Researchers have achieved a milestone in AI, teaching it to predict the complex aromas of whiskies and even identify their origins. The study, conducted in Germany, utilised AI to analyse the molecular makeup of 16 American and Scottish whiskies. It then predicted the five strongest aroma notes and distinguished between the two countries of origin with remarkable accuracy.
The AI surpassed human experts in consistency and precision, identifying aromas like menthol and citronellol for US whiskies and smoky, medicinal notes for Scotch. This innovation could ensure flavour consistency in whisky production, detect counterfeit goods, and even find applications in blending recycled materials to reduce odours.
While promising, the study was limited to a small selection of whiskies, raising questions about its performance on broader varieties or aged batches. Experts also noted that flavour perception depends on external factors, highlighting room for further exploration in this emotive domain. Nonetheless, this blend of technology and tradition signals a new step for the whisky industry.
Elon Musk’s AI company, xAI, has raised $6 billion in its latest funding round, doubling its total to $12 billion this year. The investment attracted high-profile backers such as Andreessen Horowitz, BlackRock, and Fidelity, with participation limited to existing investors. Reports suggest the company is now targeting a $50 billion valuation.
Founded last year, xAI released its flagship generative AI model, Grok, which powers features on X, formerly known as Twitter. Grok, known for its bold and unconventional responses, has integrated capabilities like image generation and news summarisation. The company has also launched APIs and a standalone app, aiming to compete with AI giants like OpenAI and Anthropic.
The company’s Memphis data centre, housing 100,000 Nvidia GPUs, is central to training the next generation of AI models. Plans are underway to double GPU capacity and secure additional power to support operations. However, these efforts have faced criticism over potential environmental impacts.
xAI envisions integrating its AI models with Musk’s other ventures, such as Tesla and SpaceX, sparking concerns among Tesla shareholders. Despite these challenges, xAI’s rapid growth positions it as a formidable contender in the expanding AI industry.
Venture funding in Europe may be headed for a flat year overall, but European AI startups are thriving, with AI companies receiving 25% of the region’s VC funding in 2024, totalling $13.7 billion. This marks a significant rise from 15% four years ago and has led to the creation of new unicorns like Poolside and Wayve. According to James Wise of Balderton Capital, breakthrough AI technology in Europe can now attract hundreds of millions, or even billions, of euros at the early stages, similar to the US.
The collective value of European AI companies has doubled in four years, reaching $508 billion, now making up nearly 15% of the region’s entire tech sector. While much of the funding still comes from outside Europe, especially the US, the local AI ecosystem is flourishing with a growing talent pool. In 2024, 349,000 people were employed by AI companies in Europe, a 168% increase since 2020, indicating a buoyant and increasingly productive sector.
Wise suggests that the rise of smaller, highly productive AI companies will be the future, with generative AI tools significantly boosting efficiency in various industries. This growing adoption of AI tools is likely to continue benefiting the European AI sector in the long run, even if the category becomes less distinct in the future.
Quantization, a technique widely used to improve AI efficiency, may have reached its limits, according to recent research. This method reduces the precision of data in AI models, making them faster and cheaper to operate. However, studies suggest that as models grow larger and are trained on vast datasets, quantization can degrade their performance.
Researchers from leading institutions found that highly trained models suffer more when quantised. For AI companies relying on massive models to enhance quality, this finding raises concerns about the long-term viability of cost-saving approaches. Quantization already impacts models like Meta’s Llama 3, which reportedly shows reduced performance compared to other AI systems.
Efforts to lower AI model costs continue as inference—using models to generate responses—remains the most significant expense for AI labs. Techniques like training in low precision and hardware supporting ultra-low bit precision are being explored. Yet, such strategies face diminishing returns and risks of quality loss if precision drops too far.
Experts believe a shift towards better data curation and filtering, alongside new architectures optimised for low-precision training, may offer solutions. These advancements could help balance efficiency and performance as AI evolves beyond traditional scaling methods.
Apple is closing in on a historic $4 trillion market valuation, driven by investor enthusiasm over its advancements in artificial intelligence and hopes for a surge in iPhone upgrades. Shares have surged 16% since November, adding $500 billion to its market cap, and positioning Apple ahead of rivals Nvidia and Microsoft in the race to this milestone. Analysts attribute the rally to expectations of a new “supercycle” in iPhone sales fueled by AI enhancements, despite modest revenue growth projections for the holiday season.
Apple’s integration of AI tools like OpenAI’s ChatGPT across its devices and apps marks a strategic pivot in a market long dominated by Microsoft, Alphabet, and Meta. Although iPhone demand remains muted, analysts forecast a rebound in 2025, as AI-powered features and broader availability drive renewed interest. Meanwhile, Apple’s premium valuation—its price-to-earnings ratio recently hit a three-year high of 33.5—has sparked mixed reactions among investors, with Warren Buffett’s Berkshire Hathaway scaling back its holdings.
Despite challenges such as geopolitical risks and fluctuating market conditions, Apple’s approach to this milestone underscores its enduring dominance in the tech sector. Analysts and investors remain optimistic about the company’s ability to navigate near-term hurdles and leverage AI innovation to maintain its leadership in a competitive landscape.
Microsoft is taking steps to diversify the AI powering its flagship product, Microsoft 365 Copilot. While OpenAI’s GPT-4 model has been a cornerstone of the AI assistant since its launch in March 2023, Microsoft is now integrating internal and third-party AI models, including its proprietary Phi-4, to reduce costs and improve efficiency. This move reflects Microsoft’s broader strategy to lessen reliance on OpenAI, its long-time partner, as it looks to offer faster, more cost-effective solutions to enterprise customers.
The shift is driven by concerns over the high costs and slower speeds associated with OpenAI’s technology for enterprise users. A company spokesperson confirmed that OpenAI remains a partner for advanced models but emphasised that Microsoft customises and incorporates a range of AI models depending on the product. Beyond its collaboration with OpenAI, Microsoft is also customising open-weight models to make its services more accessible and affordable, with potential cost savings for customers.
Microsoft’s approach mirrors similar changes in its other business units. For example, GitHub, acquired by Microsoft in 2018, has started incorporating AI models from Anthropic and Google as alternatives to OpenAI’s offerings. These efforts align with Microsoft’s goal of demonstrating the return on investment for its AI tools, particularly as some enterprises remain cautious about adopting 365 Copilot due to concerns over pricing and utility.
Despite these challenges, Microsoft reports growing adoption of 365 Copilot. The company states that 70% of Fortune 500 companies are using the AI assistant, and analysts predict that more than 10 million users will adopt it this year. As Microsoft continues refining its AI technology, leaders like CEO Satya Nadella are keeping a close watch, underscoring the company’s commitment to innovation in enterprise AI.
Google has introduced Gemini 2.0 Flash Thinking Experimental, an AI model designed for advanced reasoning, now available on its AI Studio platform. Billed as effective for multimodal understanding, coding, and complex problem-solving, it aims to enhance AI’s reasoning capabilities.
Unlike typical AI, reasoning models like Gemini fact-check themselves during response generation, improving accuracy but requiring more processing time. However, early testing shows mixed results, suggesting room for refinement in practical applications.
The rise of reasoning models reflects the industry’s search for new methods to optimise AI performance. While promising, challenges such as high computational costs and uncertain scalability remain points of debate.
Elon Musk’s AI venture, xAI, has unveiled a standalone iOS app for its chatbot, Grok, marking its first major expansion beyond the X platform. The app, currently in beta testing across Australia and a few other regions, offers users an array of generative AI features, including real-time web access, text rewriting, summarisation, and even image generation from text prompts.
Grok, described as a ‘maximally truthful and curious’ assistant, is designed to provide accurate answers, create photorealistic images, and analyse uploaded pictures. While previously restricted to paying X subscribers, a free version of the chatbot was launched in November and has recently been made accessible to all users.
The app also serves as a precursor to a dedicated web platform, Grok.com, which is in the works. xAI has touted the chatbot’s ability to produce detailed and unrestricted image content, even allowing creations involving public figures and copyrighted material. This open approach sets Grok apart from other AI tools with stricter content policies.
As the beta rollout progresses, Grok is poised to become a versatile tool for users seeking generative AI capabilities in a dynamic and user-friendly interface.
Data centres in the United States could consume up to 12% of the country’s electricity by 2028 due to the rapid growth of AI, according to a new report. The Department of Energy-backed study predicts energy usage from data centres will rise from 4% to between 6.7% and 12%, depending on GPU availability and demand.
The shift to AI-driven infrastructure is driving the surge, with GPU-accelerated servers and cooling systems responsible for doubling power use in recent years. Researchers are calling for annual reports and strategies to track trends and enhance efficiency.
The findings highlight concerns about the impact of AI on power grids, energy bills, and climate change. Researchers also suggest increased transparency in data centre energy use, aiming to encourage efficiency and sustainable growth within the industry.
NETSCOUT SYSTEMS announced significant updates to its Arbor Edge Defense (AED) and Arbor Enterprise Manager (AEM) products as part of its Adaptive DDoS Protection solution. These enhancements are designed to address the growing threats of AI-enabled DDoS attacks, which have surged in sophistication and frequency.
Application-layer and volumetric attacks have increased by 43% and 30%, respectively, with DDoS-for-hire services making attacks easier to execute. To combat these evolving threats, NETSCOUT leverages AI and machine learning (ML) within its ATLAS Threat Intelligence system, which monitors over 550 Tbps of real-time internet traffic across 500 ISPs and 2,000 enterprise sites worldwide.
The AI/ML-powered solution enables dynamic threat identification and mitigation, creating a scalable, proactive defence mechanism. The updated AED and AEM products automate a closed-loop DDoS attack detection and mitigation process, providing real-time protection by adapting to changing attack vectors and applying mitigation recommendations automatically.
NETSCOUT’s solution also offers comprehensive protection across hybrid IT environments, including on-premise infrastructure, private data centres, and public cloud platforms like AWS and Microsoft Azure, with enhancements such as 200 Gbps mitigation capacity, high-performance decryption, and visibility into non-DDoS threats.
By minimising downtime and safeguarding business-critical services, NETSCOUT’s Adaptive DDoS Protection reduces business risks and protects productivity and reputation. As the threat landscape continues to evolve, organisations can rely on NETSCOUT’s innovative technology to stay ahead of attackers and maintain IT resilience. Industry experts and agencies like the Cybersecurity & Infrastructure Security Agency (CISA) highlight the need for adaptive cybersecurity measures. NETSCOUT’s AI/ML-driven solutions meet these demands by offering robust, future-proof protection for critical IT infrastructure.
