Chamath Palihapitiya: AI has entered a recursive self-improvement loop, will be extremely intense in the next 18 months
Chamath Palihapitiya stated that the AI singularity logic chain has been initiated: after humans build AGI, AGI excels in AI research, which in turn designs stronger AI, accelerating the loop. He pointed out that results and capabilities from various labs in recent weeks show that the industry is firmly within this loop. The next 18 months will be extremely intense, with recursive self-improvement rapidly increasing capabilities, and the marginal cost of all models approaching zero. This statement reinforces market expectations for capability leaps and cost collapses. Capital is accelerating towards leading labs with self-improvement feedback loops; beneficiaries are players that have established data and computing power flywheels, while traditional human-iteration R&D models are under pressure. Source: Public information
ABAB AI Insight
Chamath directly anchors the current progress to the classic singularity argument: from human-designed AGI to AGI dominating AI research, leading to exponentially recursive capabilities. This statement aligns with the theoretical frameworks of I.J. Good, Vernor Vinge, and later Nick Bostrom, but it is the first time a leading tech investor has explicitly announced that "we have entered the loop." In terms of capital flow, resources are shifting from "training the next model" to "having models participate in designing the next model." The motivation is that once AI research itself is automated, the human R&D bottleneck is removed, and the speed of capability enhancement will be determined by computing power and algorithm search efficiency, rather than talent supply. The judgment that marginal costs are approaching zero points to continuous exponential improvements in reasoning and training efficiency. Similar cases can be seen in the early transition from manual feature extraction to end-to-end learning in deep learning, as well as the current shift from manually designed architectures to neural architecture search and automated prompt optimization. The industry is in a transitional phase where "humans still dominate direction, but AI is deeply involved in acceleration." Essentially, this is a technological replacement. The mechanism is that when AI's performance on AI research tasks exceeds the human average, the dominance of the improvement loop shifts, and subsequent capability growth no longer linearly depends on human workload. ABAB News · Cognitive Laws 1. When AI starts improving AI, the time scale is permanently changed 2. The day marginal costs approach zero is the true starting point of capability explosion 3. The singularity is not a future event, but a loop that has been announced as entered.