Digital cooperation

New findings reveal untrained AI can mirror human brain responses

Researchers at Johns Hopkins report that brain-inspired AI architectures can display human-like neural activity before any training. Structural design may provide stronger starting points than data-heavy methods. The findings challenge long-held views about how machine intelligence forms.

Researchers tested modified transformers, fully connected networks, and convolutional networks across multiple variants. They compared untrained model responses with neural data from humans and primates viewing identical images. The approach allowed a direct measure of architectural influence.

Transformers and fully connected networks showed limited change when scaled. Convolutional models, by contrast, produced patterns that aligned more closely with human brain activity. Architecture appears to be a decisive factor early in development.

Untrained convolutional models matched aspects of systems trained on millions of images. The results suggest brain-like structures could cut reliance on vast datasets and energy-intensive computation. The implications may reshape how advanced models are engineered.

Further research will examine simple, biologically inspired learning rules. The team plans to integrate these mechanisms into future AI frameworks. The goal is to combine architecture and biology to accelerate meaningful advances.

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

YouTube criticises Australia’s new youth social-media restrictions

Australia’s forthcoming ban on social media accounts for users under 16 has prompted intense criticism from YouTube, which argues that the new law will undermine existing child safety measures.

The report notes that from 10 December, young users will be logged out of their accounts and barred from posting or uploading content, though they will still be able to watch videos without signing in.

YouTube said the policy will remove key parental-control tools, such as content filters, channel blocking and well-being reminders, which only function for logged-in accounts.

Rachel Lord, Google and YouTube public-policy lead for Australia, described the measure as ‘rushed regulation’ and warned the changes could make children ‘less safe’ by stripping away long-established protections.

Communications Minister Anika Wells rejected this criticism as ‘outright weird’, arguing that if YouTube believes its own platform is unsafe for young users, it must address that problem itself.

The debate comes as Australia’s eSafety Commissioner investigates other youth-focused apps such as Lemon8 and Yope, which have seen a surge in downloads ahead of the ban.

Regulators reversed YouTube’s earlier exemption in July after identifying it as the platform where 10- to 15-year-olds most frequently encountered harmful content.

Under the new Social Media Minimum Age Act, companies must deactivate underage accounts, prevent new sign-ups and halt any technical workarounds or face penalties of up to A$49.5m.

Officials say the measure responds to concerns about the impact of algorithms, notifications and constant connectivity on Gen Alpha. Wells said the law aims to reduce the ‘dopamine drip’ that keeps young users hooked to their feeds, calling it a necessary step to shield children from relentless online pressures.

YouTube has reportedly considered challenging its inclusion in the ban, but has not confirmed whether it will take legal action.

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

Governments urged to build learning systems for the AI era

Governments are facing increased pressure to govern AI effectively, prompting calls for continuous institutional learning. Researchers argue that the public sector must develop adaptive capacity to keep pace with rapid technological change.

Past digital reforms often stalled because administrations focused on minor upgrades rather than redesigning core services. Slow adaptation now carries greater risks, as AI transforms decisions, systems and expectations across government.

Experts emphasise the need for a learning infrastructure that facilitates to reliable flow of knowledge across institutions. Singapore and the UAE have already invested heavily in large-scale capability-building programmes.

Public servants require stronger technical and institutional literacy, supported through ongoing training and open collaboration with research communities. Advocates say that states that embed learning deeply will govern AI more effectively and maintain public trust.

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

Japan plans large scale investment to boost AI capability

Japan plans to increase generative AI usage to 80 percent as officials push national adoption. Current uptake remains far lower than in the United States and China.

The government intends to raise early usage to 50 percent and stimulate private investment. A trillion yen target highlights the efforts to expand infrastructure and accelerate deployment across various Japanese sectors quickly.

Guidelines stress risk reduction and stronger oversight through an enhanced AI Safety Institute. Critics argue that measures lack detail and fail to address misuse with sufficient clarity.

Authorities expect broader AI use in health care, finance and agriculture through coordinated public-private work. Annual updates will monitor progress as Japan seeks to enhance its competitiveness and strategic capabilities.

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

Mistral AI unveils new open models with broader capabilities

Yesterday, Mistral AI introduced Mistral 3 as a new generation of open multimodal and multilingual models that aim to support developers and enterprises through broader access and improved efficiency.

The company presented both small dense models and a new mixture-of-experts system called Mistral Large 3, offering open-weight releases to encourage wider adoption across different sectors.

Developers are encouraged to build on models in compressed formats that reduce deployment costs, rather than relying on heavier, closed solutions.

The organisation highlighted that Large 3 was trained with extensive resources on NVIDIA hardware to improve performance in multilingual communication, image understanding and general instruction tasks.

Mistral AI underlined its cooperation with NVIDIA, Red Hat and vLLM to deliver faster inference and easier deployment, providing optimised support for data centres along with options suited for edge computing.

A partnership that introduced lower-precision execution and improved kernels to increase throughput for frontier-scale workloads.

Attention was also given to the Ministral 3 series, which includes models designed for local or edge settings in three sizes. Each version supports image understanding and multilingual tasks, with instruction and reasoning variants that aim to strike a balance between accuracy and cost efficiency.

Moreover, the company stated that these models produce fewer tokens in real-world use cases, rather than generating unnecessarily long outputs, a choice that aims to reduce operational burdens for enterprises.

Mistral AI continued by noting that all releases will be available through major platforms and cloud partners, offering both standard and custom training services. Organisations that require specialised performance are invited to adapt the models to domain-specific needs under the Apache 2.0 licence.

The company emphasised a long-term commitment to open development and encouraged developers to explore and customise the models to support new applications across different industries.

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

NVIDIA platform lifts leading MoE models

Frontier developers are adopting a mixture-of-experts architecture as the foundation for their most advanced open-source models. Designers now rely on specialised experts that activate only when needed instead of forcing every parameter to work on each token.

Major models, such as DeepSeek-R1, Kimi K2 Thinking, and Mistral Large 3, rise to the top of the Artificial Analysis leaderboard by utilising this pattern to combine greater capability with lower computational strain.

Scaling the architecture has always been the main obstacle. Expert parallelism requires high-speed memory access and near-instant communication between multiple GPUs, yet traditional systems often create bottlenecks that slow down training and inference.

NVIDIA has shifted toward extreme hardware and software codesign to remove those constraints.

The GB200 NVL72 rack-scale system links seventy-two Blackwell GPUs via fast shared memory and a dense NVLink fabric, enabling experts to exchange information rapidly, rather than relying on slower network layers.

Model developers report significant improvements once they deploy MoE designs on NVL72. Performance leaps of up to ten times have been recorded for frontier systems, improving latency, energy efficiency and the overall cost of running large-scale inference.

Cloud providers integrate the platform to support customers in building agentic workflows and multimodal systems that route tasks between specialised components, rather than duplicating full models for each purpose.

Industry adoption signals a shift toward a future where efficiency and intelligence evolve together. MoE has become the preferred architecture for state-of-the-art reasoning, and NVL72 offers a practical route for enterprises seeking predictable performance gains.

NVIDIA positions its roadmap, including the forthcoming Vera Rubin architecture, as the next step in expanding the scale and capability of frontier AI.

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

AWS launches frontier agents to boost software development

AWS has launched frontier agents, autonomous AI tools that extend software development teams. The first three – Kiro, AWS Security Agent, and AWS DevOps Agent – enhance development, security, and operations while working independently for extended periods.

Kiro functions as a virtual developer, maintaining context, learning from feedback, and managing tasks across multiple repositories. AWS Security Agent automates code reviews, penetration testing, and enforces organisational security standards.

AWS DevOps Agent identifies root causes of incidents, reduces alerts, and provides proactive recommendations to improve system reliability.

These agents operate autonomously, scale across multiple tasks, and free teams from repetitive work, allowing focus on high-priority projects. Early users, including SmugMug and Commonwealth Bank of Australia, report quicker development, stronger security, and more efficient operations.

By integrating frontier agents into the software development lifecycle, AWS is shifting AI from task assistance to completing complex projects independently, marking a significant step forward in what AI can achieve for development teams.

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

Honolulu in the US pushes for transparency in government AI use

Growing pressure from Honolulu residents in the US is prompting city leaders to consider stricter safeguards surrounding the use of AI. Calls for greater transparency have intensified as AI has quietly become part of everyday government operations.

Several city departments already rely on automated systems for tasks such as building-plan screening, customer service support and internal administrative work. Advocates now want voters to decide whether the charter should require a public registry of AI tools, human appeal rights and routine audits.

Concerns have deepened after the police department began testing AI-assisted report-writing software without broad consultation. Supporters of reform argue that stronger oversight is crucial to maintain public trust, especially if AI starts influencing high-stakes decisions that impact residents’ lives.

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

OpenAI faced questions after ChatGPT surfaced app prompts for paid users

ChatGPT users complained after the system surfaced an unexpected Peloton suggestion during an unrelated conversation. The prompt appeared for a Pro Plan subscriber and triggered questions about ad-like behaviour. Many asked why paid chats were showing promotional-style links.

OpenAI said the prompt was part of early app-discovery tests, not advertising. Staff acknowledged that the suggestion was irrelevant to the query. They said the system is still being adjusted to avoid confusing or misplaced prompts.

Users reported other recommendations, including music apps that contradicted their stated preferences. The lack of an option to turn off these suggestions fuelled irritation. Paid subscribers warned that such prompts undermine the service’s reliability.

OpenAI described the feature as a step toward integrating apps directly into conversations. The aim is to surface tools when genuinely helpful. Early trials, however, have demonstrated gaps between intended relevance and actual outcomes.

The tests remain limited to selected regions and are not active in parts of Europe. Critics argue intrusive prompts risk pushing users to competitors. OpenAI said refinements will continue to ensure suggestions feel helpful, not promotional.

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

Regulators question transparency after Mixpanel data leak

Mixpanel is facing criticism after disclosing a security incident with minimal detail, providing only a brief note before the US Thanksgiving weekend. Analysts say the timing and lack of clarity set a poor example for transparency in breach reporting.

OpenAI later confirmed its own exposure, stating that analytics data linked to developer activity had been obtained from Mixpanel’s systems. It stressed that ChatGPT users were not affected and that it had halted its use of the service following the incident.

OpenAI said the stolen information included names, email addresses, coarse location data and browser details, raising concerns about phishing risks. It noted that no advertising identifiers were involved, limiting broader cross-platform tracking.

Security experts say the breach highlights long-standing concerns about analytics companies that collect detailed behavioural and device data across thousands of apps. Mixpanel’s session-replay tools can be sensitive, as they can inadvertently capture private information.

Regulators argue the case shows why analytics providers have become prime targets for attackers. They say that more transparent disclosure from Mixpanel is needed to assess the scale of exposure and the potential impact on companies and end-users.

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

Diplo chatbot can make mistakes. Consider checking important information.