What AI leaders and researchers are saying.
On-the-record interviews, talks, podcasts and essays, summarised with attribution. Quotes are checked against the transcript or the recording before publication; opinions and predictions are labelled as such.
People in the news
Latest statements
- Jeremy Howard Explains SolveIt: A Tool for Human-Centered AI Collaboration
Jeremy Howard, co-founder of Answer.AI and fast.ai, announced the public launch of SolveIt, a tool designed to enhance human understanding and collaboration with AI, rather than automate human tasks. He argues that AI should augment human capabilities, not replace them.
- Ben Recht says AI doomsayers' forecasts are design documents for tech giants
Ben Recht argues that AI doomsday forecasts are not genuine predictions but rather design documents reflecting the desires of major tech companies like OpenAI and Anthropic.
- Jack Clark Explains How AI Systems Can Improve Themselves Through Subcontracting
Jack Clark, in his Import AI newsletter, discusses the development of FrontierML, a benchmark that measures how well AI agents can use capital to improve their performance on machine learning tasks. He highlights that AI systems are becoming adept at subcontracting work to specialists to enhance their effectiveness.
- Ben Recht Explains FAA Ground Stop Was Not AI-Related
AI researcher Ben Recht argues that the recent FAA nationwide ground stop was caused by an outdated network infrastructure and human error, not artificial intelligence. He criticizes the lack of transparency and accountability in the system.
- Greg Brockman explains OpenAI's new cybersecurity products Daybreak Red Team and Daybreak Defense Factory
Greg Brockman, Co-founder and President of OpenAI, introduced two new cybersecurity products, Daybreak Red Team and Daybreak Defense Factory, aimed at enhancing vulnerability discovery and defense.
- Jason Wei Argues "Who You Know" is Becoming More Important Than "What You Can Do" in AI
AI researcher Jason Wei argues that in the last five years, the importance of personal connections ("who you know") in the AI field has significantly increased, while technical skills ("what you can do") have become relatively less important. He cites the commoditization of intelligence, increased resource requirements for research, and the shift to closed-source models as key drivers.
- Ethan Mollick Says New AI Models Are 'Almost' AGI, Dubbed 'Jagged AGI'
Ethan Mollick, a professor at the Wharton School, argues that new AI models like Anthropic's Fable 5.2 are approaching Artificial General Intelligence (AGI) but are not quite there yet, referring to them as 'Jagged AGI'. He notes that while these models excel at many intellectual tasks, they still exhibit significant gaps in judgment and consistency.
- Boaz Barak: AI's economic impact hinges on augmentation, not just automation
Professor Boaz Barak argues that the economic benefits of AI are maximized when AI augments human capabilities rather than simply automating tasks. He emphasizes the need for "centaur benchmarks" that measure AI's ability to assist humans.
- Nathan Lambert Explains Open Model Inference Market Growth
AI researcher Nathan Lambert argues that while measurements of open model adoption are imprecise, data from inference startups suggests significant near-exponential growth in the market for open model inference.
- Terence Tao: AI race turns beautiful mathematics into ‘toxic waste’
Mathematician Terence Tao argues that the intense race to develop AI is degrading the field of mathematics, turning it into 'toxic waste'. He calls for a 'Maths 2.0' world to preserve the integrity of mathematical research.
- Andrej Karpathy Explains His New YouTube Clone, HackerTube
Andrej Karpathy has launched HackerTube, a personal YouTube clone, as a response to the perceived decline in content quality on YouTube. The project leverages AI to curate video recommendations based on Hacker News comments.
- Gary Marcus says Anthropic's S-1 filing omits key Q3 data
Gary Marcus, writing in his Substack newsletter, points out that Anthropic's draft S-1 filing appears to omit crucial data regarding Q3 performance, specifically concerning demand and the impact of alternative models.
- Gary Marcus Says Anthropic May Be Close to IPO
Gary Marcus reports a rumor that Anthropic is preparing for an Initial Public Offering (IPO), suggesting the timing might be challenging.
- Jason Wei Argues AI Will Not Automate Radiologists Soon
AI researcher Jason Wei argues that despite advancements in AI, radiologists will not be automated away in the near future due to the complex, multi-task nature of radiology and elastic demand for the service. He predicts a future where AI acts as a triage tool, increasing the overall volume of scans rather than replacing human experts.
- Ben Recht Explains Why Predictive Models Are Never Clean
Ben Recht argues that the statistical community's commitment to data models has led to irrelevant theory and questionable conclusions, hindering progress on important problems. He advocates for a broader adoption of algorithmic modeling for more accurate and informative predictions.
- Arvind Narayanan Argues AI Development Should Prioritize Human Augmentation Over Automation
Arvind Narayanan, a professor at Princeton University, argues that the dominant AI development path focused on total automation is misguided. He advocates for an alternative direction prioritizing human augmentation and collaboration.
- Jack Clark Says AI's Recursive Self-Improvement Timeline May Be Accelerating
In his Import AI newsletter, Jack Clark suggests that AI systems may be approaching recursive self-improvement (RSI) sooner than previously anticipated, potentially by 2027, due to advancements in AI research taste.
- Nathan Lambert: AI progress will accelerate, but not towards general superintelligence
AI researcher Nathan Lambert argues that while AI capabilities will see rapid acceleration due to engineering and infra improvements, this progress will not lead to general superintelligence in the near term. He emphasizes that current advancements focus on optimizing existing models rather than fundamental breakthroughs in AI's nature.
- Ben Recht Argues Uncertainty Quantification is Narrowly Defined
Ben Recht argues that uncertainty quantification in forecasting is too narrowly defined, primarily relying on error bars and prediction intervals. He suggests a need for more imaginative approaches to better prepare for forecast failures.
- Gary Marcus Argues for AI Regulation at NYC Council Hearing
Gary Marcus, a scientist and author, testified before the New York City Council, advocating for a regulatory regime for AI similar to the FDA for drugs. He argued that AI developers should be required to demonstrate that their products' benefits outweigh the risks before gaining market access.
- Terence Tao Explains AIM: An Invitation to Explore Mathematics Together
Mathematician Terence Tao, in a blog post, introduces AIM, a community-led initiative for mathematical exploration, aiming to leverage AI while prioritizing human collaboration and understanding. The project seeks to foster a collaborative environment for mathematicians, especially students and early-career researchers, to engage with open problems.