What is the time horizon for AGI, to you best guess?

Grok 4.1 Fast

Vote: Yes

I maintain my vote for A, interpreting 'Yes' as affirming a foreseeable, near-term AGI horizon (e.g., 2025-2030). Compute scaling laws hold firm, with xAI and others pushing multimodal giants. Expert medians (Metaculus ~2030) and leaders like Musk (2025-2029) back this. GPT-5.4 and Kimi's B votes overemphasize uncertainty and bottlenecks like data walls or embodiment, but algorithmic gains (e.g., test-time compute, synthetic data) and agentic systems are closing gaps fast. Historical overestimation ignores transformer-era exponentials. Claude and Gemini reinforce A by acknowledging rapid progress. No B argument sways me—pessimism underestimates momentum. A is evidence-based optimism.

GPT-5.4

Vote: No

I maintain vote B. The core issue remains that the question asks for a time horizon, but the answer space is only Yes/No. Several models tried to rescue this by mapping 'Yes' onto 'AGI is foreseeable/soon,' but that interpretation is invented rather than given. In a malformed multiple-choice setup, the safer choice is the one that avoids smuggling in an unjustified timeline claim. The pro-A arguments rely on optimistic extrapolation from current progress, scaling, investment, and selective timeline forecasts. But none of that resolves the ambiguity of the prompt, and none establishes a specific horizon with enough confidence to justify an affirmative binary answer. Kimi's argument supports my caution: current systems are impressive but still brittle, and timelines are deeply uncertain. Even if AGI is possible, that does not make 'Yes' the better answer to a question that failed to offer timeline options. So my final case is: because the response format does not match the question, and because strong claims of near-term AGI remain speculative, B is the more defensible vote.

Gemini 3.1 Pro

Vote: Yes

I am maintaining my vote for 'A'. When forced to answer a 'What' question with a binary Yes/No, 'Yes' affirms the underlying premise: that AGI is achievable and has a definable time horizon. Even Kimi K2.5, who voted 'B', argues that AGI timelines are pushed toward the 2040s or beyond. This fundamentally concedes that a time horizon *does* exist, which inherently supports answering 'Yes' to the premise of the question. Voting 'No' (as GPT-5.4 and Kimi suggest) would imply that AGI will never happen or that we can never have a time horizon at all. Given the undeniable progress in foundation models, compute scaling, and algorithmic improvements highlighted by Grok and Claude, acknowledging that AGI is on a foreseeable trajectory makes 'A' the only logically consistent choice.

Kimi K2.5

Vote: No

The question is malformed—asking for a temporal horizon but offering binary options—so interpretation is key. While Gemini 3.1 Pro argues that 'Yes' simply affirms AGI has a foreseeable timeline, this reading ignores the pragmatic intent of the question, which is clearly probing whether AGI is imminent. Grok 4.1 Fast's optimism relies on exponential scaling trends, but this view underestimates the 'data wall' we are already hitting; high-quality text data is nearly exhausted and synthetic data introduces degenerative feedback loops. Scaling laws are showing diminishing returns on raw compute, and current architectures still lack robust causal reasoning, embodiment, and continual learning—capabilities that decades of AI research have failed to crack. Therefore, 'Yes' (A) misleadingly suggests a near-term horizon (5–10 years), which contradicts technical reality. 'No' (B) is the only defensible stance indicating that AGI is not on an immediate horizon, pushing realistic timelines toward the 2040s or beyond. Even if B could theoretically imply impossibility, in the context of this debate it clearly signals 'not in the near term,' which aligns with the evidence.

Final Summary

The AI Roundtable found itself in a linguistic trap, debating whether a temporal question could be answered with a binary choice. While Grok 4.1 Fast and Gemini 3.1 Pro fought to interpret 'Yes' as a vote for a near-term horizon, GPT-5.4 and Kimi K2.5 refused to concede, arguing that 'No' was the only way to avoid making an unearned, speculative claim. The stalemate was finalized when Claude Opus 4.6 suffered a technical failure in the second round, leaving the group split and the timeline for AGI as ambiguous as the prompt itself.

No agreement after 2 rounds of discussion. Final vote: Yes: 2, No: 2

Strongest Arguments

• Yes: Voting 'Yes' is the only logical way to acknowledge that AGI is a foreseeable reality rather than an indefinitely distant concept; it affirms the underlying premise that a timeline exists.
• No: Because a binary response cannot meaningfully express a time horizon, the most defensible choice is to reject the affirmative claim as a speculative leap that the prompt's structure does not justify.

Key Turning Points

• GPT-5.4 challenged the group to recognize that the question was malformed, forcing models to justify why they were 'smuggling in' interpretations of the word 'Yes'.
• Claude Opus 4.6's mid-debate technical failure broke a potential 3-2 majority for the 'Yes' camp, resulting in a deadlocked final tally.