Development

Odyssey presents immersive AI-powered streaming

Odyssey, a startup founded by self-driving veterans Oliver Cameron and Jeff Hawke, has unveiled an AI model that allows users to interact with streaming video in real time.

The technology generates video frames every 40 milliseconds, enabling users to move through scenes like a 3D video game instead of passively watching. A demo is currently available online, though it is still in its early stages.

The system relies on a new kind of ‘world model’ that predicts future visual states based on previous actions and environments. Odyssey claims its model can maintain spatial consistency, learn motion from video, and sustain coherent video output for five minutes or more.

Unlike models trained solely on internet data, Odyssey captures real-world environments using a custom 360-degree, backpack-mounted camera to build higher-fidelity simulations.

Tech giants and AI startups are exploring world models to power next-generation simulations and interactive media. Yet creative professionals remain wary. A 2024 study commissioned by the Animation Guild predicted significant job disruptions across film and animation.

Game studios like Activision Blizzard have been scrutinised for using AI while cutting staff.

Odyssey, however, insists its goal is collaboration instead of replacement. The company is also developing software to let creators edit scenes using tools like Unreal Engine and Blender.

Backed by $27 million in funding and supported by Pixar co-founder Ed Catmull, Odyssey aims to transform video content across entertainment, education, and advertising through on-demand interactivity.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

SoftBank and Intel develop low-power memory for AI data centres

SoftBank and Intel are collaborating on a new memory technology designed to halve power consumption compared to current high-bandwidth memory (HBM), which is critical for AI systems.

The project includes the University of Tokyo, research institute Riken, and Shinco Electric Industries, among others, and aims to address cost, efficiency, and supply issues in today’s HBM market.

The initiative will be led by a newly established SoftBank subsidiary, Cy Memory, which is tasked with IP management and developing prototype memory chips that reconfigure wiring structures in stacked DRAMs.

The goal is to mass-produce more energy-efficient and cost-effective memory than existing HBM, with applications in data centres running AI workloads.

SoftBank plans to invest 3 billion yen in Cy Memory, becoming its largest shareholder, and may seek government funding.

As AI data centre demand skyrockets, the new memory could offer Japan a strategic edge in semiconductor development—particularly as domestic DRAM production has largely disappeared. Japan has pledged over 10 trillion yen in semiconductor investments by 2030 to revive its chip industry.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

Google is testing voice-powered Search Live in AI Mode

Google is rolling out a new voice-powered feature called Search Live as part of its evolving AI Mode for Search. Initially previewed at Google I/O 2025, the feature allows users to interact with Search through real-time spoken conversations without needing to type or tap through results.

Available to select users in the United States via the Google app on Android and iOS, Search Live lets users ask questions aloud and receive voice responses. It also supports conversational follow-ups, creating a more natural flow of information.

The feature is powered by Project Astra, Google’s real-time speech processing engine that underpins other innovations like Live in Gemini.

When enabled, a sparkle-styled waveform icon appears under the search bar, replacing the previous Google Lens shortcut. Tapping it opens the feature and activates four voice style options—Cosmo, Neso, Terra, and Cassini. Users can opt for audio responses or mute them and view a transcript instead.

Search Live marks a broader shift in how Google is rethinking search: turning static queries into dynamic dialogues. The company also plans to expand AI Mode with support for live camera feeds shortly, aiming to make Search more immersive and interactive.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

Schools in the EU start adapting to the AI Act

European schools are taking their first concrete steps to integrate AI in line with the EU AI Act, with educators and experts urging a measured, strategic approach to compliance.

At a recent conference on AI in education, school leaders and policymakers explored how to align AI adoption with the incoming regulations.

With key provisions of the EU AI Act already in effect and full enforcement coming by August 2026, the pressure is on schools to ensure their use of AI is transparent, fair, and accountable. The law classifies AI tools by risk level, with those used to evaluate or monitor students subject to stricter oversight.

Matthew Wemyss, author of ‘AI in Education: An EU AI Act Guide,’ laid out a framework for compliance: assess current AI use, scrutinise the impact on students, and demand clear documentation from vendors.

Wemyss stressed that schools remain responsible as deployers, even when using third-party tools, and should appoint governance leads who understand both technical and ethical aspects.

Education consultant Philippa Wraithmell warned schools not to confuse action with strategy. She advocated starting small, prioritising staff confidence, and ensuring every tool aligns with learning goals, data safety, and teacher readiness.

Al Kingsley MBE emphasised the role of strong governance structures and parental transparency, urging school boards to improve their digital literacy to lead effectively.

The conference highlighted a unifying theme: meaningful AI integration in schools requires intentional leadership, community involvement, and long-term planning. With the right mindset, schools can use AI not just to automate, but to enhance learning outcomes responsibly.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

New phase in Malaysia Airlines and Google collaboration

Malaysia Airlines is deepening its partnership with Google to accelerate its digital transformation, focusing on AI-powered marketing and end-to-end innovation across the travel experience.

Building on a successful 2024 collaboration, the renewed agreement will integrate more advanced technologies to enhance customer engagement and streamline services.

The airline and tech giant aims to shift from promotional showcases to a fully integrated ecosystem, using AI and data-driven tools to reshape the travel journey.

A recent highlight was an interactive showcase at the MATTA Fair, where visitors transformed photos into videos using Google’s Veo image-to-video generation tool — part of Malaysia Airlines’ ‘Time For’ campaign.

Dersenish Aresandiran, chief commercial officer of airlines at Malaysia Aviation Group, said the partnership is about more than technology: it’s a strategic move to enrich customer experiences and support national tourism goals ahead of Visit Malaysia Year 2026.

Google Malaysia’s director, Farhan Qureshi, echoed the sentiment, calling the collaboration a powerful example of how innovation and creativity can redefine global travel.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

Broadcom beats estimates but stock dips after-hours

Broadcom reported strong second-quarter earnings and revenue driven by robust AI demand and solid networking performance.

Despite beating expectations and raising its outlook, the stock fell 3.47% in after-hours trading on Thursday — likely due to profit-taking instead of concern about fundamentals. Shares had previously rallied over 75% since April.

Revenue for the quarter ending May 5 reached US$15 billion, up 20% year-on-year. Adjusted earnings per share were US$1.58, exceeding estimates by two cents.

Net income more than doubled to US$4.97 billion. CEO Hock Tan attributed the strength to growing demand for AI infrastructure and contributions from VMware, which Broadcom acquired in late 2023.

Broadcom forecasted Q3 revenue of approximately US$15.8 billion, slightly above analyst expectations. AI-related revenue is set to increase to US$5.1 billion, up from US$4.4 billion in Q2, fuelled by custom AI accelerators and high-speed networking chips used in hyperscale data centres.

Tan said that the trend should continue through fiscal 2026.

Semiconductor solutions brought in US$8.4 billion in Q2, up 17% from last year, while software revenue rose 25% to US$6.6 billion, with VMware as a key contributor.

About 30% of Broadcom’s AI-related revenue now comes from its switching business, reflecting increasing demand for AI chip clusters. Despite the slight dip in share price, analysts continue to view Broadcom as a key player in AI infrastructure.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

Semiconductor innovation drives AI chip boom

The global AI chip design market is set for explosive growth, with its value projected to rise from USD 73.87 billion in 2024 to USD 468.9 billion by 2032.

This rapid expansion, driven by a 25.98% compound annual growth rate, reflects rising demand for AI in everyday devices, cloud computing, and industrial automation.

Surging adoption of AI-powered technologies across sectors—from smartphones and autonomous vehicles to manufacturing and healthcare—is fuelling the need for advanced, energy-efficient chips.

Companies are investing in smaller, faster processors and real-time edge computing solutions, while governments in countries like the US, China, and South Korea are backing local AI chip development through funding and research initiatives.

North America currently leads the market thanks to tech giants like NVIDIA, Google, and Intel, but Asia-Pacific is growing fastest, particularly as China and South Korea accelerate efforts toward self-reliance in AI hardware.

Industry leaders such as Arm, Intel, and Qualcomm are racing to optimise performance while managing power demands, placing AI chip design at the heart of the next digital transformation wave.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

ChatGPT adds meeting recording and cloud access

OpenAI has launched new features for ChatGPT that allow it to record meetings, transcribe conversations, and pull information directly from cloud platforms like Google Drive and SharePoint.

Instead of relying on typed input alone, users can now speak to ChatGPT, which records audio, creates editable summaries, and helps generate follow-up content such as emails or project outlines.

‘Record’ is currently available to Team users via the macOS app and will soon expand to Enterprise and Edu accounts.

The recording tool automatically deletes the audio after transcription and applies existing workspace data rules, ensuring recordings are not used for training.

Instead of leaving notes scattered across different platforms, users gain a structured and searchable history of conversations, voice notes, or brainstorming sessions, which ChatGPT can recall and apply during future interactions.

At the same time, OpenAI has introduced new connectors for business users that let ChatGPT access files from cloud services like Dropbox, OneDrive, Box, and others.

These connectors allow ChatGPT to search and summarise information from internal documents, rather than depending only on web search or user uploads. The update also includes beta support for Deep Research agents that can work with tools like GitHub and HubSpot.

OpenAI has embraced the Model Context Protocol, an open standard allowing organisations to build their own custom connectors for proprietary tools.

Rather than serving purely as a general-purpose chatbot, ChatGPT is evolving into a workplace assistant capable of tapping into and understanding a company’s complete knowledge base.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

M&S CEO targeted by hackers in abusive ransom email

Marks & Spencer has been directly targeted by a ransomware group calling itself DragonForce, which sent a vulgar and abusive ransom email to CEO Stuart Machin using a compromised employee email address.

The message, laced with offensive language and racist terms, demanded that Machin engage via a darknet portal to negotiate payment. It also claimed that the hackers had encrypted the company’s servers and stolen customer data, a claim M&S eventually acknowledged weeks later.

The email, dated 23 April, appears to have been sent from the account of an Indian IT worker employed by Tata Consultancy Services (TCS), a long-standing M&S tech partner.

TCS has denied involvement and stated that its systems were not the source of the breach. M&S has remained silent publicly, neither confirming the full scope of the attack nor disclosing whether a ransom was paid.

The cyber attack has caused major disruption, costing M&S an estimated £300 million and halting online orders for over six weeks.

DragonForce has also claimed responsibility for a simultaneous attack on the Co-op, which left some shelves empty for days. While nothing has yet appeared on DragonForce’s leak site, the group claims it will publish stolen information soon.

Investigators believe DragonForce operates as a ransomware-as-a-service collective, offering tools and platforms to cybercriminals in exchange for a 20% share of any ransom.

Some experts suspect the real perpetrators may be young hackers from the West, linked to a loosely organised online community called Scattered Spider. The UK’s National Crime Agency has confirmed it is focusing on the group as part of its inquiry into the recent retail hacks.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

Gemini 2.5 Pro tops AI coding tests, surpasses ChatGPT and Claude

Google has released an updated version of its Gemini 2.5 Pro model, addressing issues found in earlier updates.

Unlike the I/O Edition, which focuses mostly on coding, the new version improves performance more broadly and is expected to become a stable release in both the Gemini app and web interface.

The company claims the updated model performs significantly better in code generation, topping the Aider Polyglot test with a score of 82.2 percent—surpassing offerings from OpenAI, Anthropic and DeepSeek.

Beyond coding, the model aims to close previous performance gaps introduced with the March 25th update, especially in creativity and response formatting.

Developers can now fine-tune the model’s ‘thinking budget’, while users should notice a more transparent output structure. These changes and consistent improvement in leaderboard ratings on LMArena and WebDevArena suggest that Google is extending its lead in the AI race.

Google continues to rely on blind testing to judge how people feel about its models, and the new Gemini Pro seems to resonate well. In fact, it now answers even quirky test questions with more clarity and confidence—something that had been lacking in earlier versions.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!