A US federal appeals court has upheld a law requiring TikTok’s Chinese parent company, ByteDance, to sell its US operations by 19 January or face a nationwide ban. The ruling marks a significant win for the Justice Department, citing national security concerns over ByteDance’s access to Americans’ data and its potential to influence public discourse. TikTok plans to appeal to the Supreme Court, hoping to block the divestment order.
The decision reflects bipartisan efforts to counter perceived threats from China, with Attorney General Merrick Garland calling it a vital step in preventing the Chinese government from exploiting TikTok. Critics, including the ACLU, argue that banning the app infringes on First Amendment rights, as 170 million Americans rely on TikTok for creative and social expression. The Chinese Embassy denounced the ruling, warning it could damage US-China relations.
Unless overturned or extended by President Biden, the law could also set a precedent for restricting other foreign-owned apps. Meanwhile, TikTok’s rivals, such as Meta and Google, have seen gains in the wake of the decision, as advertisers prepare for potential shifts in the social media landscape.
Meta Platforms has teamed up with James Cameron‘s Lightstorm Vision to enhance 3D entertainment production for its Meta Quest headsets. The partnership will focus on live sports, concerts, and TV series, with the Quest serving as Lightstorm’s exclusive mixed reality hardware platform.
This collaboration reaffirms Meta’s commitment to its mixed-reality initiatives, following significant investments in augmented reality and metaverse technologies. In addition, both companies will work on reducing the production costs of creating 3D content, making immersive media more accessible.
In September, Meta introduced a more affordable version of the Quest, aligning with its goal to expand its user base.
Cohere, a Canadian AI startup valued at $5.5 billion, is shifting its focus to developing customised AI models for businesses. Co-founder Nick Frosst explained that enterprise users prefer models tailored to specific use cases rather than larger, general-purpose ones. The company aims to refine its approach by prioritising model deployment and customisation over simply increasing model sizes.
Although Cohere will continue building foundation models, it plans to invest in training techniques to improve functionality. The startup has secured over $900 million in funding from major investors like Nvidia, Cisco, and Innovia Capital. Unlike some competitors, Cohere positions itself as an independent player, working with clients such as Oracle and Fujitsu to design models for their unique requirements.
The AI industry, once focused on scaling up models, now faces diminishing returns from increasing model size. As large language model advancements plateau, Cohere’s customised approach offers a more efficient and cost-effective solution. Frosst highlighted that this strategy aligns with the company’s enterprise-centric vision and avoids reliance on speculative breakthroughs in artificial general intelligence.
By concentrating on tailored AI solutions, Cohere aims to enhance real-world applications for its enterprise clients. This strategy positions the startup as a competitive alternative to larger AI labs such as OpenAI and Anthropic.
Google’s newest AI, the PaliGemma 2 model, has drawn attention for its ability to interpret emotions in images, a feature unveiled in a recent blog post. Unlike basic image recognition, PaliGemma 2 offers detailed captions and insights about people and scenes. However, its emotion detection capability has sparked heated debates about ethical implications and scientific validity.
Critics argue that emotion recognition is fundamentally flawed, relying on outdated psychological theories and subjective visual cues that fail to account for cultural and individual differences. Studies have shown that such systems often exhibit biases, with one report highlighting how similar models assign negative emotions more frequently to certain racial groups. Google says it performed extensive testing on PaliGemma 2 for demographic biases, but details of these evaluations remain sparse.
Experts also worry about the risks of releasing this AI technology to the public, citing potential misuse in areas like law enforcement, hiring, and border control. While Google emphasises its commitment to responsible innovation, critics like Oxford’s Sandra Wachter caution that without robust safeguards, tools like PaliGemma 2 could reinforce harmful stereotypes and discriminatory practices. The debate underscores the need for a careful balance between technological advancement and ethical responsibility
A to-do list app, Twos, is rethinking productivity with AI-driven features that go beyond simple task tracking. Instead of just helping users organise tasks, Twos offers actionable suggestions to help complete them. For instance, writing ‘Buy paper napkins’ prompts the app to suggest links to online stores like Amazon or Walmart. Planning a birthday? Twos might remind you to add a calendar event, send a message, or purchase a gift card.
Launched in 2021 by former Google engineer Parker Klein and Joe Steilberg, Twos integrates with 27 apps, including Spotify, Uber Eats, Google Maps, and Ticketmaster. While the app currently leans on US-centric services, plans for better localisation aim to broaden its appeal. Available across Android, iOS, and the web, Twos is free, with optional premium features like custom sorting and templates priced at $2 each.
Beyond task suggestions, Twos introduced an AI assistant for list creation last year, positioning itself in the growing market of AI-powered productivity tools. The app now boasts over 25,000 active users and emphasises intuitive, energy-efficient design. While other apps like Hypelist compete in this space, Twos’ holistic approach could redefine how we manage daily tasks.
Microsoft has introduced Copilot Vision, an AI-powered feature available in a limited US preview for users of Microsoft Edge. This experimental tool, part of the Copilot Labs program, can read web pages to answer user queries, summarise and translate content, and even assist with tasks like finding discounts or offering gaming tips. For example, it can provide recipes from a cooking site or strategic advice during an online chess game.
To address privacy concerns, Microsoft emphasises that Copilot Vision deletes all processed data at the end of each session and does not store information for model training. The feature is initially restricted to a pre-approved list of popular websites, excluding sensitive or paywalled content, though Microsoft plans to expand compatibility over time.
Microsoft’s cautious rollout reflects ongoing efforts to balance innovation with publisher concerns over AI’s use of web data. The company is collaborating with third-party publishers to ensure the tool benefits users without compromising website content or functionality.
Broadcom has announced breakthrough technology aimed at accelerating custom chip performance in response to rising demand for generative AI infrastructure. The innovation, known as 3.5D XDSiP, enhances memory integration and speeds up processing by directly linking critical components. Developed in collaboration with TSMC, the technique uses advanced packaging methods, including chip-on-wafer-on-substrate, which remains a bottleneck in AI chip supply chains.
The California-based chipmaker has positioned itself as a key player in the AI hardware market, supplying custom processors to cloud providers looking to diversify beyond Nvidia’s pricier options. Although Broadcom has not disclosed its partners, industry experts suggest that major companies like Google and Meta are among its clients.
Broadcom’s CEO, Hock Tan, stated in September that the company expects AI revenue to reach $12 billion in fiscal year 2024, reflecting a significant increase from earlier forecasts. Five products using the 3.5D XDSiP technology are currently in development, with production shipments planned for early 2026.
The custom chip market, valued at $45 billion by 2028, is set to be dominated by Broadcom and rival Marvell. Analysts predict continued growth for both companies as AI infrastructure demand expands globally.
Tamar Eilam, a leading scientist at IBM, is pushing boundaries in sustainable computing. With 24 years at the company, she serves as an IBM fellow and chief scientist for sustainable computing, tackling one of AI’s most pressing challenges its immense energy consumption. Eilam’s work includes spearheading projects like Kepler, an open-source tool that quantifies the energy usage of applications, helping teams adopt greener AI practices.
AI’s energy demands are staggering, with reports showing that a single ChatGPT search consumes ten times more electricity than a typical Google search. Eilam envisions AI as a double-edged sword—capable of aiding sustainability efforts like decarbonising power grids while also requiring extensive resources to function. To address these issues, her team focuses on efficient AI training methods, high-quality data use, and platform optimisations to reduce resource drain.
Eilam advocates for transparency and trust in AI development. IBM’s approach ensures energy-efficient models, like its Granite line, while addressing intellectual property concerns with verifiable data sources. She also highlights the need for startups and companies to prioritise openness about energy impacts and guardrails to prevent misuse.
Beyond technical achievements, Eilam is a vocal advocate for gender diversity in AI. She emphasises breaking unconscious biases and encourages women to persist in leadership roles. Her passion for merging climate solutions with cutting-edge technology continues to shape sustainable computing innovations.
Italian startup iGenius and Nvidia are teaming up to launch one of the largest deployments of Nvidia’s advanced servers by mid-2025. The ‘Colosseum’ data centre in southern Italy will house around 80 GB200 NVL72 servers, each powered by 72 of Nvidia’s latest Blackwell chips, enabling unprecedented AI computing capabilities.
iGenius, a European AI unicorn with over €650 million in funding this year, specialises in open-source AI models for industries like banking and healthcare, prioritising strict data security. The Colosseum system will leverage Nvidia’s tools, including the NIM software platform, to simplify AI model distribution for businesses using Nvidia hardware.
Nvidia executives lauded the partnership, with VP Charlie Boyle highlighting its scale and uniqueness. Multiple Nvidia teams are working closely with iGenius to bring the cutting-edge system online, marking a significant milestone in AI infrastructure development.
FCC Chairwoman Jessica Rosenworcel has proposed requiring US communications providers to certify annually that they have plans to defend against cyberattacks. The move comes amid growing concerns over espionage by ‘Salt Typhoon,’ a hacking group allegedly linked to Beijing that has infiltrated several American telecom companies to steal call data.
Rosenworcel highlighted the need for a modern framework to secure networks as US intelligence agencies assess the impact of Salt Typhoon’s widespread attack. A senior US official confirmed the hackers had stolen metadata from numerous Americans, breaching at least eight telecom firms.
The FCC proposal, which Rosenworcel has circulated to other commissioners, would take effect immediately if approved. The announcement follows a classified Senate briefing on the breach, but industry giants like Verizon, AT&T, and T-Mobile have yet to comment.
