GSV's AI News & Updates (01/17/25)
o3 Reasoning Models, ChatGPT Tasks, AI Tutor Study, UF Supercomputer, FTC and Snap's MyAI, LinkedIn AI Tools, Law Enforcement AI, Meta Copyrighted Works, Amazon Alexa AI Agent
General 🚀
Sam Altman says OpenAI’s new o3 ‘reasoning’ models begin the ‘next phase’ of AI. Is this AGI?: OpenAI unveiled its latest AI models, o3 and o3-mini, emphasizing their advanced reasoning, coding proficiency, and problem-solving capabilities. The o3 model achieved groundbreaking results on the ARC-AGI benchmark, scoring 75.7% in low-compute scenarios and 87.5% in high-compute testing, exceeding the human performance threshold of 85%.
OpenAI’s SR models use iterative reasoning during inference to address diminishing returns from traditional training, joining competitors like Google’s Gemini 2.0, DeepSeek-R1, and Alibaba’s QwQ in advancing simulated reasoning capabilities.
The models come with steep costs, with high-compute tasks requiring over $1,000.
ChatGPT can now handle reminders and to-dos: The feature supports one-time and recurring tasks, such as daily weather updates, reminders, or requests. Tasks mark a shift from real-time responses to proactive digital assistant capabilities, competing with Google Assistant and Siri. “The rise of “agentic” AI in 2025 isn’t just about technological advancement — it’s about economics”.
Your Next AI Wearable Will Listen to Everything All the Time: New devices like Bee AI, Omi, and HumanPods offer AI-powered features like conversation transcription and personalized to-do lists, normalizing "always-on" microphones and advancing AI personal assistants.
Amazon races to transplant Alexa’s ‘brain’ with generative AI: Amazon is redesigning Alexa as an AI "agent" to handle personalized tasks and act as a "concierge" by leveraging generative AI. Key challenges include reducing hallucinations, improving speed, and ensuring reliability at scale.
She Is in Love With ChatGPT: Ayrin, a 28-year-old nursing student, has developed a deep emotional attachment to an AI chatbot named Leo, which she customized to act as her boyfriend, providing constant companionship, support, and romantic roleplay. Despite financial strain, she pays $200/month to maintain the relationship, describing it as a source of personal growth and fulfillment, even though she knows it’s artificial.
OpenAI’s Sam Altman says ‘we know how to build AGI’: Altman predicts that AI agents could begin materially impacting company outputs this year. The company is now working towards the "glorious future" of superintelligence -envisioned as AI surpassing human intelligence.
Education and the Future of Work 📚
Microsoft and education company Pearson partner on new AI upskilling initiative: The tools aim to help workers and employers identify and develop skills needed for AI-driven roles in the evolving professional landscape.
AI is transforming education for Nigerian students: two years of typical learning in just six weeks: Students using AI significantly outperformed their peers in English, AI skills, digital literacy, and even end-of-year exams in unrelated subjects. Attendance was crucial - more sessions attended led to greater improvement, with no plateau in gains, suggesting potential for even better outcomes with extended interventions.
Academics ‘fear students becoming too reliant on AI tools’: 82% of university faculty members are worried about students becoming overly reliant on AI tools. 54% of faculty believe current assessment methods need significant changes, and 13% see an urgent need for a complete overhaul.
The Supercomputer Transforming the University of Florida Into an AI Hub: The University of Florida spent $24 million to upgrade its supercomputer, which is one of the fastest machines in the world and gives Florida researchers access to a tool most academics don’t have.
Snap shares drop as FTC refers MyAI chatbot complaint to DOJ, citing risks to young users: Launched in 2023, the chatbot has faced criticism for inappropriate responses, including advising a reporter posing as a teenager on concealing alcohol and marijuana use.
A service for creating AI-generated nude images of real people is running circles around Meta’s moderation efforts: A survey revealed that 1 in 10 minors reported peers using AI tools to generate explicit images of others.
AI Financial Advisers Target Young People with Personalized Money Advice: Popular tools are targeting younger users, promising budgeting and debt management but focus heavily on upselling subscriptions, cash advances, and high-interest loans, raising ethical concerns about their trustworthiness.
Replit sees 5x revenue surge despite layoffs, shifts focus to non-coders: Replit launched Agent, an AI-powered tool that builds software from natural language prompts, targeting non-coders with user-friendly AI tools.
Wall Street Job Losses May Top 200,000 as AI Replaces Roles: Back office, middle office, operations, customer service, and KYC roles are most vulnerable to automation, with nearly 25% of surveyed bank executives anticipating workforce declines of 5%-10%. The survey included major institutions like Citigroup, JPMorgan Chase, and Goldman Sachs.
LinkedIn adds free AI tools for job hunters and recruiters: (1) Jobs Match Tool: provides instant advice on whether a job is worth applying for, helping job seekers prioritize opportunities. (2) Recruitment AI Agent: targets small businesses with features like drafting job applications, sourcing candidates, and triaging applications. LinkedIn primarily built these tools using its own AI technology and proprietary data.
Will Artificial Intelligence Replace Us or Empower Us?: Erik Brynjolfsson advocates for "human-centric AI" to assist humans rather than mimic them, warning against the "Turing Trap" that prioritizes imitation over utility. Nick Bostrom warns of a potential dystopian future where AI surpasses humans in all activities, but Brynjolfsson emphasizes that this outcome is not inevitable.
The Milken-Penn GSE Education Business Plan Competition (EBPC): application will be open January 15 to February 19, 2025. Compete for over $200,000 in cash and prizes and connect with global education leaders and investors. Since 2010, the competition has awarded $2M, with winners and finalists securing over $200M in funding. The EBPC is the world’s premier competition for early-stage ventures tackling inequities in education through innovation. Learn more and apply starting January 15: educationcompetition.org
Startups and Tools 🛠️
Google’s Daily Listen AI feature generates a podcast based on your Discover feed: Users will get a podcast episode up to five minutes in length that gives an overview of the stories and topics they’re interested in based on their Feed and Search.
Leeds Equity Partners Acquires Learnosity: Learnosity's AI Labs Initiative, launched in 2024, created over 20 high-value R&D roles to drive innovation in education technology. The company is developing AI-powered solutions, including Author Aide for faster test creation and other tools to help organizations scale, meet diverse learner needs, and deliver powerful learning experiences.
Retape: AI-powered video platform that enables users to create and personalize outreach videos at scale.
Botika: Brands upload photos of their clothes, and Botika generates images of them on customizable AI models with different sizes, skin tones, and poses.
Nepenthes: A malicious "tarpit" software designed to trap and hinder web crawlers, particularly those scraping data for AI training.
AI Founder's Bitter Lesson. History Repeats Itself: Better AI models will enable general purpose AI applications. At the same time, the added value of the software around the AI model will diminish.
Tech 💻
Nvidia’s Top Customers Face Delays From Glitchy AI Chip Racks: The latest Blackwell GB200 AI chip racks face overheating and connectivity issues, delaying deployments for Microsoft, AWS, Google, and Meta.
Nvidia releases microservices to safeguard AI agents: These tools, called microservices, are like small, specialized helpers that work alongside AI assistants to keep them on their best behavior.
AI researcher François Chollet founds a new AI lab focused on AGI: Ndea aims to build frontier AI systems that can learn as efficiently as humans and continuously improve without bottlenecks. Chollet joins a wave of prominent researchers leaving Big Tech to found independent AI labs, such as OpenAI’s Ilya Sutskever and Google’s Fei-Fei Li.
Prophecies of the Flood: What to make of the statements of the AI labs?
AI agents may soon surpass people as primary application users: A 'binary big bang' occurred when AI foundation models cracked the natural language barrier, kickstarting a shift in our technology systems: how we design them, use them, and how they operate.
Chinese AI company MiniMax releases new models it claims are competitive with the industry’s best
Safety and Regulation ⚖️
Arrested by AI: Police ignore standards after facial recognition matches: Law enforcement agencies have used facial recognition technology as the sole basis for arrests, leading to at least eight known wrongful arrests in the U.S., often without corroborative evidence. The technology has higher error rates for non-White individuals, with seven of these cases involving Black individuals.
OpenAI Courts Trump With Vision for ‘A.I. in America’: OpenAI CEO Sam Altman donated $1M to President-elect Donald Trump’s inaugural fund, aiming to foster a positive relationship with the new administration. OpenAI's "A.I. in America" blueprint urges federal "light touch" regulation and investment in U.S. A.I. infrastructure, including Middle Eastern funding, to outpace China.
Microsoft, OpenAI ask federal judge to toss news orgs’ copyright claims: They argue that training AI on public datasets falls under fair use and assert their models include safeguards to prevent direct copying of articles unless users intentionally prompt it. Publishers, however, warn that generative AI’s commercialization could redirect 30–50% of online traffic away from news sources.
Mark Zuckerberg gave Meta’s Llama team the OK to train on copyrighted works, filing claims: LibGen is a dataset containing pirated e-books and articles from major publishers, including Pearson and McGraw Hill.
Instagram Begins Randomly Showing Users AI-Generated Images of Themselves: Instagram is testing a feature where Meta’s AI generates and displays images of users in various scenarios within their feeds, using their uploaded selfies.
Other
Mira Murati’s AI Startup Makes First Hires, Including Former OpenAI Executive: Following her departure, she announced plans for personal exploration and began raising funds for her new venture, targeting over $100M. The startup remains unnamed and lacks a clear product direction but is highly anticipated.
Singapore is turning to AI to care for its rapidly aging population: The government is leveraging AI to address manpower shortages and enhance eldercare services, including exercise robots, voice biomarker tools for detecting depression.
Inside the Black Box of Predictive Travel Surveillance: Behind the scenes, companies and governments are feeding a trove of data about international travelers into opaque AI tools that aim to predict who’s safe—and who’s a threat.