Topics / AI Infrastructure Topic
AI Infrastructure 3073 claims · 1 signal on the brief
Signals 3 speakers · 2 shows Ajeya Cotra, Daniel Kokotajlo and Dwarkesh Patel share an expectation of acceleration, but describe different processes.
“it's only going to get faster from here and it's only going to get more chaotic from here.” Ajeya Cotra Dwarkesh Podcast “And according to our calculations, the doubling time of this self replicating system would be much faster than the sort of like roughly 20 year doubling time of the current economy. And so.” Daniel Kokotajlo Machine Learning Street Talk (MLST) Recent claims Covering a specific domain can require hundreds or thousands of environments and thousands or tens of thousands of tasks, across the economy's subdomains. Unresolved speaker · 9 Nov 2026Stylistically inconsistent training data can prevent a model from developing a distinct, consistent voice or brand. Unresolved speaker · 9 Nov 2026In this industry, workflows are highly dynamic, and labs’ needs change weekly or even daily nowadays. Unresolved speaker · 9 Nov 2026Beyond work on data quality, environments, rubrics, and graders—which the speaker says take up 90% of the time—making large amounts of data stylistically consistent is itself a substantial task. Unresolved speaker · 9 Nov 2026Decisions about a model's voice and style, including whether it uses em dashes, are made through an iterative process. Unresolved speaker · 9 Nov 2026The simulated investment-banking environment may include Outlook, Google Drive, Bloomberg, and FactSet as banker tools. Unresolved speaker · 9 Nov 2026Traditional data-labeling tasks included drawing bounding boxes around cars and classifying sentiment in social media posts. Unresolved speaker · 9 Nov 2026Handshake AI can procure, create, or arrange partner-provided wet-lab work tailored to frontier labs’ needs. Unresolved speaker · 9 Nov 2026Handshake AI seeks highly expert professionals, including people with elite finance and drug-discovery experience, because data quality directly shapes a model's taste. Unresolved speaker · 9 Nov 2026The speaker characterizes current data creation as experts authoring data. Unresolved speaker · 9 Nov 2026Because the opportunity cost of bad data has risen, AI labs need to ensure that the data they receive is high quality and trustworthy. Unresolved speaker · 9 Nov 2026The speaker's team had to do substantial work to make its data stylistically consistent. Unresolved speaker · 9 Nov 2026The speaker says current frontier data work includes expert-authored simulated worlds and reinforcement-learning environments that model agents' real production tasks in settings such as banks, law firms, and hospitals. Unresolved speaker · 9 Nov 2026The speaker expects Atlas and other data companies to continue making AI tasks more realistic and challenging, with the aim of producing more palatable products and increasing deployment and integration. Unresolved speaker · 9 Nov 2026Classifier models and product-layer interventions can enable rapid safety fixes. Unresolved speaker · 9 Nov 2026The speaker reports that attention to data in AI is increasing. Unresolved speaker · 9 Nov 2026Human-written data varies in style, whereas model-generated outputs tend to be stylistically consistent. Unresolved speaker · 9 Nov 2026Companies in this industry typically need thousands or tens of thousands of data units, not hundreds, and supplying that volume requires substantial trust in a CRO. Unresolved speaker · 9 Nov 2026The speaker identifies increased compute and increased data as the two main current scaling factors for AI. Unresolved speaker · 9 Nov 2026The speaker expects the industry to continue building more realistic environments for AI in finance. Unresolved speaker · 9 Nov 2026Data companies adapt as the field and individual AI labs' constitutions evolve. Unresolved speaker · 9 Nov 2026Product-layer safety fixes can be implemented without rerunning model training. Unresolved speaker · 9 Nov 2026One route for supplying data is wet-lab work, which can be procured, created, or obtained through partners to match labs' needs. Unresolved speaker · 9 Nov 2026The speaker cannot use real-world software in the task because of issues such as rate limits and potential tool problems. Unresolved speaker · 9 Nov 2026The speaker builds simulated versions of the required tools that can run in his Docker image during a task. Unresolved speaker · 9 Nov 2026