Today's AI news: Google launches Gemini 3.8 Live, Meta unveils Muse glasses, AI tutoring matches human tutors, and Sam Altman addresses the UN Security Council.
Here is what happened in AI today — Thursday, 24 September 2026.
Google DeepMind Launches Gemini 3.8 Live with Live Avatar
Google DeepMind has announced Gemini 3.8 Live, a new version of its flagship Gemini model that introduces a Live Avatar feature. The release marks a significant step forward in real-time, multimodal AI interaction. DeepMind released the announcement on 24 September 2026, though full technical details of the model's capabilities are still emerging. This is one of the most notable model releases of the week and signals Google's continued push to keep Gemini competitive at the frontier of conversational and visual AI.
Meta Connect 2026: Muse Glasses, Voice, Video, and More
Meta held its Connect 2026 event, unveiling a range of new AI-powered hardware and software. The headline product is the Muse glasses, which integrate voice and video capabilities directly into wearable eyewear. The announcement reinforces Meta's aggressive push into AI-first consumer hardware and suggests the company is doubling down on the intersection of AI and augmented reality. Latent Space's AINews described the event as showing that Team Zuck is — in their words — absolutely on fire, reflecting the broad enthusiasm around Meta's current product momentum.
Sam Altman Addresses the United Nations Security Council
OpenAI CEO Sam Altman delivered remarks to the United Nations Security Council on 23 September 2026, focusing on three key themes: AI safety, human control over AI systems, and the need for international cooperation on AI governance. The appearance at the UN Security Council — one of the world's most prominent diplomatic forums — underscores how seriously world leaders are beginning to treat AI as a matter of global security and policy. It is rare for a technology company executive to be invited to address the Security Council, marking a notable moment for the broader AI industry.
Landmark Study: AI Tutoring Matches Expert Human Tutoring in GRE Outcomes
A major new study released on arXiv introduces StudentBench, a large-scale evaluation platform for AI teaching. The research involved 2,383 human participants across AI tutoring, human tutoring, and no-tutoring conditions, measuring learning gains on Quantitative and Verbal GRE questions. The headline finding is striking: AI tutoring is statistically equivalent to expert human tutoring for GRE learning gains. In five of seven GRE domains, the best-performing AI tutor actually surpassed the human tutor on average.
Perhaps most remarkable is the cost differential. One AI tutor achieved equivalent learning gains at 918 times lower cost — approximately USD $0.0052 per percentage point gained for AI, compared to USD $4.81 for a human tutor. The study also found that faster AI responses correlated with greater student engagement, more practice, and ultimately larger learning gains. The StudentBench platform is freely available at studentbench.org, enabling ongoing large-scale research into AI-assisted education.
OpenAI Academy Marks Two Years
OpenAI has released a milestone update marking two years of its OpenAI Academy programme, which is focused on bringing AI skills and education to a broader range of communities. The anniversary update signals that OpenAI continues to invest in AI literacy and access as a strategic priority alongside its model development work. While lighter on technical news than the other announcements today, it reflects an important dimension of how the major AI labs are thinking about their role in society — not just building powerful models, but actively working to ensure more people can benefit from and understand them.
The Foundries vs. Navigators Thesis: How AI Is Reshaping Scientific Research
A thought-provoking guest post on Latent Space explores a growing asymmetry in scientific research: thinking has become cheap thanks to AI, but doing — running experiments, acquiring lab equipment, managing physical processes — remains expensive. The piece introduces the framing of Foundries versus Navigators to describe two emerging categories of research organisation adapting to this new reality. As AI continues to lower the cognitive cost of science, this structural shift in how research is organised and funded may prove to be one of the more consequential — if currently underappreciated — trends in the industry.
Frequently Asked Questions
What did Google DeepMind announce on 24 September 2026?
Google DeepMind announced Gemini 3.8 Live, a new version of its Gemini model that introduces a Live Avatar feature for real-time, multimodal AI interaction.
What were the key findings of the StudentBench AI tutoring study?
The StudentBench study, involving 2,383 participants, found that AI tutoring is statistically equivalent to expert human tutoring for GRE learning gains. In five of seven GRE domains, the best AI tutor surpassed the human tutor. One AI tutor achieved equivalent outcomes at 918 times lower cost than a human tutor.
What did Sam Altman say at the United Nations Security Council?
OpenAI CEO Sam Altman addressed the UN Security Council on 23 September 2026, discussing AI safety, the importance of human control over AI systems, and the need for international cooperation on AI governance.
What did Meta announce at Connect 2026?
Meta unveiled the Muse glasses at its Connect 2026 event, featuring integrated voice and video capabilities. The event highlighted Meta's continued push into AI-powered consumer hardware and augmented reality wearables.