Capacity development

Meta’s Behemoth AI model faces setback

Meta Platforms has postponed the release of its flagship AI model, known as ‘Behemoth,’ due to internal concerns about its performance, according to a report by the Wall Street Journal.

Instead of launching as planned, engineers are struggling to deliver improvements that would meaningfully advance the model beyond earlier versions.

Behemoth was originally scheduled for release in April to coincide with Meta’s first AI developer conference but was quietly delayed to June. The latest update suggests the launch has now been pushed to autumn or later, as internal doubts grow over whether it is ready for public deployment.

In April, Meta previewed Behemoth under the Llama 4 line, calling it ‘one of the smartest LLMs in the world’ and positioning it as a teaching model for future AI systems. Instead of Behemoth, Meta released Llama 4 Scout and Llama 4 Maverick as the latest iterations in its AI portfolio.

The delay comes amid intense competition in the generative AI space, where rivals like Google, OpenAI, and Anthropic continue advancing their models. Meta appears to be opting for caution instead of rushing an underwhelming product to market.

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

DeepMind unveils AlphaEvolve for scientific breakthroughs

Google DeepMind has unveiled AlphaEvolve, a new AI system designed to help solve complex scientific and mathematical problems by improving how algorithms are developed.

Rather than acting like a standard chatbot, AlphaEvolve blends large language models from the Gemini family with an evolutionary approach, enabling it to generate, assess, and refine multiple solutions at once.

Instead of relying on a single output, AlphaEvolve allows researchers to submit a problem and potential directions. The system then uses both Gemini Flash and Gemini Pro to create various solutions, which are automatically evaluated.

The best results are selected and enhanced through an iterative process, improving accuracy and reducing hallucinations—a common issue with AI-generated content.

Unlike earlier DeepMind tools such as AlphaFold, which focused on narrow domains, AlphaEvolve is a general-purpose AI for coding and algorithmic tasks.

It has already shown its value by optimising Google’s own Borg data centre management system, delivering a 0.7% efficiency gain—significant given Google’s global scale.

The AI also devised a new method for multiplying complex matrices, outperforming a decades-old technique and even beating DeepMind’s specialised AlphaTensor model.

AlphaEvolve has also contributed to improvements in Google’s hardware design by optimising Verilog code for upcoming Tensor chips.

Though not publicly available yet due to its complexity, AlphaEvolve’s evaluation-based framework could eventually be adapted for smaller AI tools used by researchers elsewhere.

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

Amazon launches end-to-end AI audiobook production for publishers

Amazon has announced the expansion of AI-generated audiobook narration through its Audible platform, offering publishers the option to convert books into audio using synthetic voices.

The company’s new tools include more than 100 AI-generated voices across English, Spanish, French, and Italian, with variations in accents and dialects. Publishers will be able to choose between two production workflows: a fully managed Audible-led process or a self-service option using the same underlying technology.

According to Amazon, the initiative aims to scale audiobook production and increase accessibility. Audible CEO Bob Carrigan said the company’s long-term goal is to support ‘every book in every language’ through its end-to-end AI production system.

Later in 2024, Amazon will begin beta testing AI-powered translation services for audiobooks.

The services will include text-to-text translation of manuscripts—with the option of either AI or professional narration—and speech-to-speech translation designed to preserve the original narrator’s voice and tone across languages.

Human linguists will be available to review translations for accuracy and cultural context. Amazon did not indicate whether the new AI tools would replace human narrators but positioned the offering as an additional option for publishers.

The expansion comes as generative AI continues to reshape content production workflows across the publishing and media industries.

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

Apple partners with BCI developers to expand accessibility

Apple has announced a new set of accessibility features arriving later this year, including a breakthrough initiative that allows users to control their devices using brain signals. The company also plans to streamline its Personal Voice tool and expand accessibility support across its platforms.

In partnership with Brain Computer Interface (BCI) developers like Synchron, Apple is working to enable users to navigate iPhone, iPad, Mac, and Vision Pro using neural input.

Synchron’s device, called the Stentrode, is a stent-like implant placed in a vein near the brain’s motor cortex via a minimally invasive procedure. It reads brain signals and translates them into commands for digital interfaces.

The Wall Street Journal reports that Apple is building system-level support for BCIs in iOS 19 and visionOS 3, including a new protocol compatible with Switch Control.

This will allow users to operate Apple devices without physical interaction, an innovation aimed at supporting individuals with conditions such as amyotrophic lateral sclerosis (ALS).

One early tester of the technology, Mark Jackson, who has ALS, described using his Stentrode in conjunction with Apple’s Vision Pro headset to virtually explore environments like the Swiss Alps.

While Jackson noted limitations in cursor control and speed, the collaboration marks a significant step forward in assistive technology.

In addition to brain control support, iOS 19 will enhance Apple’s Personal Voice feature, originally introduced in iOS 17. This feature enables users at risk of losing their speech to create a personalised synthetic voice.

Previously, the tool required users to record 150 phrases and process them overnight. In iOS 19, the process has been reduced to just 10 phrases, with the model processed in under a minute. Apple says the updated output will sound smoother and more natural.

Other upcoming accessibility features include a Magnifier app for Mac and App Store Nutrition Labels to help users assess the accessibility of third-party apps.

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

Amazon to invest in Saudi AI Zone

Amazon has announced a new partnership with Humain, an AI company launched by Saudi Arabia’s Crown Prince Mohammed bin Salman, to invest over $5 billion in creating an ‘AI Zone’ in the kingdom.

The project will feature Amazon Web Services (AWS) infrastructure, including servers, networks, and training programmes, while Humain will develop AI tools using AWS and support Saudi startups with access to resources.

A move like this adds Amazon to a growing list of tech firms—such as Nvidia and AMD—that are working with Humain, which is backed by Saudi Arabia’s Public Investment Fund. American companies like Google and Salesforce have also recently turned to the PIF for funding and AI collaborations.

Under a new initiative supported by former US President Donald Trump, US tech firms can now pursue deals with Saudi-based partners more freely.

Instead of relying on foreign data centres, Saudi Arabia has required AI providers to store data locally, prompting companies like Google, Oracle, and now Amazon to expand operations within the region.

Amazon has already committed $5.3 billion to build an AWS region in Saudi Arabia by 2026, and says the AI Zone partnership is a separate, additional investment.

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

TikTok unveils AI video feature

TikTok has launched ‘AI Alive,’ its first image-to-video feature that allows users to transform static photos into animated short videos within TikTok Stories.

Accessible only through the Story Camera, the tool applies AI-driven movement and effects—like shifting skies, drifting clouds, or expressive animations—to bring photos to life.

Unlike text-to-image tools found on Instagram and Snapchat, TikTok’s latest feature takes visual storytelling further by enabling full video generation from single images. Although Snapchat plans to introduce a similar function, TikTok has moved ahead with this innovation.

All AI Alive videos will carry an AI-generated label and include C2PA metadata to ensure transparency, even when shared beyond the platform.

TikTok emphasises safety, noting that every AI Alive video undergoes several moderation checks before it appears to creators.

Uploaded photos, prompts, and generated videos are reviewed to prevent rule-breaking content. Users can report violations, and final safety reviews are conducted before public sharing.

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

Harvey adds Google and Anthropic AI

Harvey, the fast-growing legal AI startup backed early by the OpenAI Startup Fund, is now embracing foundation models from Google and Anthropic instead of relying solely on OpenAI’s.

In a recent blog post, the company said it would expand its AI model options after internal benchmarks showed that different tools excel at different legal tasks.

The shift marks a notable win for OpenAI’s competitors, even though Harvey insists it’s not abandoning OpenAI. Its in-house benchmark, BigLaw, revealed that several non-OpenAI models now outperform Harvey’s original system on specific legal functions.

For instance, Google’s Gemini 2.5 Pro performs well at legal drafting, while OpenAI’s o3 and Anthropic’s Claude 3.7 Sonnet are better suited for complex pre-trial work.

Instead of building its own models, Harvey now aims to fine-tune top-tier offerings from multiple vendors, including through Amazon’s cloud. The company also plans to launch a public legal benchmark leaderboard, combining expert legal reviews with technical metrics.

While OpenAI remains a close partner and investor, Harvey’s broader strategy signals growing competition in the race to serve the legal industry with AI.

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

iOS 18.5: Satellite SOS, Screen Time alerts, and bug fixes

Apple has released iOS 18.5, bringing its life-saving satellite emergency features to iPhone 13 models for the first time. Previously available only on iPhone 14 and newer, the feature allows users to connect with emergency services via satellite when cellular or Wi-Fi networks are unavailable.

The update expands access to satellite services provided by mobile carriers, including those like T-Mobile working with Starlink. iPhone 13 users can check for availability by visiting the Cellular menu in Settings.

The satellite feature has already been credited with multiple life-saving interventions, including rescuing hikers, wildfire victims, and others in remote areas. With this update, a wider group of users can now benefit from the added layer of safety.

Alongside the satellite expansion, iOS 18.5 introduces several smaller but notable features. Screen Time now alerts parents if a child successfully guesses the parental passcode to override restrictions. The Mail app has been updated with a dedicated ‘All Mail’ tab for easier navigation.

Other changes include a new Pride wallpaper, a simplified method for subscribing to Apple TV content on smart TVs, and a fix for a bug that caused Apple Vision Pro to launch with a black screen. The update also resolves issues with Siri, enterprise app performance, and other minor bugs.

iOS 18.5 launched alongside updates for iPadOS (18.5), watchOS (11.5), visionOS (2.5), and security patches for macOS Ventura and Sonoma.

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

Click To Do and Settings agent bring AI to Windows 11 beta

Microsoft has rolled out Windows 11 Insider Preview Build 26120.3964 to the Beta Channel, marking the official start of the 24H2 version. Available to Insider users starting this week, the update delivers key AI-driven enhancements—most notably, a new agent built into the Settings app and upgraded text actions.

The AI agent in Settings allows users to interact using natural language instead of simple keywords. Microsoft says users can ask questions like ‘how to control my PC by voice’ or ‘my mouse pointer is too small’ to receive personalised help navigating and adjusting system settings.

Initially, the feature is limited to Copilot+ PCs powered by Snapdragon processors and set to English as the primary language. Microsoft plans to expand support to AMD and Intel devices in the near future.

The update also introduces a new FAQs section on the About page under Settings > System. The company says this addition will help users better understand their device’s configuration, performance, and compatibility.

Microsoft is also enhancing its ‘Click To Do’ feature. On Copilot+ PCs with AMD or Intel chips, users can now highlight text (10 words or more) and press Win + Click or Win + Q to access quick AI actions like Summarise, Rewrite, or Create a bulleted list.

These tools are powered by Phi Silica, an on-device small language model. The features require the system language to be English and the user to be signed in with a Microsoft account.

Microsoft notes that Rewrite is temporarily unavailable for users with French or Spanish as their default language but will return in a future update.

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

SoftBank profit jumps on AI-driven rebound

SoftBank Group reported a 124% surge in quarterly profit, driven by booming AI demand that lifted chip sales and startup valuations. Net income reached ¥517.18 billion ($3.5 billion) in the fiscal fourth quarter, with the Vision Fund swinging back to a profit of ¥26.1 billion.

The results provide momentum for SoftBank’s ambitions to invest heavily in OpenAI and US-based AI infrastructure. Plans include a $30 billion stake in OpenAI and leading a $100 billion push into data centres under the Stargate project, which could eventually grow to $500 billion.

However, investor caution amid tariffs and tech protectionism has delayed detailed financing discussions. Despite these hurdles, SoftBank’s chip unit Arm Holdings has benefited from rising global AI investments, even as near-term forecasts remain mixed.

For the full year, SoftBank earned ¥1.15 trillion, reversing a significant loss from the previous year. The company continues to navigate risks tied to the volatile tech start-up market, especially as Vision Fund portfolio firms go public in India.

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