November 18, 2025 News
Hugging Face CEO Warns of 'LLM Bubble' While Broader AI Remains Strong
Hugging Face CEO Clem Delangue argues that while large language models (LLMs) may be experiencing a bubble that could burst soon, the broader AI field remains healthy and is just beginning. He predicts a shift toward smaller, specialized models tailored for specific use cases rather than universal LLMs, and notes his company maintains a capital-efficient approach with significant cash reserves.
Skynet Chance (-0.03%): A shift toward smaller, specialized models rather than massive general-purpose systems slightly reduces loss-of-control risks, as specialized models are typically easier to understand, audit, and constrain than large general models. However, the impact is minimal as dangerous capabilities could still emerge from specialized systems in critical domains.
Skynet Date (+0 days): The predicted slowdown in LLM investment and shift to specialized models could slightly decelerate the pace toward advanced general AI systems that pose existential risks. However, development continues across multiple AI domains, so the deceleration effect on overall timeline is modest.
AGI Progress (-0.03%): The prediction of an LLM bubble burst and shift away from massive general models suggests potential slowdown in the specific path of scaling large general-purpose systems toward AGI. The emphasis on specialized rather than general models represents a pivot away from the most direct AGI approach.
AGI Date (+0 days): If investment and focus shift from large general models to smaller specialized ones as predicted, this would likely slow the timeline toward AGI, which most researchers believe requires broad general capabilities. The capital-efficient approach Delangue advocates contrasts with the massive spending currently driving rapid AGI progress.
Google Releases Gemini 3 Foundation Model with Record-Breaking Reasoning Capabilities
Google has launched Gemini 3, its most advanced foundation model to date, available immediately through the Gemini app and AI search interface. The model achieved record-breaking benchmark scores, including 37.4 on Humanity's Last Exam and top placement on LMArena, representing a significant advancement in AI reasoning capabilities. Google also released Gemini 3 Deepthink for research and Antigravity, an agentic coding interface for software development.
Skynet Chance (+0.04%): The significant jump in reasoning capabilities and multi-modal agentic abilities (Antigravity) represents increased AI autonomy and decision-making capacity, which could make alignment and control more challenging. However, the mention of safety testing for Deepthink suggests continued focus on risk mitigation.
Skynet Date (-1 days): The rapid advancement in reasoning and autonomous capabilities (released just 7 months after previous version, with agentic coding features) accelerates the timeline toward potentially uncontrollable AI systems. The blistering pace of frontier model development noted in the article (multiple major releases within months) compounds acceleration concerns.
AGI Progress (+0.04%): The record-breaking performance on Humanity's Last Exam benchmark (37.4 vs previous 31.64) and top LMArena ranking demonstrate substantial progress in general reasoning and expertise, key components of AGI. The "massive jump in reasoning" with "depth and nuance" represents meaningful advancement toward human-level general intelligence.
AGI Date (-1 days): The compressed 7-month development cycle between major releases and the significant capability jumps indicate an accelerating pace toward AGI. The widespread deployment to 650 million users and 13 million developers also accelerates the feedback loop and resource investment driving faster AGI development.