Summarized articles

Geoffrey Hinton, AI pioneer and Nobel laureate, meets for lunch in Toronto, reflecting on his legacy and growing concerns about artificial intelligence.
He helped lay the foundations for modern AI through neural networks, now powering tools like ChatGPT.
Once a champion of AI’s potential, Hinton has become a vocal critic, warning of existential risks.
He fears AI could enable ordinary people to build bioweapons or even nuclear devices.
Hinton believes AI is intelligent by any definition and will inevitably become smarter.
He advocates for a “mother-child” relationship with AI—benevolent, protective, and nurturing.
Despite his achievements, he remains humble, often saying “I don’t know” during deep discussions.
He contrasts his academic path with tech leaders like Altman and Zuckerberg, noting their capitalist motivations.
Hinton criticizes Western governments for failing to regulate AI, praising China’s engineering-led approach.
He sold his company to Google in 2013 to fund care for his neurodiverse son, later retiring in 2023.
Though excited by AI’s potential in healthcare and education, he warns of mass unemployment and inequality.
He uses ChatGPT for research and daily tasks, even citing its role in a recent breakup.
Hinton sees AI as a powerful tool, but worries about cognitive offloading and loss of human creativity.
He’s skeptical of utopian visions, believing capitalism will exploit AI for profit.
He argues universal basic income won’t solve the dignity crisis caused by job loss.
Hinton reflects on mortality and the unknown future of humanity in an AI-driven world.
He believes we’re at a historical inflection point—AI could be amazingly good or devastatingly bad.
He urges society to prepare seriously, not optimistically, for what’s coming.
His message: survival may depend on how we shape our relationship with intelligent machines.
And he leaves us with a chilling analogy: choosing between tech leaders may be like choosing between being shot or poisoned.

Source : FT

AI companies like OpenAI, Google, Microsoft, and Perplexity are betting big on shopping as a key use case for AI agents.

These agents can search for products, navigate websites, and even complete purchases on behalf of users.

This shift is disrupting traditional ecommerce, as consumers increasingly rely on chatbots instead of visiting brand websites.

Advertisers are adapting by optimizing content for AI visibility, using keyword-rich URLs and authoritative site mentions.

Startups like Profound and Refine help brands monitor their presence in chatbot responses.

AI agents are becoming the primary interface between consumers and online shopping platforms.

Google reports that nearly 60% of searches in Europe now end without a click, thanks to AI-generated overviews.

Gartner predicts a 25% drop in traditional search volume by next year due to generative AI.

OpenAI’s Agent system can shop autonomously, adding items to carts and handing off to users for checkout.

Perplexity’s Comet and Microsoft’s Action features also enable cross-platform shopping tasks.

Google’s AI tools personalize shopping experiences, reducing the need for multiple tabs and complex research.

Brands are rethinking SEO strategies to ensure visibility in AI-generated results.

Fast-loading websites and specific product descriptions are now critical for AI prioritization.

Semantic search is rising, with users asking for broader concepts like “wedding clothes in the south of France.”

Text-based advertising is proving more effective than image-heavy formats for chatbot visibility.

Transactions are shifting from brand websites to chatbot interfaces, challenging traditional ecommerce models.

Inrupt, co-founded by Tim Berners-Lee, aims to give users control over personal data shared with AI agents.

Experts warn that AI agents may limit consumer choice by selecting products rather than showing full options.

The ecommerce landscape is rapidly evolving, with AI agents reshaping how, where, and why people shop.

Brands must adapt quickly or risk losing visibility and relevance in this new AI-driven marketplace.

This September 2025, two major global gatherings—the Shanghai Cooperation Organization Plus summit and the United Nations General Assembly—highlight the starkly contrasting visions of world order. Rather than one model triumphing, the future is likely to be shaped by a complex blend of old and new governance structures.

China, hosting the SCO Plus meeting, positioned itself as a beacon of stability, with Xi Jinping using the platform to showcase military strength and diplomatic reach. In contrast, the United States, under the Trump administration, appears increasingly unpredictable, retreating from multilateral institutions and shifting its strategic focus toward the Indo-Pacific.

The UN General Assembly’s 80th session is overshadowed by global outrage over the war in Gaza and the US’s suspension of visas for Palestinians. Fractures within the Western alliance are deepening, as countries like the UK, France, and Canada call for a two-state solution, aligning with Global South nations supporting South Africa’s genocide case against Israel.

Meanwhile, global governance is faltering in other areas too—arms control treaties are fragmenting, and new threats like space militarization lack regulation. Amid this uncertainty, China unveiled its Global Governance Initiative, pledging support for the UN while promoting alternatives like a development bank for SCO members.

This signals a shift: China is blending participation in established institutions with efforts to reshape them. As the post-1945 and post-Cold War US-led order fades, the world is entering a new era of multipolarity. Navigating this transition will require both innovation and a commitment to preserving the most effective elements of existing systems.

The AI Race for Data Centres: Energy, Regulation, and Trade-offs

The rapid growth of artificial intelligence is fueling an unprecedented demand for data centres. These facilities are the backbone of AI training and inference, but they come with immense energy requirements. A modern hyperscale data centre can consume as much power as a medium-sized city, with cooling systems alone accounting for 30–40% of total electricity use. As rack densities rise to support AI workloads, cooling loads grow even faster.

The Energy Imperative

Electricity is the single largest operating cost for data centres. AI accelerators such as GPUs and TPUs are power-hungry, and their waste heat must be removed to avoid hardware failure. This drives operators to secure lower-cost, stable energy resources. While renewables are the preferred option, supply is increasingly constrained: building new solar and wind capacity requires land, permitting, and grid integration that lag far behind AI’s exponential energy demand.

Policy and Regulation

Governments are being lobbied to loosen restrictions on large power infrastructure and to accelerate permitting. Yet, policymakers are caught between enabling digital growth and meeting climate goals. Across Europe, “renewables crackdowns” — a combination of stricter siting rules, moratoriums in grid-constrained areas, reporting obligations on energy and water use, and stricter carbon targets — have made it harder to build hyperscale data centres quickly.

A striking example is the Netherlands, where hyperscale projects faced moratoriums due to land scarcity, nitrogen emissions rules, and public opposition. Microsoft and other hyperscale operators have scaled back expansion there, shifting capacity elsewhere.

Industry Responses

Operators like Equinix highlight the challenge and opportunity. Equinix reports that it now sources 96% of its global electricity from renewables. To cut cooling loads, it has adopted higher operational temperature standards (ASHRAE A1A) and invests in free cooling and waste-heat reuse projects, such as its Paris data centre that supplies local heating. The company also calls for more supportive public policy — streamlined permitting, grid upgrades, and renewable expansion — to balance sustainability with digital growth.

The Trade-offs

The AI/data centre race exposes difficult trade-offs:

Cost vs. Sustainability – Cheap fossil fuel power vs. higher-priced renewables.

Reliability vs. Intermittency – AI workloads demand 24/7 uptime, while renewables fluctuate.

Growth vs. Regulation – Hyperscale expansion often collides with environmental limits, land use, and grid capacity.

Without careful planning, surging AI demand could undermine renewable transition targets, lock in fossil fuel dependency, and inflame community opposition.

Example: Energy for This Write-Up

Even writing this short piece with AI consumed energy. The text is ~1,600 words, or about 2,080 tokens. Based on estimates that each token of AI output requires about 0.000002 kWh, generating this document consumed:

0.0042 kWh (≈ 4.2 Wh)

≈ 15,000 joules

That’s roughly the energy needed to power a 10 W LED bulb for 25 minutes.

This may seem small, but scaled across millions of AI queries daily — and trillions of tokens per year — the cumulative energy use becomes material, highlighting why energy efficiency and clean power are critical for AI’s future.