AI: OpenAI trims AI Applications down to 'AGI' basics. RTZ #1037

AI: OpenAI trims AI Applications down to 'AGI' basics. RTZ #1037

OpenAI founder/CEO Sam Altman is choosing to do more with less.

Fewer consumer focused AI application endeavors. Especially ones like the two year old text to video, briefly viral, Sora App, with high AI Compute requirements relative to subscription and/or ad revenues coming in the door.

Also on the cutting room floor ‘indefinitely’, are ‘erotic chatbots’ planned for ChatGPT later this year. An earlier aspiration to emulate Elon Musk’s Grok/xAI, which is still offering ‘spicy’ chatbots. FT reports OpenAI’s “Decision follows staff and investor concerns about sexual AI content”. Likely a good move following this week’s verdicts on a couple of cases around child safety online. Where Meta is in the regulatory cross-hairs.

OpenAI is shifting focus for its core LLM AI models and underlying AI Infrastructure, to applications that tilt more to the enterprise than consumer. Even while its organization has bulked up of late with hundreds of Meta people who know how to build ad sales atop core consumer products. All ahead of a mega-IPO race with Anthropic this year.

Led by an ex-Meta exec Fidji Simo, who was hired as OpenAI’s ‘CEO of Applications’. Now retitled to head ‘AGI Deployment’. Let’s unpack.

The Information lays it out well in “OpenAI CEO Shifts Responsibilities, Preps ‘Spud’ AI Model”:

“The company is narrowing CEO Altman’s direct reports and winding down Sora.”

OpenAI CEO Sam Altman has relinquished direct oversight of the company’s safety and security teams so he can focus on raising capital, supply chains and “building datacenters at unprecedented scale,” he told staff on Tuesday.”

And they’re certainly raising more capital, as CFO Sarah Friar just topped up an additional $10 billion from investors to bring current record fundraise pre-yearend IPO to “north of $120 billion”.

Back to Sam Altman’s fewer priorties:

“At the same time, he said the company had completed the initial development of its next major AI model, codenamed Spud, and would wind down the Sora AI video mobile app, which employees had complained was a drag on the company’s computing resources during a time of heightened competition with foes such as Anthropic and Google. The shutdown of Sora prompted Walt Disney Co. to abandon a planned $1 billion investment in OpenAI related to a Sora licensing agreement struck last December, according to a person familiar with the situation.”

Sora, a technically brilliant AI app that had caught so much global attention and peer competition, including the partnership with Disney now being wound down.

A victim of too much AI compute resources needed for too little revenues. As Implicator AI summarizes in a separate piece, OpenAI Killed Its Best Demo. Strategy Is Subtraction”:

“The arithmetic speaks for itself. Industry estimates put Sora’s compute costs at roughly $15 million per day, about $5 billion per year, for a consumer app that generated less than $500,000 a month in revenue. Across its entire lifetime, Sora earned about $2.1 million in in-app purchases, according to Appfigures. Polling told the rest of the story. About two in three Americans disapproved of AI-generated video. Nearly three-quarters said they would rather not watch fully synthetic creative content. The audience had already voted.”

“The math required no brilliance. It required a CFO.”

Never mind that competitors like Meta with its Vibes, Google with its ‘nano banana’ leveraging its lower-cost, ubiquitous, vertical TPU AI compute tech stack, and others are still in the AI text to video app game. For now.

Back to the Information and Sam’s hard choices:

“The moves are the latest by the company, recently valued at $730 billion, to narrow its focus to boost its coding and services aimed at business customers. Inside OpenAI, shutting down Sora was widely viewed as a likely first step in the process of culling such side quests.”

A continuing ‘Code Red’, as I summarized this weekend in topic #2.

“Altman didn’t specify what he meant by supply chains, but the company has been developing its own custom AI server chips and discussing deals for its own data centers and power sources, though it recently shifted focus to rent more cloud servers instead.”

And some additional re-assignments of responsibilities to other senior people at OpenAI:

“Altman moved the safety team under OpenAI’s research organization led by Chief Research Officer Mark Chen, and its security team will move to the “scaling” organization led by cofounder and president Greg Brockman.”

And a shift to OpenAI’s version of ‘AGI’, however defined out there.

“Altman also said that the company would be renaming senior executive Fidji Simo’s product organization to “AGI Deployment,” a reference to artificial general intelligence, or AI that’s roughly on par with humans. Last week, the company told employees that it would create a desktop “superapp” by combining ChatGPT with its coding agent Codex and browser Atlas.”

All to focus on pre-IPO focus on Anthropic, which is hurtling down its Enterprise AI ramp threatening to become the Coke vs remaining the Pepsi in this sibling IPO race.

“The changes come as the company competes with Anthropic for both customers and investor attention as both head toward expected blockbuster IPOs in the next year. Anthropic has been taking the business world by storm with its AI agents for coding and other white collar tasks, and Altman has publicly said OpenAI is racing to launch similar tools.”

And of course, the company’s next major iteration of its core GPT/ChatGPT LLM AIs:

“The success of its next model will help determine how well those tools perform. The company has finished pretraining “Spud,” Altman said in the memo. He told staff that the company expects to have a “very strong model” in “a few weeks” that the team believes “can really accelerate the economy.”

“He added: “Things are moving faster than many of us expected.”

Releasing additional, precious AI token compute resources from Nvidia and others.

“OpenAI’s plan to wind down its Sora mobile app and application programming interface will free up computational resources to run Spud, Altman told staff. It also shelved its original plan to ship video capabilities in ChatGPT. That’s a change from the company’s earlier plans to integrate its AI-generated video capabilities in ChatGPT.”

And the crack Sora team is being re-purposed towards AI World Models, an area I’ve highlighted as being important beyond core LLM AIs. Leading to robotics, self driving cars and more.

“Shortly after Altman’s announcement, Bill Peebles, OpenAI’s head of Sora, said in a Slack post, “Incredibly proud of what the small and extremely cracked Sora product team has accomplished; the app brought a lot of joy to the world, and it let us battle-harden Sora infrastructure by doing deployments at this scale.”

“He added that Sora research would now focus on “systems that deeply understand the world by learning to simulate arbitrary environments at high fidelity,” otherwise known as world models, and the “prize is automating the physical economy.” Altman similarly said in the memo that the Sora research team will “prioritize longer-term world simulation research especially as it pertains to robotics.”

The whole piece is worth a full read for additional detail and nuance.

Overall, it’s a shift to the enterprise specifically to head off a fast ramping Anthropic. Especially in a mega-IPO race in the second half of this year. Also an eye to Elon Musk’s June SpaceX/xAI $1.25+ trillion dollar IPO, already potentially rushed to file with the SEC next week.

All of it setting up a very busy AI Tech Wave, where the leading AI company tries to do far more with a bit less. Stay tuned.

(NOTE: The discussions here are for information purposes only, and not meant as investment advice at any time. Thanks for joining us here)





Want the latest?

Sign up for Michael Parekh's Newsletter below:


Subscribe Here