--- title: "This Week in AI | More Powerful AI Is Coming — Will the Benefits Be Shared?" author: "Unique Research" sourcePublication: "Unique Research (FFCAP English library)" originalPublishedAt: "2025-01-09T16:56:15+08:00" canonical: "https://ffcap.cn/en/research/20250109a070ct00" source: "https://ffcap.cn/en/research/20250109a070ct00" language: "en" cnSourceUrl: "https://view.inews.qq.com/a/20250109A070CT00" --- # This Week in AI | More Powerful AI Is Coming — Will the Benefits Be Shared? _Unique Research · This week in AI · January 9, 2025_ From OpenAI’s “12 Days of Shipmas” to DeepSeek’s Christmas Day model release, progress in AI flashes by almost too quickly to track. That pace has not slowed. On Sunday, OpenAI CEO Sam Altman wrote on his personal blog that OpenAI already knows how to build AGI — and has begun shifting its aim toward superintelligence. AGI is a fuzzy term; OpenAI’s own definition is “highly autonomous systems that outperform humans at most economically valuable work.” Superintelligence, in Altman’s framing, is a step beyond AGI — tools that could “vastly accelerate” innovation beyond what humans alone can achieve. “OpenAI still believes that iteratively putting powerful tools in people’s hands leads to great, broadly distributed outcomes,” Altman wrote. Altman shares an optimistic outlook with Anthropic CEO Dario Amodei: that AGI and superintelligence will bring wealth and prosperity for everyone. But even if AGI and superintelligence are feasible without new scientific breakthroughs, how do we ensure the gains reach everyone? Earlier this month, Wharton professor Ethan Mollick shared concerning research on X. Researchers from the National University of Singapore, the University of Rochester, and Tsinghua University studied how ChatGPT affected freelancers in different labor markets. They identified an economic “AI tipping point” by occupation. Before the tipping point, AI raised freelancers’ income — web developers, for example, saw roughly a **65%** increase. After the tipping point, AI began displacing freelancers; translators saw income fall by about **30%**. The study suggests that once AI starts replacing a profession, there is no going back. If more powerful AI is coming, that should concern all of us. Altman wrote that he is “pretty sure” that in an age of AGI and superintelligence, “everyone” will see how critical it is to “maximize broad benefit and empowerment.” But what if he is wrong — and only corporations capture the upside? The result would not be a better world, but more inequality. That would be a deeply disappointing outcome for AI. ## News **Silicon Valley shrugs off doomsday warnings.** For years, technologists warned that AI could cause catastrophic harm. In 2024 those voices were largely drowned out. **OpenAI is losing money on Pro.** Altman said the $200/month ChatGPT Pro plan is currently loss-making because usage exceeded expectations. **Record generative-AI funding.** Investment in generative AI — apps, tools, and services that produce text, images, video, voice, music, and more — hit new highs last year. **Microsoft doubles down on data centers.** Microsoft allocated **$80 billion** in fiscal 2025 to build data centers for AI workloads. **Grok 3 missed its window.** xAI’s next-generation model Grok 3 did not ship on schedule — part of a wider pattern of flagship models missing announced release windows. ## Paper of the week AI can make many mistakes — and still sharply lift expert productivity. That is the finding of a University of Chicago and MIT team. In a new study, investors who used OpenAI’s GPT-4o to summarize earnings calls earned higher returns than those who did not. Researchers recruited investors and had GPT-4o produce AI summaries tuned to each person’s expertise: seasoned investors got more professional notes; novices got simpler ones. After using GPT-4o, experienced investors’ one-year returns rose **9.6%**; novices’ rose **1.7%**. For human–AI collaboration, that is a solid result. ## Model of the week Prime Intellect, a startup building infrastructure for decentralized AI training, released a model said to help detect pathogens. **METAGENE-1** was trained on a dataset of more than **1.5 trillion** DNA and RNA base pairs sequenced from human wastewater samples, created with the University of Southern California and SecureBio’s Nucleic Acid Observatory. It is aimed at metagenomics uses such as studying organisms. “METAGENE-1 achieves state-of-the-art performance on various genomic benchmarks and new evaluations focused on human pathogen detection,” Prime Intellect wrote on X. “After pretraining, the model is designed to assist work in biosurveillance, outbreak monitoring, and pathogen detection.” ## Other In response to legal action by major music publishers, Anthropic has agreed to put guardrails in place so Claude will not share copyrighted song lyrics. Universal Music Group, Concord Music Group, and ABKCO sued Anthropic in 2023, alleging unauthorized training on lyrics from at least 500 songs. The suit is unresolved, but Anthropic has agreed to stop Claude from providing those publishers’ lyrics and from creating new song lyrics based on copyrighted material. “We still look forward to proving that, under existing copyright law, using potentially copyrighted material in training generative AI models is essential fair use,” Anthropic said in a statement. — END — --- Editorial English edition published on UniqueCapital: https://ffcap.cn/en/research/20250109a070ct00 On-site reading page: https://ffcap.cn/en/research/20250109a070ct00