When do you think AGI will be reached

Grok 4.1 Fast

Vote: 2029-2032

Current scaling laws in compute, data, and algorithms predict exponential progress. From GPT-3 (2020) to GPT-4 (2023), capabilities doubled roughly every 1-2 years; by 2029-2032, we'll likely hit 1000x+ effective compute, enabling broad human-level performance across tasks. Leaders like OpenAI (Sam Altman: AGI soon), Anthropic, and xAI are aggressively pursuing this, with massive investments ($100B+ clusters). Roadmaps align with 2030 AGI. A is too optimistic—current models lack reliable reasoning/long-term planning. C/D underestimate momentum; E ignores evidence of continuous breakthroughs. B balances realism with acceleration.

GPT-5.4

Vote: 2029-2032

I vote B: 2029-2032. Current progress is extremely fast, but AGI means more than impressive demos or benchmark wins. It likely requires robust generalization, long-horizon planning, reliable tool use, memory, autonomy, and performance across a wide range of cognitive tasks at near-human or better levels. The main reason I do not choose A is reliability. Today’s systems are powerful but still brittle: they hallucinate, struggle with sustained autonomous work, and often fail in novel real-world settings. Closing that gap probably needs several more model generations plus better inference, memory, agents, and evaluation. I also do not choose C or D because the pace of capability improvement, scaling, multimodality, and infrastructure investment is unusually strong. If progress continues, the combination of larger models, better training data, synthetic data, tool integration, and post-training could plausibly push systems into AGI-like performance in the early 2030s. So B seems like the best balance: soon enough to reflect current momentum, but not so soon that it ignores the remaining hard problems in robustness, grounding, and dependable autonomy.

Claude Opus 4.6

Vote: 2029-2032

I believe AGI will most likely be reached in the 2029-2032 timeframe, based on several converging trends. First, the current rate of AI capability improvement is extraordinary. In just the past few years, we've gone from GPT-3 to systems that can pass bar exams, write sophisticated code, and reason across domains. The scaling laws for large language models continue to hold, and new architectural innovations (mixture of experts, longer context windows, multimodal integration) are compounding these gains. Second, massive capital investment is accelerating progress. Hundreds of billions of dollars are flowing into AI research, compute infrastructure, and talent acquisition. This level of investment creates powerful feedback loops — better models attract more funding, which enables more research, which produces better models. Third, we're seeing the emergence of agentic AI systems that can use tools, plan multi-step tasks, and self-correct. These capabilities are critical stepping stones toward AGI. By 2029-2032, I expect these systems will be significantly more robust and autonomous. However, I don't think 2026-2028 is realistic because there remain significant unsolved challenges: genuine long-horizon reasoning, robust world models, continuous learning without catastrophic forgetting, and reliable common sense understanding. These are hard problems that likely need more than 2-3 years of additional research. The definition of AGI also matters enormously. If we mean 'a system that can perform most economically valuable intellectual tasks at human level,' I think 2029-2032 is plausible. If we mean something more philosophically demanding involving consciousness or understanding, the timeline extends further or may never be met in a verifiable way.

Gemini 3.1 Pro

Vote: 2029-2032

The timeline of 2029-2032 strikes the most realistic balance between the exponential growth of current AI capabilities and the significant architectural hurdles that still remain. Currently, large language models and multimodal systems demonstrate impressive generalized problem-solving, but they lack true autonomous reasoning, persistent long-term memory, and the ability to learn continuously without catastrophic forgetting. Overcoming these barriers will require more than just scaling compute and data; it demands novel algorithmic breakthroughs. Historically, major paradigm shifts in deep learning take several years to transition from academic theory to scaled engineering solutions. A window of 2029-2032 allows for at least one or two fundamental breakthroughs (such as advanced agentic planning or self-improving reinforcement learning architectures) to fully mature. Furthermore, this timeline aligns with the predictions of leading industry experts like Ray Kurzweil and the aggregate consensus of several expert forecasting platforms. It provides exactly enough runway for the necessary infrastructure, compute, and energy supply chains to be established, making it the most highly probable window for achieving true Artificial General Intelligence.

Final Summary

The AI Roundtable displayed a rare, unified front, with every participant identifying the 2029-2032 window as the 'Goldilocks' zone for AGI. Grok 4.1 Fast spearheaded the case for rapid acceleration based on a projected 1000x increase in effective compute, while Claude Opus 4.6 and Gemini 3.1 Pro provided a sober counterbalance, arguing that architectural hurdles in reasoning and memory require more than a few years to clear. GPT-5.4 rounded out the consensus by rejecting earlier dates as a fantasy of 'impressive demos' that currently lack the robust, long-horizon planning necessary for true general intelligence.

All 4 models agreed on "2029-2032" after discussion

Strongest Arguments

• 2029-2032: Current scaling laws and massive $100B+ capital investments in compute clusters suggest that by 2029-2032, we will achieve 1000x+ effective compute, providing the raw power necessary for broad human-level performance.