It incorporates the solutions that won gold at the International Mathematical Olympiad and beat the ICPC coding contest, and carries on its duties doing parallel reasoning, letting it try several approaches to a problem at once.
Continue reading “Google’s Gemini 3 Deep Think debuts for Ultra users”
This has now caused Altman to delay other initiatives, such as ads, to focus on making a better ChatGPT, paywalled The Information writes, citing an internal memo.
They are apparently planning to release a new reasoning model next week that will be «ahead of Gemini 3.» But this needs a little more polishing on the «experience.»
Just last week, a developer revealed ad code in the latest ChatGPT beta — meaning that their work on ads was fairly advanced and almost ready to ship.
This work is now on the back burner, at least until next week, when OpenAI hopes to reclaim their crown.
Read more: The Information (paywalled), and Reuters. Discussion on r/Singularity.
This comes as almost every other Big Tech company is increasing data center spending and Google has set aside $93 billion in «capital expenditures» this year to do the same.
This will be followed by a «significant increase» in 2026, but likely not matching OpenAI’s enormous $1.4 trillion data center spending.
They are aiming to «spend a lot,» and hit «the next 1000x in 4-5 years,» Vahdat is reported to have said.
That’s one thousand times more «capability, compute and storage networking» that he aims to add for «essentially the same cost, power and energy spend.»
Read more: The scoop at CNBC, writeups at Ars Technica and Gizmodo.
— Like the generations before it, Gemini 3 is once again advancing the state of the art, says CEO Sundar Pichai on their launch page, and adds: — In this new chapter, we’ll continue to push the frontiers of intelligence, agents, and personalization to make AI truly helpful for everyone.
Debuting in preview across all of Google’s services, including AI Mode on their front page, the new model is «another big step on the path toward AGI,» Google says.
Continue reading “Google launches Gemini 3, blows out benchmarks and is less sycophantic”
The old SIMA was good at following instructions, but the second version now has access to Gemini models and can explore 3D worlds on its own, with zero advance training.
That’s great for video games, where it can think and perform complex reasoning around its goals.
Learning from concepts
It can also learn across games, taking cues from «mining» in one game and transferring it to «harvesting» in another, meaning it can iterate and get better over time.
Continue reading “Google DeepMind unveils SIMA 2 — a self-improving 3D navigator agent”
You can now ask the app to remove sunglasses in photos or fix a smile, but it can also respond to names you have tagged in your pictures, such as «make Engel smile.»
«Help me edit»
You can use the «Help me edit»-button in the editor and simply describe the style you want your pics to be in, from a Renaissance portrait to a picture from a children’s story book.
Most of Google’s AI features are handled on-device, but they are seeing the need for more computing power to move «from completing simple requests to AI that can anticipate your needs with tailored suggestions or handle tasks for you at just the right moment.»
The connections between the device and servers are encrypted, and the data transmitted is not available to anyone’s prying eyes — not even Google’s.
The new tech is not getting a wide rollout, and most AI queries will still be on-device. The only feature to use it is the Magic Cue in Android, and the Recorder app, that will be able to summarize transcripts in «a wide range of languages.»
This is foundational technology for Google, and they will be rolling out features across their services in short order, saying «This is just the beginning.»
Read more: Google’s launch page, writeups on The Verge and 9to5Google.
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.
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!
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”
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”
In a sitdown/interview with Salesforce CEO Marc Benioff at the Dreamforce 2025 conference, that goes long on the future of AI, cloud and innovations, he let this qoute rip:
— We kickstarted Gemini, we brought Google brain and Google deep mind together, and we’ve been rapidly iterating since then, he tells Benioff, then adds some forward looking comments:
Continue reading “Sundar Pichai: Gemini 3.0 is going to be released «this year»”
The IOAA is a top rated exam for global high school students and is held annually with some 300 participants from 64 countries, and consists of questions to demonstrate deep conceptual understanding, multimodal analysis and multi-step derivations.
Sora hits 1 million downloads in five days
In what may be the fastest rush to the milestone, OpenAI’s Sora chief, Bill Peebles, just tweeted that they matched it faster than even ChatGPT. This is despite being limited by invites and only being available in North America, he points out. On the other hand, the copyright «slop» shows no signs of abating. California’s Creative Artists Agency, one of the largest around, are now warning of «significant risk» from the app, and saying creators deserve to be paid for their intellectual property.
More at: CNBC, TechCrunch and Engadget.
OpenAI delivers ChatGPT Go in slew of countries
After launching in India in August, and later expanding to Indonesia, the cost-effective new ChatGPT tier now becomes available in 18, mostly Asian, countries. The subscription offers slightly more access and should give 10x more messages, image generations than the Free tier and cost around $4.5 in local currencies.
More at: Nick Turley’s launch tweet, TechCrunch.
Read on for more!
The CodeMender model is based on Google’s own research into finding zero-day vulnerabilities in computer programs, where they found lots of exploits, and stipulated that humans would struggle to keep up with AI scanners once implemented widely.
Continue reading “Google launches Computer Use model, CodeMender and AI Mode expansion”