirm it as a source.
Murati states Sora will “definitely” launch by the end of the year. She didn’t give an exact date although it could happen within the coming months. For now, the developers are safety testing the engine looking for any “vulnerabilities, biases, and other harmful results”.
If you're thinking of one day trying out Sora, we suggest learning how to use editing software. Remember, it makes many errors and might continue to do so at launch. For recommendations, check out TechRadar's best video editing software for 2024.
In case you somehow missed it, Sora is a text-to-video AI meaning you can write a simple request and it’ll compose a video (just as image generation previously worked, but obviously a much more complex endeavor).
(Image credit: OpenAI)
Now OpenAI’s Sora research lead Tim Brooks has released some new content generated by Sora on X (formerly Twitter).
This is Sora’s crack at fulfilling the following request: “Fly through tour of a museum with many paintings and sculptures and beautiful works of art in all styles.”
Pretty impressive to say the least. On top of that, Bill Peebles, also a Sora research lead, showed us a clip generated from the following prompt: “An alien blending in naturally with new york city, paranoia thriller style, 35mm film.”
(Image credit: OpenAI)
Content creator Blaine Brown then stepped in to embellish the above clip, cutting it to repeat the footage and make it longer, while having the alien rapping, complete with lip-syncing. The music is generated by Suno AI by the way (with the lyrics written by Brown, mind), and lip-syncing is done with Pika Labs AI.
This Sora clip is 🔥 when the alien guy busts out in a lip-synced rap about how tough it is being different than everyone else. Workflow in the thread.@suno_ai_ @pika_labs (lip sync)Alienate Yourself 🆙🔊🔊 pic.twitter.com/kc5FI83q5RMarch 3, 2024
See more
Analysis: Still early days for Sora
(Image credit: OpenAI)
It’s worth underlining how fast things seem to be progressing with the capabilities of AI. Image creation powers were one thing – and extremely impressive in themselves – but this is entirely another. Especially when you remember that Sora is still just in testing at OpenAI, with a limited set of ‘red teamers’ (testers hunting out bugs and smoothing over those wrinkles).
The camera work in the museum fly-through flows realistically and feels nicely imaginative in the way it swoops around (albeit with the occasional judder). And the last tweet shows how you can take a base clip and flesh it out with content including AI-generated music.
Of course, AI can write a script as well, and so it begs the question: how long will it be before a blue alien is starring in an AI-generated post-apocalyptic drama. Or an (unintentional) comedy perhaps?
You get the idea, and we’re getting carried away, of course, but still – what AI could be capable of in just a few years is potentially mind-blowing, frankly.
Naturally, we’ll be seeing the cream of the crop of what Sora is capable of in these teasers, and there have been some buggy and weird efforts aired too. (Just as when ChatGPT and other AI chatbots first rolled onto the scene, we saw AI hallucinations and general unhinged behavior and replies).
Perhaps the broader worry with Sora, though, is how this might eventually displace, rather than assist, content creators. But that’s a fear to chew over on another day – not forgetting the potential for misuse with AI-created videos which we recently discussed in more depth here.
ChatGPT maker OpenAI has now unveiled Sora, its artificial intelligence engine for converting text prompts into video. Think Dall-E (also developed by OpenAI), but for movies rather than static images.
It's still very early days for Sora, but the AI model is already generating a lot of buzz on social media, with multiple clips doing the rounds – clips that look as if they've been put together by a team of actors and filmmakers.
Here we'll explain everything you need to know about OpenAI Sora: what it's capable of, how it works, and when you might be able to use it yourself. The era of AI text-prompt filmmaking has now arrived.
OpenAI Sora release date and price
In February 2024, OpenAI Sora was made available to “red teamers” – that's people whose job it is to test the security and stability of a product. OpenAI has also now invited a select number of visual artists, designers, and movie makers to test out the video generation capabilities and provide feedback.
“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,” says OpenAI.
In other words, the rest of us can't use it yet. For the time being there's no indication as to when Sora might become available to the wider public, or how much we'll have to pay to access it.
(Image credit: OpenAI)
We can make some rough guesses about timescale based on what happened with ChatGPT. Before that AI chatbot was released to the public in November 2022, it was preceded by a predecessor called InstructGPT earlier that year. Also, OpenAI's DevDay typically takes place annually in November.
It's certainly possible, then, that Sora could follow a similar pattern and launch to the public at a similar time in 2024. But this is currently just speculation and we'll update this page as soon as we get any clearer indication about a Sora release date.
As for price, we similarly don't have any hints of how much Sora might cost. As a guide, ChatGPT Plus – which offers access to the newest Large Language Models (LLMs) and Dall-E – currently costs $ 20 (about £16 / AU$ 30) per month.
But Sora also demands significantly more compute power than, for example, generating a single image with Dall-E, and the process also takes longer. So it still isn't clear exactly how well Sora, which is effectively a research paper, might convert into an affordable consumer product.
What is OpenAI Sora?
You may well be familiar with generative AI models – such as Google Gemini for text and Dall-E for images – which can produce new content based on vast amounts of training data. If you ask ChatGPT to write you a poem, for example, what you get back will be based on lots and lots of poems that the AI has already absorbed and analyzed.
OpenAI Sora is a similar idea, but for video clips. You give it a text prompt, like “woman walking down a city street at night” or “car driving through a forest” and you get back a video. As with AI image models, you can get very specific when it comes to saying what should be included in the clip and the style of the footage you want to see.
To get a better idea of how this works, check out some of the example videos posted by OpenAI CEO Sam Altman – not long after Sora was unveiled to the world, Altman responded to prompts put forward on social media, returning videos based on text like “a wizard wearing a pointed hat and a blue robe with white stars casting a spell that shoots lightning from his hand and holding an old tome in his other hand”.
How does OpenAI Sora work?
On a simplified level, the technology behind Sora is the same technology that lets you search for pictures of a dog or a cat on the web. Show an AI enough photos of a dog or cat, and it'll be able to spot the same patterns in new images; in the same way, if you train an AI on a million videos of a sunset or a waterfall, it'll be able to generate its own.
Of course there's a lot of complexity underneath that, and OpenAI has provided a deep dive into how its AI model works. It's trained on “internet-scale data” to know what realistic videos look like, first analyzing the clips to know what it's looking at, then learning how to produce its own versions when asked.
So, ask Sora to produce a clip of a fish tank, and it'll come back with an approximation based on all the fish tank videos it's seen. It makes use of what are known as visual patches, smaller building blocks that help the AI to understand what should go where and how different elements of a video should interact and progress, frame by frame.
Sora starts messier, then gets tidier (Image credit: OpenAI)
Sora is based on a diffusion model, where the AI starts with a 'noisy' response and then works towards a 'clean' output through a series of feedback loops and prediction calculations. You can see this in the frames above, where a video of a dog playing in the show turns from nonsensical blobs into something that actually looks realistic.
And like other generative AI models, Sora uses transformer technology (the last T in ChatGPT stands for Transformer). Transformers use a variety of sophisticated data analysis techniques to process heaps of data – they can understand the most important and least important parts of what's being analyzed, and figure out the surrounding context and relationships between these data chunks.
What we don't fully know is where OpenAI found its training data from – it hasn't said which video libraries have been used to power Sora, though we do know it has partnerships with content databases such as Shutterstock. In some cases, you can see the similarities between the training data and the output Sora is producing.
What can you do with OpenAI Sora?
At the moment, Sora is capable of producing HD videos of up to a minute, without any sound attached, from text prompts. If you want to see some examples of what's possible, we've put together a list of 11 mind-blowing Sora shorts for you to take a look at – including fluffy Pixar-style animated characters and astronauts with knitted helmets.
“Sora can generate videos up to a minute long while maintaining visual quality and adherence to the user’s prompt,” says OpenAI, but that's not all. It can also generate videos from still images, fill in missing frames in existing videos, and seamlessly stitch multiple videos together. It can create static images too, or produce endless loops from clips provided to it.
It can even produce simulations of video games such as Minecraft, again based on vast amounts of training data that teach it what a game like Minecraft should look like. We've already seen a demo where Sora is able to control a player in a Minecraft-style environment, while also accurately rendering the surrounding details.
OpenAI does acknowledge some of the limitations of Sora at the moment. The physics don't always make sense, with people disappearing or transforming or blending into other objects. Sora isn't mapping out a scene with individual actors and props, it's making an incredible number of calculations about where pixels should go from frame to frame.
In Sora videos people might move in ways that defy the laws of physics, or details – such as a bite being taken out of a cookie – might not be remembered from one frame to the next. OpenAI is aware of these issues and is working to fix them, and you can check out some of the examples on the OpenAI Sora website to see what we mean.
Despite those bugs, further down the line OpenAI is hoping that Sora could evolve to become a realistic simulator of physical and digital worlds. In the years to come, the Sora tech could be used to generate imaginary virtual worlds for us to explore, or enable us to fully explore real places that are replicated in AI.
How can you use OpenAI Sora?
At the moment, you can't get into Sora without an invite: it seems as though OpenAI is picking out individual creators and testers to help get its video-generated AI model ready for a full public release. How long this preview period is going to last, whether it's months or years, remains to be seen – but OpenAI has previously shown a willingness to move as fast as possible when it comes to its AI projects.
Based on the existing technologies that OpenAI has made public – Dall-E and ChatGPT – it seems likely that Sora will initially be available as a web app. Since its launch ChatGPT has got smarter and added new features, including custom bots, and it's likely that Sora will follow the same path when it launches in full.
Before that happens, OpenAI says it wants to put some safety guardrails in place: you're not going to be able to generate videos showing extreme violence, sexual content, hateful imagery, or celebrity likenesses. There are also plans to combat misinformation by including metadata in Sora videos that indicates they were generated by AI.
OpenAI breaks new ground as the AI giant has revealed its first text-to-video model called Sora, capable of creating shockingly realistic content.
We’ve been wondering when the company was finally going to release its own video engine as so many of its rivals, from Stability AI to Google, have beaten them to the punch. Perhaps OpenAI wanted to get things just right before a proper launch. At this rate, the quality of its outputs could eclipse its contemporaries. According to the official page, Sora can generate “realistic and imaginative scenes” from a single text prompt; much like other text-to-video AI models. The difference with this engine is the technology behind it.
Lifelike content
Open AI claims its artificial intelligence can understand how people and objects “exist in the physical world”. This gives Sora the ability to create scenes featuring multiple people, varying types of movement, facial expressions, textures, and objects with a high amount of detail. Generated videos lack the plastic look or the nightmarish forms seen in other AI content – for the most part, but more on that later.
Sora is also multimodular. Users will reportedly be able to upload a still image to serve as the basis of a video. The content inside the picture will become animated with a lot of attention paid to the small details. It can even take a pre-existing video “and extend it or fill in missing frames.”
Prompt: A litter of golden retriever puppies playing in the snow. Their heads pop out of the snow, covered in. pic.twitter.com/G1qhJRV9tgFebruary 15, 2024
See more
You can find sample clips on OpenAI’s website and on X (the platform formerly known as Twitter). One of our favorites features a group of puppies playing in the snow. If you look closely, you can see their fur and the snow on their snouts have a strikingly lifelike quality to them. Another great clip shows a Victoria-crowned pigeon bobbing around like an actual bird.
A work in progress
As impressive as these two videos may be, Sora is not perfect. OpenAI admits its “model has weaknesses.” It can have a hard time simulating the physics of an object, confuse left from right, as well as misunderstand “instances of cause and effect.” You can have an AI character bite into a cookie, but the cookie lacks a bite mark.
It makes a lot of weird errors too. One of the funnier mishaps involves a group of archeologists unearthing a large piece of paper which then transforms into a chair before ending up as a crumpled piece of plastic. The AI also seems to have trouble with words. “Otter” is misspelled as “Oter” and “Land Rover” is now “Danover”.
even the sora mistakes are mesmerizing pic.twitter.com/OvPSbaa0L9February 15, 2024
See more
Moving forward, the company will be working with its “red teamers” who are a group of industry experts “to assess critical areas for harms or risks.” They want to make sure Sora doesn’t generate false information, hateful content, or have any bias. Additionally, OpenAI is going to implement a text classifier to reject prompts that violate their policy. These include inputs requesting sexual content, violent videos, and celebrity likenesses among other things.
No word on when Sora will officially launch. We reached out for info on the release. This story will be updated at a later time. In the meantime, check out TechRadar's list of the best AI video editors for 2024.
The hits keep on coming at OpenAI. After dismissing CEO and Co-Founder Sam Altman, inviting him back, and reinstating him as CEO just a few days ago, the ChatGPT developer is apparently pulling back, at least temporarily, on its plans for a GPT Store.
Earlier this month during the first OpenAI Dev Day, Altman introduced the concept of “GPTs”, basically custom versions of the ChatGPT generative AI model. These bespoke versions would use custom data and therefore do what you wanted them to do. They'd be less generally smart, like the current ChatGPT that turned 1 this week, and much more specifically smart to your needs. Perhaps the most exciting part of this announcement was that you'd be able to post and buy these custom GPTs in an online GPT Store.
Now, Axios is reporting based on a developer memo it obtained that OpenAI is pressing pause on the GPT Store launch at least until early 2024. In the memo, according to the report, OpenAI wrote, “While we had expected to release it this month, a few unexpected things have been keeping us busy!“
That comment, if real, might be a nod toward the tumult that consumed OpenAI over the last two weeks.
A new beginning
Of course, all that is in the past now. On November 29, Altman posted a message on the company blog praising his team and even holding out an olive branch to former board member and computer scientist Ilya Sutskever who may have promoted the fire drill that prompted Altman's removal.
“I love and respect Ilya, I think he's a guiding light of the field and a gem of a human being. I harbor zero ill will towards him,” wrote Altman on Wednesday.
Perhaps notably, Altman made no mention of “GPTs” or the GPT store in his post. Instead, he focused significant attention on AI safety, writing the company has three immediate priorities and listing this as the first: “Advancing our research plan and further investing in our full-stack safety efforts, which have always been critical to our work.”
There's no indication in the reported memo that OpenAI is pausing GPTs work or really any part of its march toward Artificial General Intelligence (AGI), of which Altman writes, “One of the most important things for the team that builds AGI safely is the ability to handle stressful and uncertain situations, and maintain good judgment throughout.”
Assuming this pause memo is real, the delay to early 2024 is just a matter of a few months. Knowing OpenAI and the rapid development pace of ChatGPT and the large language model (LLM) powering it, the delay could shrink to weeks.
It now seems entirely possible that ChatGPT parent company OpenAI has solved the 'superintelligence' problem, and is now grappling with the implications for humanity.
In the aftermath of OpenAI's firing and rehiring of its co-founder and CEO Sam Altman, revelations about what sparked the move keep coming. A new report in The Information pins at least the internal disruption on a significant Generative AI breakthrough that could lead to the development of something called 'superintelligence' within this decade or sooner.
Superintelligence is, as you might have guessed, intelligence that outstrips humanity, and the development of AI that's capable of such intelligence without proper safeguards is, naturally, a major red flag.
According to The Information, the breakthrough was spearheaded by OpenAI Chief Scientist (and full-of-regrets board member) Ilya Sutskever.
It allows AI to use cleaner and computer-generated data to solve problems the AI has never seen before. This means the AI is trained not on many different versions of the same problem, but on information not directly related to the problem. Solving problems in this way – usually math or science problems – requires reasoning. Right, something we do, not AIs.
OpenAI's primary consumer-facing product, ChatGPT (powered by the GPT large language model [LLM]) may seem so smart that it must to be using reason to craft its responses. Spend enough time with ChatGPT, however, and you soon realize it's just regurgitating what it's learned from the vast swaths of data it's been fed, and making mostly accurate guesses about how to craft sentences that make sense and which apply to your query. There is no reasoning involved here.
The Information claims, though, that this breakthrough – which Altman may have alluded to in a recent conference appearance, saying, “on a personal note, just in the last couple of weeks, I have gotten to be in the room, when we sort of like push the sort of the veil of ignorance back and the frontier of discovery forward,” – sent shockwaves throughout OpenAI.
Managing the threat
While there's no sign of superintelligence in ChatGPT right now, OpenAI is surely working to integrate some of this power into, at least, some of its premium products, like GPT-4 Turbo and those GPTs chatbot agents (and future 'intelligent agents').
Connecting superintelligence to the board's recent actions, which Sutskever initially supported, might be a stretch. The breakthrough reportedly came months ago, and prompted Sutskever and another OpenAI scientist, Jan Leike, to form a new OpenAI research group called Superaligment with the goal of developing superintelligence safeguards.
Yes, you heard that right. The company working on developing superintelligence is simultaneously building tools to protect us from superintelligence. Imagine Doctor Frankenstein equipping the villagers with flamethrowers, and you get the idea.
What's not clear from the report is how internal concerns about the rapid development of superintelligence possibly triggered the Altman firing. Perhaps it doesn't matter.
At this writing, Altman is on his way back to OpenAI, the board is refashioned, and the work to build superintelligence – and to protect us from it – will continue.
If all of this is confusing, I suggest you ask ChatGPT to explain it to you.
1 844.742.7448 
Email Us
Remote Support 
Contact
