Author: Tor Fosheim

OpenAI estimated to lose daily $15 million on Sora videos

Video is more expensive than text, and OpenAI is currently bleeding money on free Sora 2 generations.
Sora 2 generations are bleeding money for OpenAI, and they say they will reduce the number of free «gens.» (Picture: generated)
There are no official figures on just how much the free Sora service costs, but Forbes has done the math.

It works out by putting a price of $1.30 per video from GPU costs, multiplying it to an estimated 4.5 million users and then presuming that about 25% on average generate ten videos per day.

$15 million per day works out to a cool $5.4 billion annual bill.

That’s a lot of money, and Bill Peebles, head of Sora, recently said that «eventually we will need to bring the free gens down to accommodate growth (we won’t have enough gpus to do it otherwise!), but we’ll be transparent as it happens.»

Read the full story on Forbes

Rumor roundup for Monday, November 10.

(picture: Adobe)
Not much to publish today. There are, however, lots of rumors on the grapevine:

There’s talk about Google deprecating models like Gemini 2.5 Pro and its whole family by the end of November to make way for the upcoming Gemini 3.0, said by CEO Sundar Pichai to arrive this year. Many say a preview is launching soon.

There is even a purported leak of Nano Bana 2, the next generation of Gemini’s image generator, and a lot of people are blown out of the water by it.

Some people are also claiming to have spotted GPT-5.1 in OpenAI source code, which could mean we are getting closer to release of that model, too.

Then rumor has it that the new Kimi 2 model from China, said to have been developed for a paltry 4.6 million dollars, is going head to head with GPT-5 in a benchmark collection. It’s made by a company called Moonshot that is backed by Alibaba.

None of this is confirmed news, though, and I’ll be closely watching for anything official, which might be coming sooner than we think.

US Army to purchase one million drones in the next three years

Drones might well be the future of war, and the Pentagon is late to the party.
The USA seems to finally have woken up about drone warfare, but it needs Congress aboard for investments. (Picture: Adobe)
After «learning lessons» from the war in Ukraine, U.S. Army Secretary Daniel Driscoll said they will up its drone game, according to Reuters.

The target is to produce millions of drones in the next three years, and to build «robust supply lines» to be able to increase production «to as however much we need.»

There is a bill going through Congress for a megacenter in Texas to produce a million drones, but Driscoll doesn’t want to rely on a single company.

— We want to partner with other drone manufacturers who are using them for Amazon deliveries and all the different use cases, he says.

Meanwhile, the Replicator initiative and Valkyrie fighter drones have spent years in limbo waiting for Congressional approval.

Read more: this is from a Reuters exclusive. See also Defense News.

Meta commits to «at least» $600 billion spend on AI infrastructure

Meta commits to spending big its AI infrastructure investments in the next three years.
Meta will be a big AI infrastructure player if their investments pan out. (Picture: Adobe)
The outlay is only for next three years, according to Reuters, who says Meta has been doubling down on AI recently, with massive outlays for their Superintelligence Lab and data center buildouts.

They are building out data centers because «it’s the right strategy to aggressively front-load capacity so we’re prepared for the most optimistic cases,» according to CEO Mark Zuckerberg.

Meta saw an 83% drop in its operating income the last quarter, and plans for capital expenditures of $70 billion just this year, but will have increase it threefold to reach their target by 2028.

Read more: Reuters, The Register and Business Insider.

Weekend roundup: Gemini is going to space, Apple chooses Google, and Amazon’s had it with Perplexity

(Picture: generated)

Gemini is launching to orbit
Google’s latest moonshot might almost be literal. They are preparing for sending their TPU processors into low-earth orbit, and maybe then build a proper AI data center in space — where there is ample sunlight to provide it with energy. They have already tested a TPU in orbit conditions in a particle accelerator and it survived, and the next step is the launch of two prototype satellites in early 2027. They call it Project Suncatcher, and say that «in the future, space may be the best place to scale AI compute.»
More at: Sundar Pichai’s tweet, Google’s announcement blog

Google close to Apple deal for AI Siri
Apparently, Apple has chosen Gemini for its upcoming AI version of the Siri assistant. They will use what is likely a custom version of the model with 1.2 trillion parameters, running on Apple’s Private Cloud Compute servers. Apple supposedly also tested options from OpenAI and Anthropic, but Anthropic’s fees were too high and Apple already partners with Google for search results. The deal will cost Apple $1 billion a year, far less than the $20 billion Google pays Apple to be their search provider.
More at: Bloomberg, MacRumors, TechCrunch.

Read on for more!

Continue reading “Weekend roundup: Gemini is going to space, Apple chooses Google, and Amazon’s had it with Perplexity”

Altman: There is a «chance» GPT-6 will deliver real scientific discovery

GPT-6 will get massive levels of compute, and should be able to do scientific discovery, Altman said.
GPT-6 will be paired with historic levels of computing power, with OpenAI having invested in excess of 40 GW of capacity. (Picture: Generated)
In a recent podcast called Conversations with Tyler, he also said we are getting close to AI running entire corporations.

— GPT-5 is the first moment where you see a glimmer of AI doing new science. It’s very tiny things, but here and there someone’s posting like, “Oh, it figured this thing out,” or “Oh, it came up with this new idea,” or “Oh, it was a useful collaborator on this paper,” Altman said.

Continue reading “Altman: There is a «chance» GPT-6 will deliver real scientific discovery”

OpenAI says policy on health not changed, ChatGPT still answering

You can still ask ChatGPT about health, just not presonal advice.
OpenAI clarified its policy, but it was not a new one, they say. (Picture: Adobe)
According to several social media posts about an updated OpenAI guidance this weekend, it could seem that professional advice had been banned on the platform. It has not, according to OpenAI themselves.

The policy wasn’t new, nor a change, but a consolidation of several different ones, and led to some confusion, it would seem.

OpenAI’s head of health, Karan Singhal, denies any changes entirely, saying ChatGPT will «continue to be a great resource for health information:»

Continue reading “OpenAI says policy on health not changed, ChatGPT still answering”

OpenAI partners with AWS for $38 billion of compute power

OpenAI are free to negotiate deals with Microsoft's rivals, but is still on tap to buy $240 billion of compute from them.
Microsoft no longer has dibs on OpenAI compute, and AWS are moving in to stifle their hunger. (Picture: generated)
After removing Microsofts «first right of refusal» in their Public Benefit makeover last week, OpenAI is already busy making deals with Microsofts competitors.

The new seven-year agreement doesn’t put a gigawatt number on the compute OpenAI is buying, but it costs roughly $40B to build out a 1 GW data center.

Continue reading “OpenAI partners with AWS for $38 billion of compute power”

Google to start experimenting with ads in AI content

Robby Stein, of Google Search doesn't see ads going away, and lots of opportunity for them in Google's AI products.
Robby Stein, VP of Product for Google Search comments on advertising in Google’s AI. (Picture: screenshot).
In a wide-ranging interview on the podcast Silicon Valley Girl, Robby Stein, VP of Product for Google Search is positive about how advertising could get even more granular with all the extra information people can provide in their AI products, saying;

Ads not going away
— I don’t see them [ads] going away. The way people are using Google Search isn’t really changing, what is happening is that it’s expanding [with AI services], he opines.

Continue reading “Google to start experimenting with ads in AI content”

OpenAI shuts down pipeline of professional advice on ChatGPT [updated]

Reddit used to be democratizing tool for expert professional advice, but now it's all over. OpenAI lawyered up.
You can no longer use ChatGPT as your personal doctor, as it defies the EU AI Act and FDA guidance, according to OpenAI. (Picture: Adobe)
UPDATE: OpenAI says there are no changes, simply a consolidation of several usage policies that might have lead to confusion.

—

OpenAI has updated ChatGPT’s usage policies of October 29, banning a vast swath of content where it was arguably the most useful — as in interpreting medical imagery and helping with medical diagnosis, and offering legal or financial advice.

The idea is to stop ChatGPT (and any other OpenAI model) from giving advice that could be interpreted as professional, fiduciary, or legally binding guidance, as required by the EU AI Act and American FDA guidance.

Continue reading “OpenAI shuts down pipeline of professional advice on ChatGPT [updated]”

Weekend roundup; expanded Sora, security research and the battle for India

For a limited time, Sora is available without invite codes for select countries, but the 30 generations per day limit may have to go.
No more invite codes for select countries in Sora 2, and bevy of new features. (Picture: generated)

Sora 2 expands, is now available without invite codes
Following the massive success of the Sora 2 video generator, OpenAI is opening up the service for those without invite codes in the USA, Canada, Japan and Korea «for a limited time.» Simultaneously, they are announcing reusable characters that can feature in more than one video and an easier way to stitch videos together. If that wasn’t enough, OpenAI is adding more video generations for power users hitting the 30-per-day generation limits and letting them pay for more gens. They are also musing about letting rightsholders get compensation for the reuse of their characters, as a means of getting paid for your work on the platform. They do warn that 30 gens needs too many GPUs and will be throttled at some stage.
More at: MacRumors and a Twitter announcement, list of available countries.

OpenAI reveals security research agent in beta
The new agent, Aardvark, will look through code repositories at scale almost like a human would, and find errors and exploits before the bad guys do. It will continually analyze your source code and find vulnerabilities. The agent has already been used to find «numerous» vulnerabilities in open source software, and OpenAI will provide pro bono scanning to «select, non-commercial» OSS systems. Aardvark is not being widely released, existing instead as a private beta inside OpenAI’s offices, kind of like Google’s CodeMender.
More at OpenAI’s announcement and ZDNet.

Read on for more!

Continue reading “Weekend roundup; expanded Sora, security research and the battle for India”

Big Tech doubles down on even bigger AI spend

Big tech will spend about a quarter of a trillion dollars on "capital expenditures" this year, according to quarterly reports.
Data center deals are flourishing and none of the big tech spenders feel they can afford not be in the race. (Picture: Adobe)
Quarterly results are in for Microsoft, Alphabet and Meta — and while the numbers are mixed, they all agree on big capital expenditures — needed for building data centers — for fiscal year 2025.

The coming AI wars will be fought with data centers and gigawatts, and nobody wants to lose out.

Continue reading “Big Tech doubles down on even bigger AI spend”

OpenAI completes transition to a public benefit corporation

OpenAI will still be controlled by a non-profit, but is now easier to invest in and might go public sometime in the future.
The AI lab is now open to investments, but will have a purpose baked into its corporate structure. (Picture: Adobe)
The AI lab is now a less regulated $500 billion public benefit corporation, controlled by a non-profit arm with about $130 billion in equity.

The new governance makes it possible for them to reach for a market debut at some time in the future, and unlocks the investment of some $30 billion from SoftBank, which was contingent on the regulatory change.

Continue reading “OpenAI completes transition to a public benefit corporation”

ChatGPT reports millions of troubling mental health conversations

OpenAI reports high numbers of mental health issues and says they are mitigating the issue with an update.
Once you look at actual numbers, the number of mental health issues with ChatGPT is astonishingly high. (Picture: generated)
OpenAI says remedies are being taken against three categories of troubling interactions with ChatGPT, and readies expert psychologist responses to them.

The categories are «psychosis, mania or other severe mental health symptoms,» which consists of around 0.07% of conversations — or 550,000 people in actual numbers.

Second is «self-harm and suicide,» ticking in at 0.15% of ChatGPT users — or 1.2 million in actual numbers.

And the third is «emotional reliance on AI,» accounting for 0.15% of users in a given week, also 1.2 million in real numbers.

Continue reading “ChatGPT reports millions of troubling mental health conversations”

OpenAI said to be working on a music generation tool

OpenAI is said to be making an AI music generator, but results seem far off.
Music generation is on a tear these days, with people hardly noticing any difference from the hand made stuff. (Picture: Adobe)
The Information (paywalled) is reporting that OpenAI is well underway to create a music generator.

It could be used to generate music from a text or audio prompt — or make a guitar background to a vocal track, or add music to videos.

Recruiting at The Juilliard School
Apparently, the AI lab has partnered with students from The Juilliard School of music to annotate scores that could be used to train the generator.

Continue reading “OpenAI said to be working on a music generation tool”