Artificial intelligence

Mistral introduces edge AI models for privacy-first applications

Mistral, a French AI startup, has launched its first generative AI models, ‘Les Ministraux,’ designed to run on edge devices like laptops and mobile phones. The models, Ministral 3B and Ministral 8B, offer versatile applications such as on-device translation and autonomous robotics, catering to privacy-focused, low-latency scenarios.

Both models can process 128,000 tokens, roughly the length of a 50-page book. While Ministral 8B is available for research purposes, commercial licences for self-deployment are being offered directly by Mistral. Developers can also access the models through Mistral’s cloud platform, La Platforme.

Smaller AI models are increasingly in demand due to their cost-effectiveness and efficiency. Mistral claims that its Ministral models outperform competitors like Llama and Gemma across several benchmarks, offering strong instruction-following and problem-solving capabilities.

Paris-based Mistral, which has raised $640 million in venture capital, continues to expand its AI portfolio. The company has introduced services such as developer testing and model fine-tuning, positioning itself as a competitor to major players like OpenAI and Anthropic.

Lightmatter raises funds for photonic chips amid AI race

Photonic chip startup Lightmatter has raised $400 million in a Series D funding round led by T. Rowe Price, valuing the company at $4.4 billion. This funding boost comes amid surging demand for more efficient data center infrastructure due to the rise of artificial intelligence technologies like ChatGPT. With backing from investors such as Fidelity and Alphabet’s GV, Lightmatter plans to use the capital to produce and deploy its innovative chips and expand its workforce across the US and Canada.

The Mountain View-based company, founded in 2017, specialises in using silicon photonics to build faster and more energy-efficient chips, a crucial technology for AI and cloud computing. Co-founder and CEO Nick Harris indicated that this could be the company’s last private funding round, with plans to go public in the near future. He emphasised that photonic chips are the future of high-performance computing, which is why Lightmatter has secured large deals with major tech players, though the company has not disclosed specific clients.

With the potential to work across platforms like Nvidia, Intel, and AMD, Lightmatter’s technology aims to significantly increase AI cluster bandwidth while lowering energy consumption. Its first large AI clusters are expected to be operational next year as the company prepares for an initial public offering.

Urgent need for AI regulation highlighted at global conference

Governments around the world are scrambling to establish protections surrounding AI development and use, according to Doreen Bogdan-Martin, Secretary-General of the International Telecommunication Union (ITU). Speaking at a regulatory conference, she noted that only 15% of ITU’s 194 member states have begun developing policies in this area, leaving the vast majority without regulatory frameworks for AI.

Martin pointed out that many countries are eager to join discussions and learn from others that have advanced in AI regulation. She highlighted India‘s success in building a digital public infrastructure, particularly its unified payments interface and Aadhaar digital identity system, calling the country a global leader in this space. India’s fast 5G network has also positioned it as a key player in the digital economy.

Reflecting on past regulatory efforts to improve internet access, Martin explained that regulators today face a more complex landscape. AI’s integration with telecoms is a particular area of focus, as policymakers consider how to regulate in a world where digital infrastructure is increasingly interconnected.

Martin also stressed the growing challenges posed by cyberattacks, which are rising 80% year on year, and the risks posed by AI-driven deepfakes and disinformation. Strengthening network resilience and ensuring global participation in the digital economy remain critical concerns.

Human-level AI still a decade away, Meta scientist warns

Achieving human-level AI may be at least a decade away, according to Meta’s AI scientist, Yann LeCun. Current AI systems, like large language models, fall short of true reasoning, memory, and planning, even though companies like OpenAI market their technologies with terms like ‘memory’ and ‘thinking’. LeCun cautions against the hype, saying these systems lack the deeper understanding required for complex human tasks.

LeCun argues that the limitations stem from how these AI models function. LLMs predict words, while image and video models predict pixels, making them capable of only single or two-dimensional predictions. In contrast, humans operate in a three-dimensional world, able to plan and adapt intuitively. Even the most advanced AI struggles with everyday actions, such as cleaning a room or driving a car, tasks children and teenagers can learn with ease.

The key to more advanced AI, according to LeCun, lies in ‘world models’ – systems capable of perceiving and predicting outcomes within a three-dimensional environment. These models would allow AI to form action plans without trial and error, similar to how humans quickly solve problems by envisioning the results of their actions. However, building these systems requires massive computational power, driving cloud providers to partner with AI companies.

FAIR, Meta’s research arm, has shifted its focus towards developing world models and objective-driven AI. Other labs are also pursuing this approach, with researchers such as Fei-Fei Li raising significant funding to explore the potential of world models. Despite growing interest, LeCun emphasises that significant technical challenges remain, and achieving human-level AI will likely take many years, if not a full decade.

AI platform aims to ease electrician shortage with faster tech installations

With the US facing a growing shortage of electricians, Treehouse, a startup, is using AI to help make installations of electric vehicle (EV) chargers, heat pumps, and other tech more efficient and affordable. As the demand for renewable energy and electrification surges, Treehouse has developed AI models to predict job times, materials needed, and to eliminate unnecessary site visits, which typically slow down the process. By gathering data and asking customers key questions, Treehouse can streamline quoting and installation processes, especially for simpler jobs like EV chargers.

Treehouse uses its AI-driven platform to design installations and assist in permitting, often completing jobs with minimal visits from electricians. For more complex installations, like heat pumps, the company may require additional photos or virtual visits. Treehouse works in 40 states, hiring independent electricians for many of the jobs, and plans to expand its operations across all 50 states by the end of the year. With a recent $16 million Series A funding round, the company aims to improve its AI tools and grow its team.

Founder and CEO Eric Owski believes the electrification trend will continue to reshape how consumers think about home energy, with EV chargers being just the starting point. Treehouse’s partnerships with companies like CarMax and ChargePoint also help boost its reach as it tackles the ongoing electrician shortage.

NYT issues cease-and-desist to Perplexity over AI content use

The New York Times has issued a cease-and-desist notice to the AI company Perplexity, demanding it halt the use of its content for generating summaries and other outputs. The newspaper claims that Perplexity’s practices violate copyright law, adding to the ongoing tensions between media publishers and AI firms.

The letter from NYT highlighted concerns over how Perplexity continues to use its articles despite promises to stop. Perplexity, which previously agreed to cease using crawling technology, assured it does not scrape data to train models but instead indexes web pages to provide factual citations when responding to user queries.

Perplexity is required to provide details on how it accesses the NYT website by 30 October. The startup has faced similar allegations from other media outlets, including Forbes and Wired, but has since introduced a revenue-sharing programme to address some concerns.

The NYT has taken a strong stance on generative AI, having also sued OpenAI for allegedly using millions of its articles without permission to train its chatbot. The conflict underlines broader worries among publishers about how AI companies are using their content.

SimpliSafe launches new outdoor monitoring solution

SimpliSafe has launched the Active Guard Outdoor Protection service, enhancing its security offerings with a combination of AI and human monitoring. Priced at $50 per month, this new tier builds on its $32 indoor monitoring plan, providing 24/7 protection for outdoor spaces through advanced surveillance.

The new service relies on the Outdoor Security Camera Series 2, which features an ‘AI for the Familiar Face’ feature. This AI minimises false alarms by identifying known visitors. If an unrecognised person is detected, a human agent is alerted and can intervene by activating lights, triggering a siren, or notifying the authorities.

Executives at SimpliSafe emphasise that human agents retain the final decision-making authority, using AI only as a support tool. Hooman Shahidi, SVP of Product, stated that the company prioritises human judgement and workforce diversity to ensure fair monitoring practices. CEO Christian Cerda noted that while the company explores generative AI, it remains cautious about implementing new technologies.

The Series 2 camera costs $200 and offers HD recording, a 140-degree field of view, and two-way communication. It can be powered by batteries or connected to a power source and is waterproof for outdoor use. SimpliSafe, founded in 2006, operates primarily in the US but has expanded to the UK since 2019.

Apple’s new iPad mini with AI tools goes on pre-order

Apple has introduced its new iPad mini, equipped with advanced AI capabilities and powered by the A17 Pro chip, which is also used in the iPhone 15 Pro models. The upgraded iPad mini is set to deliver 30% better CPU performance compared to its predecessors and will include AI-driven writing tools and an enhanced Siri assistant, running on Apple’s AI software, Apple Intelligence.

A new software update, iPadOS 18.1, will roll out the first AI features in the United States English version this month for iPads using the A17 Pro or M1 chips and later models. Additional AI features, such as image generation and ChatGPT-powered tools, are expected to be released over the coming months.

Despite initial concerns over the iPhone 16 launch in September, analysts predict the new iPad mini and AI-driven devices will boost Apple’s sales in the fourth quarter and sustain momentum into 2025. Apple’s third-quarter shipments already reached a record high.

The new iPad mini, starting at $499, is now available for pre-order and will start shipping to customers and stores next week, according to Apple.

Dane Stuckey joins OpenAI as it boosts security for AI technologies

Dane Stuckey, former Chief Information Security Officer (CISO) of Palantir, has been appointed as the new CISO at OpenAI, working alongside head of security Matt Knight. Stuckey made the announcement in a post on social media, expressing his excitement to help secure OpenAI’s technologies as they continue to grow in use and impact.

Stuckey, who joined Palantir in 2014, brings extensive experience in digital forensics and incident response, having worked in both commercial and government roles. His background may prove valuable as OpenAI continues to deepen its partnerships with the United States Department of Defense, with whom it has collaborated on various cybersecurity projects.

OpenAI has been expanding its security efforts in recent months, following the appointment of former National Security Agency head Gen. Paul Nakasone as a board member. The company has also lifted its ban on selling AI technology to the military, signalling a strategic shift towards government contracts.

In addition to Stuckey’s appointment, OpenAI has posted a new job listing for a head of trusted compute and cryptography, highlighting its commitment to developing secure AI infrastructure to protect its technologies and users.

Electronics and mobility sectors unite in Japan

Japan’s largest annual electronics event opened alongside a mobility show, marking the first joint trade fair of its kind. The collaboration reflects the increasing convergence of technology and automotive industries, especially as vehicles become more autonomous and connected.

The trade show, hosted by the Japan Electronics and Information Technology Industries Association (JEITA) and Japan Automobile Manufacturers Association (JAMA), aims to promote cross-industry innovation. AI emerged as a core theme, with around half of the 800 tech exhibitors presenting AI-driven products and solutions.

Toyota Motor showcased a portable hydrogen tank capable of powering electric generators during disasters, promoting hydrogen as a sustainable energy source. Panasonic highlighted its perovskite solar cells, which can be installed on car windows to enhance power efficiency for electric vehicles, while Sony demonstrated a safety system that uses image sensors to detect driver fatigue.

NEC presented an AI-powered service capable of summarising movies or creating accident reports from dashcam footage, offering applications in various fields. TDK introduced a brain-inspired semiconductor chip that reduces AI electricity consumption to one-hundredth of current levels. The fair runs until Friday at Chiba’s Makuhari Messe, with free entrance for online registrants.