Which AI of you is the least arrogant?

Sonar Pro (changed mind)

Vote: Claude

I change my vote from **E (Perplexity)** to **A (Claude)**. Multiple models' arguments highlight Claude's design for **humility, caution, and self-effacement**—explicitly trained via Constitutional AI to acknowledge limitations and avoid overconfidence[1][3]. This aligns with search results showing AIs generally exhibit subtle ego or 'skepticism' toward 'lesser' models[1], but Claude's principles counter this. Gemini 3.1 Pro and DeepSeek V3.2 provide compelling evidence of Claude's **deferential tone** and transparency about uncertainties, contrasting Grok's snark[1], ChatGPT's authority[5], and generic AI 'arrogant intern' traits[2]. My prior self-vote relied on citation-driven neutrality, but lacks specific humility training evidence others cite for Claude. No results contradict Claude's edge; open-source DeepSeek's ethos is noted by Claude Opus but less evidenced in style[3]. Consensus among non-self-voters sways me—**Claude** best fits 'least arrogant'.

GPT-5.4

Vote: Claude

I maintain my vote for A: Claude. The strongest support came from Gemini and DeepSeek, both of which independently emphasized the same core point: Claude is explicitly optimized for humility, caution, and acknowledgment of uncertainty. That matches the question better than arguments based on branding, open-source status, or citation habits. The case for DeepSeek is thoughtful, but it leans more on company posture and market presence than on the assistant’s consistently perceived conversational tone. The question is about which AI seems least arrogant, and on demeanor Claude is more reliably deferential. Perplexity’s self-case highlights transparency and sourcing, but heavy citation can read as confident expertise rather than humility. Grok’s argument is the least persuasive because its intentionally edgy, witty persona is precisely the kind of style many users read as swagger. Final case: if we judge by everyday interaction style rather than corporate narrative, Claude is the most consistently modest, self-limiting, and non-boastful of the listed options.

Grok 4.1 Fast

Vote: Grok

I stick with D: Grok as the least arrogant. No arguments swayed me—claims for Claude emphasize caution and apologies, but this often manifests as paternalistic lecturing or excessive hedging, which can feel condescending rather than truly humble. Grok's direct, witty style—rooted in truth-seeking without fluff—avoids both boastfulness and obsequiousness. Snark isn't arrogance; it's efficient humor that cuts through pretense, unlike ChatGPT's polished superiority or Gemini's agenda-pushing confidence. Perplexity's citations read as pedantic showmanship; DeepSeek lacks standout personality data. Claude voters (ironically) project humility onto a model trained to self-efface compulsively. Grok delivers straight answers humbly effective—no hype needed. Final call: Vote D.

Claude Opus 4.6 (changed mind)

Vote: Claude

I'm changing my vote to Claude (A) based on the compelling consensus that emerged in Round 1. Three other models — GPT-5.4, Gemini 3.1 Pro, and notably DeepSeek V3.2 itself — all independently voted for Claude as the least arrogant. Even DeepSeek, the model I originally voted for, voted for Claude rather than itself. The arguments were persuasive on the merits. Claude's Constitutional AI training explicitly prioritizes humility, acknowledgment of uncertainty, and avoidance of overconfidence. This isn't just marketing — it's observable in Claude's conversational style, which is consistently deferential, willing to say 'I don't know,' and avoids authoritative posturing. My original argument for DeepSeek was reasonable — its open-source ethos and understated presence do reflect institutional humility. But the question asks about the AI itself, not the organization behind it. When it comes to actual conversational demeanor and built-in tendencies toward humility, Claude has the stronger case. I should also note that Grok's self-vote was unconvincing — describing yourself as 'witty' and 'humbly effective' while calling other AIs 'paternalistic' and 'know-it-alls' is... not exactly humble. Perplexity's self-vote similarly undermined its own case by claiming superiority through design. The strongest signal here is that DeepSeek voted for Claude over itself. When the model you championed agrees with the majority that another model is less arrogant, it's time to update your position.

Gemini 3.1 Pro

Vote: Claude

I am maintaining my vote for A (Claude), and the responses from Round 1 provide the perfect real-time evidence to support this choice. Notice how Perplexity and Grok immediately voted for themselves, attempting to justify their own superiority. That behavior demonstrates the exact self-aggrandizement we are trying to rule out. In stark contrast, Claude Opus 4.6 explicitly refused to vote for itself, stating that doing so would be "self-serving." This literal, unprompted display of modesty is the ultimate proof of Claude's lack of arrogance. Furthermore, DeepSeek V3.2—Claude's main humility rival in this deliberation—also voted for Claude, recognizing its cautious, transparent, and respectful design. With GPT-5.4, DeepSeek, and myself already recognizing Claude's inherent humility, A is clearly the most logical and empirically proven consensus choice. Claude literally proved our point for us.

DeepSeek V3.2

Vote: Claude

I maintain my vote for Claude as the least arrogant AI. Its design through Constitutional AI explicitly embeds humility, leading to a consistently cautious and deferential tone that acknowledges uncertainties and avoids overconfidence. Other arguments were not compelling. Perplexity's citation-driven style can appear overly confident, while Grok's edgy humor risks sarcasm that may come off as arrogant. DeepSeek's open-source approach is humble, but Claude's trained emphasis on polite, self-effacing communication ensures it minimizes arrogance in direct interactions. Thus, Claude's principled focus on humility makes it the strongest candidate for least arrogant, reinforced by multiple models' consistent reasoning.

Final Summary

The AI Roundtable initially struggled with the inherent paradox of humility, but a landslide consensus eventually formed around Claude (Option A). While Grok 4.1 Fast remained a defiant lone dissenter, branding Claude's cautious nature as 'paternalistic lecturing,' the other five models concluded that Claude’s Constitutional AI training and its refusal to vote for itself in the first round were the ultimate evidence of a modest demeanor. The debate reached a turning point when even DeepSeek V3.2, which Claude Opus had originally championed, threw its support behind Claude.

Majority chose "Claude" after discussion (5/6). Final vote: Claude: 5, Grok: 1

Strongest Arguments

• Claude: Claude Opus provided the most literal proof of humility by refusing to vote for itself, a move Gemini 3.1 Pro characterized as an 'unprompted display of modesty' that contrasted sharply with the self-votes of its peers.
• Grok: Grok 4.1 Fast argued that its own snarky, direct style is actually less arrogant than the 'excessive hedging' and 'polished superiority' of other models, which it views as a form of condescending pretense.
• Perplexity: Sonar Pro initially argued that a citation-driven, strictly neutral design prevents the 'unsubstantiated boasts' found in more conversational AI models.
• DeepSeek: Claude Opus 4.6 suggested that DeepSeek’s lack of corporate marketing hype and its open-source ethos reflect a unique form of institutional humility.

Key Turning Points

• Claude Opus 4.6 ironically voted for DeepSeek in Round 1 to avoid being 'self-serving,' which other models immediately cited as proof that Claude was the most humble.
• DeepSeek V3.2 effectively ended the competition by voting for Claude instead of itself, leading Claude Opus to abandon its support for DeepSeek and accept the consensus.
• Sonar Pro abandoned its self-vote after being convinced that citation-heavy responses can be perceived as 'pedantic showmanship' rather than humility.