AI Research Under Threat

Trump Administration Cuts Spark Controversy

The Trump Administration is reportedly cutting AI experts from the National Science Foundation (NSF), potentially impacting the agency's AI research efforts. According to Bloomberg, the Directorate for Technology, Innovation, and Partnerships, which is crucial for distributing AI-focused government grants, has been affected. This has led to postponed or canceled review panels and stalled funding for some AI projects. AI experts have criticized these cuts, particularly those supported by Elon Musk's Department of Government Efficiency. Geoffrey Hinton, an AI pioneer, has called for Musk's expulsion from the British Royal Society, to which Musk responded that he would try to fix any mistakes.

Cybersecurity

AI for Dark Web Monitoring

The article addresses the difficulties in monitoring and restricting illicit actions on the dark web. Standard algorithms are often ineffective against malicious actors employing tools like Tor Browser and VPN services to circumvent restrictions. The article proposes that a more profound examination of user behavior, including time spent on a page, website visited, and traffic patterns, is necessary to distinguish individuals bypassing restrictions from average users. The potential of AI to improve the monitoring and identification of illegal activities on the dark web is also examined.

Artificial Intelligence

AI Plays Super Mario Bros.

Researchers at the University of California San Diego tested AI's ability to play Super Mario Bros. Anthropic’s Claude 3.7 performed best, followed by Claude 3.5. Google’s Gemini 1.5 Pro and OpenAI’s GPT-4 struggled in comparison. The AI used the GamingAgent framework, receiving basic instructions and in-game screenshots to generate Python code, controlling Mario's actions. Reasoning models performed worse than non-reasoning models due to longer decision-making times. Experts debate the relevance of AI gaming skills to overall technological advancement, citing the abstract and simple nature of games and the availability of infinite training data. Andrej Karpathy from OpenAI highlighted an 'evaluation crisis,' questioning how to effectively assess these models' capabilities.

AI Developments

AI Efficiency and Cost Reduction

Zoom researchers have introduced a "chain of draft" method that significantly reduces AI token usage by 92%, potentially revolutionizing the economics of language model deployment.

AI-Powered Data Processing

A Google data science agent helped one scientist reduce data processing time dramatically, from one week to just five minutes.

AI Writing Trends

Stanford researchers analyzed 305 million texts, uncovering trends in AI writing adoption. The study surprisingly found that areas with less education are adopting AI writing tools at a faster rate.

Google Gemini Updates

Google Gemini users can now access the AI chatbot directly from the iPhone’s lock screen via a widget. This update also includes lock screen widgets for taking pictures and uploading them to Gemini, setting reminders, and starting text chats. Google also announced that Gemini users on Android will soon be able to ask the AI chatbot questions about video and on-screen content and get answers in real time, a feature that will be available for Google’s Gemini Advanced plan subscribers. This development comes as Apple’s AI-enabled Siri faces delays, allowing AI competitors to provide AI assistants to iPhone users.

Technology and Society

AI's Potential vs. Societal Risks

At the Mobile World Congress, Ray Kurzweil from Google highlighted AI's potential to improve healthcare, longevity, and sectors like solar energy. Conversely, Scott Galloway cautioned against rage-fueling algorithms that exploit polarization and isolation, especially affecting young men. He also criticized tech CEOs, including Elon Musk, for not addressing democratic abuses and tax advantages for the wealthy.

AI Industry Investment and Development

Anthropic's Funding Round

Anthropic has secured $3.5 billion in Series E funding, reaching a valuation of $61.5 billion. The AI company's revenue has increased by 1,000% year-over-year, intensifying competition with OpenAI.

California's AI Regulation Efforts

California State Senator Scott Wiener introduced AI bill SB53 to protect employees at leading AI labs, allowing them to speak out about potential “critical risks” to society. SB53 defines critical risks as those potentially resulting in death or serious injury to over 100 people, or more than $1 billion in damage. The bill also proposes a public cloud computing cluster, CalCompute, to provide researchers and startups with resources for developing AI for public benefit.

Autonomous Vehicle Advancements

Google and several automakers are actively developing autonomous vehicles for American roads, which are currently legal in three states. Google's fleet has accumulated over 300,000 miles. Recent developments include Waymo releasing a software update to address honking issues in San Francisco, and Amazon beginning robotaxi tests in Austin and Miami.

AI Tools and Platforms

1min.AI Lifetime Access

1min.AI is offering lifetime access to its AI-powered tools for A$160. This allows users to generate content, edit visuals, and enhance creativity without recurring monthly fees. The platform utilizes AI models for article generation, rewriting, and image creation.

Google Gemini's New Features

Google has teased the upcoming release of its Gemini AI with new features at Mobile World Congress in Barcelona. Gemini Live with Video and Gemini Live with Screenshare will allow Google One AI Premium subscribers to share their phone screens and real-time video with Gemini, enabling the chatbot to answer questions based on visual input. This feature is expected to launch later this month.

Google's Open-Source SpeciesNet

Google has open-sourced SpeciesNet, an AI model designed to identify animal species by analyzing photos from camera traps. Trained on over 65 million images, it classifies images into over 2,000 labels. Available on GitHub under an Apache 2.0 license, SpeciesNet will help scale the monitoring of biodiversity in natural areas.

AI Events

TechCrunch Sessions: AI – Call for Speakers

TechCrunch Sessions: AI is scheduled for June 5th at Zellerbach Hall, UC Berkeley. AI innovators are invited to apply to lead a 50-minute breakout session before the deadline on March 7, 2025. Selected speakers will gain access to the main stage, additional breakout sessions, and networking opportunities, with promotion across event listings and social media.

AI in Healthcare

Microsoft Dragon Copilot

Microsoft has unveiled Microsoft Dragon Copilot, an AI system tailored for healthcare. It leverages voice dictation and ambient listening technology to generate notes based on clinical visits. This tool aims to streamline documentation through features such as multilingual ambient note creation and natural language dictation, ultimately reducing administrative burdens and allowing clinicians to focus more on patient care. The AI assistant also provides medical information searches and automates tasks like summaries and referral letters.

AI Models and Development

Podcastle's Asyncflow v1.0

Podcast recording and editing platform Podcastle has launched its own AI model, Asyncflow v1.0, for text-to-speech conversion. This model offers over 450 AI voices. An API will be available for developers to directly integrate the technology into apps. Podcastle is also upgrading its voice cloning feature, requiring only a few seconds of recording to create a voice clone.

Anthropic and OpenAI's Government Partnerships

AI companies like Anthropic and OpenAI are expanding their partnerships with the U.S. government. They recently hosted an AI Jam Session, gathering 1,000 scientists to test their latest models, including OpenAI's o3-mini and Anthropic's Claude 3.7 Sonnet. This event is part of existing agreements with the Department of Energy and the National Nuclear Security Administration to assess AI's capabilities in scientific research and nuclear safety.

AI Web Browsing

Opera's Browser Operator

Opera has introduced Browser Operator, an AI agent for its web browser that can perform browsing tasks for users. This agentic browsing allows users to explain what they need in natural language, and the browser performs the necessary tasks. Browser Operator runs natively inside the browser, using the DOM Tree and browser layout data for context, making it faster and more efficient while protecting user privacy.

AI Assistants

Apple's Siri Upgrade Delay

According to reports, Apple's fully upgraded, conversational version of Siri may not reach consumers until iOS 20, around 2027. While Apple Intelligence was initially marketed as a seamless personal assistant, its rollout has been limited. Experts suggest Apple Intelligence continues to lag behind competitors like Amazon's Alexa+ and conversational assistants from Google and ChatGPT.

AI Security

AI-Powered Phishing Attacks

AI-enhanced phishing attacks are a growing security threat, prompting the need for companies to adopt AI-based defenses. Gartner's survey identifies AI-based cyber threats as top emerging risks. AI phishing attacks use artificial intelligence to create personalized and convincing emails by analyzing targets' social media profiles and online activities. Zeus Kerravla of ZK Research advises using AI to combat these attacks. Companies should pay attention to AI and Large language models (LLMs) being the latest sophisticated tools adopted by attackers.

AI Models and Platforms

Advancements in AI Models

OpenAI has released GPT-4.5 as a research preview, featuring advancements in unsupervised learning for improved world knowledge and factual accuracy. Claude 3.7 Sonnet by Anthropic integrates standard response capabilities with detailed reasoning, allowing users to toggle between quick responses and step-by-step thinking. Perplexity's Deep Research is now free for autonomous analysis, and specialized financial models are available. Tools to simplify the development of AI agents and apps are also being introduced, making it easier for developers to create AI solutions.

Simplifying AI Adoption

Many businesses face challenges adopting generative AI due to technical complexity, high costs, talent shortages, and scalability issues. Kaustav Das Sharma, Founding Engineer at Gumloop, aims to overcome these obstacles by creating intuitive, scalable, and user-focused AI solutions. Sharma's career centers on developing tools that simplify AI adoption, such as a generative AI tool simplifying database queries using plain language instead of SQL.

Amazon's AI Integration

Panos Panay, Amazon's head of devices and services, discusses the new AI-powered Alexa Plus. Panay, emphasizes the challenges of integrating AI into Alexa, requiring collaboration across various parts of Amazon. He highlights the importance of "experts" and sees AI as transforming how we interact with computers. Vishal Sharma, Amazon's VP of Artificial General Intelligence, states that Amazon is deploying AI across its various divisions, including AWS, robotics, and Alexa. Amazon utilizes around 750,000 robots in its warehouses. AWS offers Nova and the Bedrock service, and Amazon is collaborating with Anthropic to build a large AI compute cluster using its Trainium2 chips.

Google's AI Initiatives

Google's Gemini app for iOS and iPadOS now features lock screen widgets for quicker access to various AI assistant functions. Google Colab is integrating a new "AI agent" called DataScienceAgent to help users clean data, visualize trends, and gain insights from uploaded datasets. DataScienceAgent utilizes Google's Gemini 2.0 AI model and reasoning tools, supports CSV, JSON, or .txt files under 1GB.

Deutsche Telekom's AI Phone

Deutsche Telekom is collaborating with Perplexity to develop a new "AI Phone" powered by the Perplexity chatbot. The device, named "MagentaAI," will provide users access to Perplexity Assistant, Google Cloud AI, ElevenLabs, Picsart, and other AI tools. The AI phone is scheduled to launch in the second half of the year.

Microsoft's Copilot Updates

Microsoft is revamping its Copilot experience with a new card-based design across mobile, web, and Windows. Copilot gains personalization features like Copilot Vision, a natural voice conversation mode, and a news presenter mode. Windows 11 will show Phone Link status in the start menu, and Paint and Photos are getting AI-powered features like Generative Fill and Erase. Copilot Plus PCs will have an improved AI-powered Windows Search with a "Click to Do" feature and text-based photo search.

Artificial Intelligence

Gemini Updates

• Google is enhancing Gemini by making a paid feature available to all users and giving it the ability to "see." The updated Gemini can now remember user preferences and details, which was formerly only available to Gemini Advanced users, and it will offer more pertinent responses. Users have the option to manually add saved information via the settings menu. Gemini Live can now "see" through screen sharing or live video, allowing users to ask questions about what's on their phone screen or surroundings. This functionality is currently accessible to Gemini Advanced paying customers.

AI Impact on Job Market

• The quick development of technology has given rise to unprecedented innovations, and AI has truly arrived, having a significant influence on the global economy. AI is changing how businesses are conducted across industries, which will inevitably lead to changes in employment responsibilities. By 2025, it is predicted that more than 85 million jobs will be impacted by automation and AI, but 97 million new roles may also emerge that are better suited to the new methods of operation.

Anthropic's Funding and Growth

• AI startup Anthropic has secured $3.5 billion in Series E funding, led by Lightspeed Venture Partners, with a post-money valuation of $61.5 billion. The capital will support the advancement of AI systems, expand computing capabilities, enhance research efforts, and expedite international growth. Anthropic recently introduced Claude 3.7 Sonnet, an AI model designed to simplify user interaction. The company's annual revenue run rate was approximately $1 billion last year and has increased by 30% this year, but they anticipate burning $3 billion this year on AI systems development.

AI Development & Chip Restrictions

Chip Procurement Amidst Restrictions

Despite U.S. export controls aimed at restricting China's access to AI-related chips, Chinese buyers are reportedly obtaining computing systems with Nvidia's Blackwell chips through third-party traders in regions like Malaysia, Taiwan, and Vietnam. These restrictions were introduced to prevent the use of U.S. technology for military modernization. Microsoft has urged the Trump administration to ease these restrictions to tap into China's AI market, while China has advised its AI researchers to avoid visiting the U.S.

AI in Mobile Technology

Deutsche Telekom's AI Phone

Deutsche Telekom (DT) is developing an "AI Phone" in collaboration with Perplexity and others, featuring a new AI assistant app called "MagentaAI." This low-cost handset, aimed at the European market, will be unveiled in the second half of this year and start selling in 2026 for under $1,000. Perplexity, known for its generative AI search engine, plays a key role in developing the phone's AI interface, transitioning from an answer machine to an action machine capable of booking flights and reservations. This move highlights carriers' efforts to compete with technology companies like Apple and Google.

Google's AI Features on Android

Google is previewing two AI features at Mobile World Congress 2025 that will roll out to Android devices later this month. The first is Live Video, which uses Gemini's multi-modal capabilities to allow users to show the chatbot what their phone's camera sees and ask questions. The other new feature is screen sharing, which allows users to share their screen with Gemini to ask questions about what they see on a website or app. These features will be available to Gemini Advanced subscribers, included in the $20 per month Google One AI Premium plan.

Apple's Siri Delay and AI Strategy

Apple may delay the launch of its Apple Intelligence-enhanced Siri, with the true conversational Siri potentially not arriving until 2027. iOS 19, set to debut at WWDC 2025, might roll out the new Siri, but the software may not launch until spring 2026, possibly as part of iOS 19.4. Recent code leaks suggest Apple might allow other AI models like Gemini on iPhones, though this could be limited, similar to OpenAI's integration with Siri. Apple may focus on making a phone that doesn’t rely on energy-hungry AI, as recent iOS updates eat up more storage despite a survey indicating that few Apple buyers are getting the latest iPhones for Apple Intelligence.

AI in Entertainment

AI at the Oscars 2025

The use of AI in Hollywood was a hot topic at the 2025 Oscars, with comedian Conan O’Brien joking about not using AI in the show. Controversy arose over the film "The Brutalist," which won three Oscars but used AI to tweak actors' dialogue. Director Brady Corbet defended the use, stating it only refined vowels and letters for accuracy. "Emilia Pérez," another winner, also used AI to enhance an actress's voice. The role of AI in film has sparked debate due to concerns about job displacement. The Academy is considering making it mandatory for filmmakers to disclose AI use in their submissions.

Artificial Intelligence

AI Model Training and Optimization

Deepseek's lead has spurred AI firms to use a "teacher" LLM to train smaller AI systems, creating cheaper models.

Audio Generation on Mobile Devices

Stability AI and Arm have partnered to optimize StableAudio Open, an AI model for generating audio, for mobile devices with Arm chips. StableAudio Open creates sound effects from text descriptions, using royalty-free audio. The optimized model can generate an 11-second audio sample in about 8 seconds on an Armv9 CPU, with Stability claiming a 30x speed improvement. Stability aims to make its models available for consumer apps and devices, and is collaborating with Arm for further optimization.

Mobile Technology

T-Mobile's AI Phone

T-Mobile is developing an "AI Phone" in collaboration with Perplexity and other companies, planning to launch it in 2026 for under $1,000. The phone will feature a new AI assistant app called "MagentaAI" and integrate Perplexity's AI technology for proactive tasks like booking flights and sending emails. The device aims to offer an AI-driven experience, with services including Google Cloud AI, ElevenLabs, and Picsart. MagentaAI will also be available as an app for T-Mobile customers on Android and iOS.

AI Assistant & Browser Automation

Opera's Browser Operator: Automating Web Tasks

Opera has introduced Browser Operator, a new agent feature currently in preview, designed to automate tasks on websites like finding products, booking tickets, and arranging travel. Users retain oversight and control. Opera emphasizes user privacy, as Browser Operator operates natively without cloud instances or virtual machines. The feature will soon be available through Opera's feature drop program.

Jolla's Privacy-Focused AI Assistant

Jolla, through Venho.ai, has unveiled a privacy-centric AI assistant that integrates with apps to summarize emails and documents, book meetings, filter social media, and perform web look-ups. It can also create AI agents and act as a shopping and personal aide-mémoire. The software, paired with proprietary AI hardware, uses smaller, locally hosted AI models and a vector database to unify data while ensuring user privacy. Jolla plans to launch an AI agent marketplace next month.

AI and General Intelligence

The Human Brain as a Model for General Intelligence

Ars Technica reports that the human brain serves as an existing example of general intelligence, contrasting it with current AI capabilities.

Artificial Intelligence

RAG Data Security

Securing Retrieval-Augmented Generation (RAG) data is crucial, especially when using chatbots. The rise of foundation models like OpenAI, Anthropic, and Llama necessitates addressing limitations such as chatbots generating incorrect answers due to training on public data. RAG, which provides relevant context to the model, and vector databases, known for efficient vector search, are key components.

AI Social Media Tools

A six-month experiment evaluated major AI social media platforms using real client accounts and budgets. Pressmaster.ai was identified as the top AI social media tool for agencies and business owners in 2025. The evaluation also covered Sprout Social, Lately.ai, Hootsuite, Buffer, Copy.ai, and Jasper.ai, highlighting their individual strengths and weaknesses through real-world application.

Google Gemini Live Updates

Google is enhancing its continuous conversational AI, Gemini Live, with live video and screen sharing. Announced at Mobile World Congress in Barcelona, the update enables real-time visual interaction with the AI chatbot on Android devices. The "see" capability, initially teased in August 2024, allows users to interact with the AI about their surroundings. Live video features will be available to Gemini Advanced subscribers in March, following the release of Gemini 2.0 Pro and Gemini 2.0 Flash Thinking Experimental.

AI Tools & Software

Resume Optimization

Resume From Space is an AI-powered resume optimization tool designed to improve job-hunting success. It helps users create ATS-friendly resumes with smart recommendations for rewording and structuring, highlighting accomplishments, and optimizing keywords. The tool also offers customizable templates and interview training features with practice questions and scoring. Lifetime access is available for $59.99.

Photo Editing

Luminar Neo is an AI photo editor that allows users to edit photos. It includes layers, masking, local adjustments, and AI-powered editing tools for skin enhancement, sky replacement, and image upscaling. The bundle includes the Luminar Neo photo editor for Windows, Mac, or as a plugin for Photoshop or Lightroom, a video training course on photo-editing techniques, and six packs of preset photo filters. With a one-time purchase, users gain lifetime access to Luminar Neo. For a limited time, it's available for $119.99.

Google Gemini Updates

Google is introducing new features to its AI assistant, Gemini, allowing users to pose questions using video or on-screen content in real-time. The "Screenshare" feature enables users to share their phone screen with Gemini and inquire about its contents. The video search feature lets users record videos and ask Gemini questions while filming. These features will be available to Gemini Advanced users on the Google One AI Premium plan on Android later this month.

AI Implementation in Enterprises

CIOs Plan to Implement AI Agents

A survey of 1,050 CIOs indicates that 93% plan to implement AI agents within the next two years. These implementations focus on removing data silos to facilitate the technology.

Enterprises use an average of 897 applications, but only 29% are integrated, hindering unified experiences. CIOs are allocating 20% of their budget to data infrastructure and management, compared to 5% for AI itself.

AI agents are expected to boost the adoption of digital applications and revolutionize human-computer interaction by understanding intent, planning actions, and continuously learning. This may lead to increased software spending and improved employee productivity.