The rise of large language models and the question of ownership

The divide defining AI’s future through large language models

What are large language models? Large language models (LLMs) are advanced AI systems that can understand and generate various types of content, including human-like text, images, video, and more audio.

The development of these large language models has reshaped ΑΙ from a specialised field into a social, economic, and political phenomenon. Systems such as GPT, Claude, Gemini, and Llama have become fundamental infrastructures for information processing, creative work, and automation.

Their rapid rise has generated an intense debate about who should control the most powerful linguistic tools ever built.

The distinction between open source and closed source models has become one of the defining divides in contemporary technology that will, undoubtedly, shape our societies.

gemini chatgpt meta AI antitrust trial

Open source models such as Meta’s Llama 3, Mistral, and Falcon offer public access to their code or weights, allowing developers to experiment, improve, and deploy them freely.

Closed source models, exemplified by OpenAI’s GPT series, Anthropic’s Claude, or Google’s Gemini, restrict access, keeping architectures and data proprietary.

Such a tension is not merely technical. It embodies two competing visions of knowledge production. One is oriented toward collective benefit and transparency, and the other toward commercial exclusivity and security of intellectual property.

The core question is whether language models should be treated as a global public good or as privately owned technologies governed by corporate rights. The answer to such a question carries implications for innovation, fairness, safety, and even democratic governance.

Innovation and market power in the AI economy

From an economic perspective, open and closed source models represent opposing approaches to innovation. Open models accelerate experimentation and lower entry barriers for small companies, researchers, and governments that lack access to massive computing resources.

They enable localised applications in diverse languages, sectors, and cultural contexts. Their openness supports decentralised innovation ecosystems similar to what Linux did for operating systems.

Closed models, however, maintain higher levels of quality control and often outperform open ones due to the scale of data and computing power behind them. Companies like OpenAI and Google argue that their proprietary control ensures security, prevents misuse, and finances further research.

The closed model thus creates a self-reinforcing cycle. Access to large datasets and computing leads to better models, which attract more revenue, which in turn funds even larger models.

The outcome of that has been the consolidation of AI power within a handful of corporations. Microsoft, Google, OpenAI, Meta, and a few start-ups have become the new gatekeepers of linguistic intelligence.

OpenAI Microsoft Cloud AI models

Such concentration raises concerns about market dominance, competitive exclusion, and digital dependency. Smaller economies and independent developers risk being relegated to consumers of foreign-made AI products, instead of being active participants in the creation of digital knowledge.

As so, open source LLMs represent a counterweight to Big Tech’s dominance. They allow local innovation and reduce dependency, especially for countries seeking technological sovereignty.

Yet open access also brings new risks, as the same tools that enable democratisation can be exploited for disinformation, deepfakes, or cybercrime.

Ethical and social aspects of openness

The ethical question surrounding LLMs is not limited to who can use them, but also to how they are trained. Closed models often rely on opaque datasets scraped from the internet, including copyrighted material and personal information.

Without transparency, it is impossible to assess whether training data respects privacy, consent, or intellectual property rights. Open source models, by contrast, offer partial visibility into their architecture and data curation processes, enabling community oversight and ethical scrutiny.

However, we have to keep in mind that openness does not automatically ensure fairness. Many open models still depend on large-scale web data that reproduce existing biases, stereotypes, and inequalities.

Open access also increases the risk of malicious content, such as generating hate speech, misinformation, or automated propaganda. The balance between openness and safety has therefore become one of the most delicate ethical frontiers in AI governance.

Socially, open LLMs can empower education, research, and digital participation. They allow low-resource languages to be modelled, minority groups to build culturally aligned systems, and academic researchers to experiment without licensing restrictions.

ai in us education

They represent a vision of AI as a collaborative human project rather than a proprietary service.

Yet they also redistribute responsibility: when anyone can deploy a powerful model, accountability becomes diffuse. The challenge lies in preserving the benefits of openness while establishing shared norms for responsible use.

The legal and intellectual property dilemma

Intellectual property law was not designed for systems that learn from millions of copyrighted works without direct authorisation.

Closed source developers defend their models as transformative works under fair use doctrines, while content creators demand compensation or licensing mechanisms.

3d illustration folder focus tab with word infringement conceptual image copyright law

The dispute has already reached courts, as artists, authors, and media organisations sue AI companies for unauthorised use of their material.

Open source further complicates the picture. When model weights are released freely, the question arises of who holds responsibility for derivative works and whether open access violates existing copyrights.

Some open licences now include clauses prohibiting harmful or unlawful use, blurring the line between openness and control. Legal scholars argue that a new framework is needed to govern machine learning datasets and outputs, one that recognises both the collective nature of data and the individual rights embedded in it.

At stake is not only financial compensation but the broader question of data ownership in the digital age. We need to question ourselves. If data is the raw material of intelligence, should it remain the property of a few corporations or be treated as a shared global resource?

Economic equity and access to computational power

Even the most open model requires massive computational infrastructure to train and run effectively. Access to GPUs, cloud resources, and data pipelines remains concentrated among the same corporations that dominate the closed model ecosystem.

Thus, openness in code does not necessarily translate into openness in practice.

Developing nations, universities, and public institutions often lack the financial and technical means to exploit open models at scale. Such an asymmetry creates a form of digital neo-dependency: the code is public, but the hardware is private.

For AI to function as a genuine global public good, investments in open computing infrastructure, public datasets, and shared research facilities are essential. Initiatives such as the EU’s AI-on-demand platform or the UN’s efforts for inclusive digital development reflect attempts to build such foundations.

3d united nations flag waving wind with modern skyscraper city close up un banner blowing soft smooth silk cloth fabric texture ensign background 1

The economic stakes extend beyond access to infrastructure. LLMs are becoming the backbone of new productivity tools, from customer service bots to automated research assistants.

Whoever controls them will shape the future division of digital labour. Open models could allow local companies to retain more economic value and cultural autonomy, while closed models risk deepening global inequalities.

Governance, regulation, and the search for balance

Governments face a difficult task of regulating a technology that evolves faster than policy. For example, the EU AI Act, US executive orders on trustworthy AI, and China’s generative AI regulations all address questions of transparency, accountability, and safety.

Yet few explicitly differentiate between open and closed models.

The open source community resists excessive regulation, arguing that heavy compliance requirements could suffocate innovation and concentrate power even further in large corporations that can afford legal compliance.

On the other hand, policymakers worry that uncontrolled distribution of powerful models could facilitate malicious use. The emerging consensus suggests that regulation should focus not on the source model itself but on the context of its deployment and the potential harms it may cause.

An additional governance question concerns international cooperation. AI’s global nature demands coordination on safety standards, data sharing, and intellectual property reform.

The absence of such alignment risks a fragmented world where closed models dominate wealthy regions while open ones, potentially less safe, spread elsewhere. Finding equilibrium requires mutual trust and shared principles for responsible innovation.

The cultural and cognitive dimension of openness

Beyond technical and legal debates, the divide between open and closed models reflects competing cultural values. Open source embodies the ideals of transparency, collaboration, and communal ownership of knowledge.

Closed source represents discipline, control, and the pursuit of profit-driven excellence. Both cultures have contributed to technological progress, and both have drawbacks.

From a cognitive perspective, open LLMs can enhance human learning by enabling broader experimentation, while closed ones can limit exploration to predefined interfaces. Yet too much openness may also encourage cognitive offloading, where users rely on AI systems without developing independent judgment.

Ai brain hallucinate

Therefore, societies must cultivate digital literacy alongside technical accessibility, ensuring that AI supports human reasoning rather than replaces it.

The way societies integrate LLMs will influence how people perceive knowledge, authority, and creativity. When language itself becomes a product of machines, questions about authenticity, originality, and intellectual labour take on new meaning.

Whether open or closed, models shape collective understanding of truth, expression, and imagination for our societies.

Toward a hybrid future

The polarisation we are presenting here, between open and closed approaches, may be unsustainable in the long run. A hybrid model is emerging, where partially open architectures coexist with protected components.

Companies like Meta release open weights but restrict commercial use, while others provide APIs for experimentation without revealing the underlying code. Such hybrid frameworks aim to combine accountability with safety and commercial viability with transparency.

The future equilibrium is likely to depend on international collaboration and new institutional models. Public–private partnerships, cooperative licensing, and global research consortia could ensure that LLM development serves both the public interest and corporate sustainability.

A system of layered access (where different levels of openness correspond to specific responsibilities) may become the standard.

google translate ai language model

Ultimately, the choice between open and closed models reflects humanity’s broader negotiation between collective welfare and private gain.

Just as the internet or many other emerging technologies evolved through the tension between openness and commercialisation, the future of language models will be defined by how societies manage the boundary between shared knowledge and proprietary intelligence.

So, in conclusion, the debate between open and closed source LLMs is not merely technical.

As we have already mentioned, it embodies the broader conflict between public good and private control, between the democratisation of intelligence and the concentration of digital power.

Open models promote transparency, innovation, and inclusivity, but pose challenges in terms of safety, legality, and accountability. Closed models offer stability, quality, and economic incentive, yet risk monopolising a transformative resource so crucial in our quest for constant human progression.

Finding equilibrium requires rethinking the governance of knowledge itself. Language models should neither be owned solely by corporations nor be released without responsibility. They should be governed as shared infrastructures of thought, supported by transparent institutions and equitable access to computing power.

Only through such a balance can AI evolve as a force that strengthens, rather than divides, our societies and improves our daily lives.

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

Cloudflare chief warns AI is redefining the internet’s business model

AI is inserting itself between companies and customers, Cloudflare CEO Matthew Prince warned in Toronto. More people ask chatbots before visiting sites, dulling brands’ impact. Even research teams lose revenue as investors lean on AI summaries.

Frontier models devour data, pushing firms to chase exclusive sources. Cloudflare lets publishers block unpaid crawlers to reclaim control and compensation. The bigger question, said Prince, is which business model will rule an AI-mediated internet.

Policy scrutiny focuses on platforms that blend search with AI collection. Prince urged governments to separate Google’s search access from AI crawling to level the field. Countries that enforce a split could attract publishers and researchers seeking predictable rules and payment.

Licensing deals with news outlets, Reddit, and others coexist with scraping disputes and copyright suits. Google says it follows robots.txt, yet testimony indicated AI Overviews can use content blocked by robots.txt for training. Vague norms risk eroding incentives to create high-quality online content.

A practical near-term playbook combines technical and regulatory steps. Publishers should meter or block AI crawlers that do not pay. Policymakers should require transparency, consent, and compensation for high-value datasets, guiding the shift to an AI-mediated web that still rewards creators.

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

Qwen3-Max-Thinking hits perfect scores as Alibaba raises the bar on AI reasoning

Alibaba unveiled Qwen3-Max-Thinking, which scored 100 percent on AIME 2025 and HMMT, matching OpenAI’s top model on reasoning tests. It targets high-precision problem-solving across algebra, number theory, and probability. Researchers regard elite maths contests as strong proxies for reasoning.

Built on Qwen3-Max, a trillion-parameter flagship, the thinking variant emphasises step-by-step solutions. Alibaba says it matches or beats Claude Opus 4, DeepSeek V3.1, Grok 4, and GPT-5 Pro. Positioning stresses accuracy, traceability, and controllable latency.

Signal from a live trading trial added momentum. In a two-week crypto experiment, Qwen3-Max returned 22.3 percent on 10,000 US dollars. Competing systems underperformed, with DeepSeek at 4.9 percent and several US models booking losses.

Access is available via the Qwen web chatbot and Alibaba Cloud APIs. Early adopters can test tool use and stepwise reasoning on technical tasks. Enterprises are exploring finance, research, and operations cases requiring reliability and auditability.

Alibaba researchers say further tuning will broaden task coverage without diluting peak maths performance. Plans include multilingual reasoning, safety alignment, and robustness under distribution shift. Community benchmarks and contests will track progress.

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

Microsoft deal signals pay-per-use path for AI access to People Inc. content

People Inc. has joined Microsoft’s publisher content marketplace in a pay-per-use deal that compensates media for AI access. Copilot will be the first buyer, while People Inc. continues to block most AI crawlers via Cloudflare to force paid licensing.

People Inc., formerly Dotdash Meredith, said Microsoft’s marketplace lets AI firms pay ‘à la carte’ for specific content. The agreement differs from its earlier OpenAI pact, which the company described as more ‘all-you-can-eat’, but the priority remains ‘respected and paid for’ use.

Executives disclosed a sharp fall in Google search referrals: from 54% of traffic two years ago to 24% last quarter, citing AI Overviews. Leadership argues that crawler identification and paid access should become the norm as AI sits between publishers and audiences.

Blocking non-paying bots has ‘brought almost everyone to the table’, People Inc. said, signalling more licences to come. Such an approach by Microsoft is framed as a model for compensating rights-holders while enabling AI tools to use high-quality, authorised material.

IAC reported People Inc. digital revenue up 9% to $269m, with performance marketing and licensing up 38% and 24% respectively. The publisher also acquired Feedfeed, expanding its food vertical reach while pursuing additional AI content partnerships.

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

Salesforce’s Agentforce helps organisations deliver 24/7 support

Organisations across public and private sectors are using Salesforce’s Agentforce to engage people whenever and wherever they need support.

From local governments to hospitals and education platforms, AI systems are transforming how services are delivered and accessed.

In the city of Kyle, Texas, an Agentforce-driven 311 app enables residents to report issues such as potholes or water leaks. The city plans to make the system voice-enabled, reducing traditional call volumes while maintaining a steady flow of service requests and faster responses.

At Pearson, AI enables students to access their online learning platforms instantly, regardless of their time zone. The company stated that the technology fosters loyalty by providing immediate assistance, rather than requiring users to wait for human support.

Meanwhile, UChicago Medicine utilises AI to streamline patient interactions, from prescription refills to scheduling, while ambient listening tools enable doctors to focus entirely on patients rather than typing notes.

Salesforce said Agentforce empowers organisations to save resources while enhancing trust, accessibility, and service quality. By meeting people on their own terms, AI enables more responsive and human-centred interactions across various industries.

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

Study finds AI summaries can flatten understanding compared with reading sources

AI summaries can speed learning, but an extensive study finds they often blunt depth and recall. More than 10,000 participants used chatbots or traditional web search to learn assigned topics. Those relying on chatbot digests showed shallower knowledge and offered fewer concrete facts afterwards.

Researchers from Wharton and New Mexico State conducted seven experiments across various tasks, including gardening, health, and scam awareness. Some groups saw the same facts, either as an AI digest or as source links. Advice written after AI use was shorter, less factual, and more similar across users.

Follow-up raters judged AI-derived advice as less informative and less trustworthy. Participants who used AI also reported spending less time with sources. Lower effort during synthesis reduces the mental work that cements understanding.

Findings land amid broader concerns about summary reliability. A BBC-led investigation recently found that major chatbots frequently misrepresented news content in their responses. The evidence suggests that to serves as support for critical reading, rather than a substitute for it.

The practical takeaway for learners and teachers is straightforward. Use AI to scaffold questions, outline queries, and compare viewpoints. Build lasting understanding by reading multiple sources, checking citations, and writing your own synthesis before asking a model to refine it.

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

UK teachers rethink assignments as AI reshapes classroom practice

Nearly eight in ten UK secondary teachers say AI has forced a rethink of how assignments are set, a British Council survey finds. Many now design tasks either to deter AI use or to harness it constructively in lessons. Findings reflect rapid cultural and technological shifts across schools.

Approaches are splitting along two paths. Over a third of designers create AI-resistant tasks, while nearly six in ten purposefully integrate AI tools. Younger staff are most likely to adapt; yet, strong majorities across all age groups report changes to their practices.

Perceived impacts remain mixed. Six in ten worry about their communication skills, with some citing narrower vocabulary and weaker writing and comprehension skills. Similar shares report improvements in listening, pronunciation, and confidence, suggesting benefits for speech-focused learning.

Language norms are evolving with digital culture. Most UK teachers now look up slang and online expressions, from ‘rizz’ to ‘delulu’ to ‘six, seven’. Staff are adapting lesson design while seeking guidance and training that keeps pace with students’ online lives.

Long-term views diverge. Some believe AI could lift outcomes, while others remain unconvinced and prefer guardrails to limit misuse. British Council leaders say support should focus on practical classroom integration, teacher development, and clear standards that strike a balance between innovation and academic integrity.

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

Mustafa Suleyman warns against building seemingly conscious AI

Mustafa Suleyman, CEO of Microsoft AI, argues that AI should be built for people, not to replace them. Growing belief in chatbot consciousness risks campaigns for AI rights and a needless struggle over personhood that distracts from human welfare.

Debates over true consciousness miss the urgent issue of convincing imitation. Seemingly conscious AI may speak fluently, recall interactions, claim experiences, and set goals that appear to exhibit agency. Capabilities are close, and the social effects will be real regardless of metaphysics.

People already form attachments to chatbots and seek meaning in conversations. Reports of dependency and talk of ‘AI psychosis‘ show persuasive systems can nudge vulnerable users. Extending moral status to uncertainty, Suleyman argues, would amplify delusions and dilute existing rights.

Norms and design principles are needed across the industry. Products should include engineered interruptions that break the illusion, clear statements of nonhuman status, and guardrails for responsible ‘personalities’. Microsoft AI is exploring approaches that promote offline connection and healthy use.

A positive vision keeps AI empowering without faking inner life. Companions should organise tasks, aid learning, and support collaboration while remaining transparently artificial. The focus remains on safeguarding humans, animals, and the natural world, not on granting rights to persuasive simulations.

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

Live exploitation of CVE-2024-1086 across older Linux versions flagged by CISA

CISA’s warning serves as a reminder that ransomware is not confined to Windows. A Linux kernel flaw, CVE-2024-1086, is being exploited in real-world incidents, and federal networks face a November 20 patch-or-disable deadline. Businesses should read it as their cue, too.

Attackers who reach a vulnerable host can escalate privileges to root, bypass defences, and deploy malware. Many older kernels remain in circulation even though upstream fixes were shipped in January 2024, creating a soft target when paired with phishing and lateral movement.

Practical steps matter more than labels. Patch affected kernels where possible, isolate any components that cannot be updated, and verify the running versions against vendor advisories and the NIST catalogue. Treat emergency changes as production work, with change logs and checks.

Resilience buys time when updates lag. Enforce least privilege, require MFA for admin entry points, and segment crown-jewel services. Tune EDR to spot privilege-escalation behaviour and suspicious modules, then rehearse restores from offline, immutable backups.

Security habits shape outcomes as much as CVEs. Teams that patch quickly, validate fixes, and document closure shrink the blast radius. Teams that defer kernel maintenance invite repeat visits, turning a known bug into an avoidable outage.

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

Microsoft AI chief rules out machine consciousness as purely biological phenomenon

Microsoft’s AI head, Mustafa Suleyman, has dismissed the idea that AI could ever become conscious, arguing that consciousness is a property exclusive to biological beings.

Speaking at the AfroTech Conference in Houston, Suleyman said researchers should stop exploring the notion of sentient AI, calling it ‘the wrong question’.

He explained that while AI can simulate experience, it cannot feel pain or possess subjective awareness.

Suleyman compared AI’s output to a narrative illusion rather than genuine consciousness, aligning with the philosophical theory of biological naturalism, which ties awareness to living brain processes.

Suleyman has become one of the industry’s most outspoken critics of conscious AI research. His book ‘The Coming Wave’ and his recent essay ‘We must build AI for people;’ not to be a person warn against anthropomorphising machines.

He also confirmed that Microsoft will not develop erotic chatbots, a direction that has been embraced by competitors such as OpenAI and xAI.

He stressed that Microsoft’s AI systems are designed to serve humans, not mimic them. The company’s Copilot assistant now includes a ‘real talk’ mode that challenges users’ assumptions instead of offering flattery.

Suleyman said responsible development must avoid ‘unbridled accelerationism’, adding that fear and scepticism are essential for navigating AI’s rapid evolution.

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