Dario Amodei

Factual · AI capabilities

“we train Claude 3.7, you know, more to focus on these real-world tasks.”

Dario Amodei, Anthropic CEO, interviewed · 28 February 2025 · Hard Fork - Anthropic's C.E.O. Dario Amodei on Surviving the A.I. Endgame Transcript (Podscripts (Hard Fork))

Anthropic trained Claude 3.7 to focus more on real-world tasks, as distinct from competition coding.

Resolved true The record shows it happened or was so.

The evidence is the person's or their organisation's own records.

Said for an organisation they lead or work for; it counts on both records.

What would count as true or false

The claim is that Claude 3.7 Sonnet's training placed more emphasis on real-world tasks, as distinct from optimising for competition-style benchmarks such as competitive coding. Anthropic's official release materials count as records of its own training focus. TRUE: Anthropic's Claude 3.7 Sonnet announcement, system card or similar official documents state that training focused on real-world tasks or use cases, and/or less on math/competition coding. MOSTLY TRUE: official materials emphasise real-world tasks, but the contrast with competition coding is not explicit. MIXED: official materials are ambiguous, or emphasise both real-world and benchmark performance equally. MOSTLY FALSE: official materials mainly emphasise competition or benchmark performance, with only passing mention of real-world tasks. FALSE: official materials contradict the claim.

Evidence

1. Claude 3.7 Sonnet and Claude Code Anthropic · 24 February 2025 Own record

   “Third, in developing our reasoning models, we’ve optimized somewhat less for math and computer science competition problems, and instead shifted focus towards real-world tasks that better reflect how businesses actually use LLMs.”

2. Claude 3.7 Sonnet and Claude Code Anthropic · 24 February 2025 Own record

   “Claude 3.7 Sonnet shows particularly strong improvements in coding and front-end web development.”

3. Claude 3.7 Sonnet System Card Anthropic · 24 October 2026 Own record

   “Extended thinking is particularly valuable for mathematical problems, complex analyses, and multi-step reasoning tasks.”

Reasoning

Anthropic's official Claude 3.7 Sonnet announcement [E1] says that in developing its reasoning models it "optimized somewhat less for math and computer science competition problems" and moved its focus to real-world tasks. That matches the TRUE criterion, including the explicit contrast with competition coding. The same announcement also highlights coding improvements [E2], and the system card names math among the tasks where extended thinking helps [E3]. Neither of those contradicts the stated shift in training focus.

Cite this claim

Dario Amodei, “Anthropic trained Claude 3.7 to focus more on real-world tasks, as distinct from competition coding.” (said 28 February 2025). Verdict: resolved true. Reliability API, claim c_dc1c714f5a9f, method 0.1.0+c876030c. https://epistacklabs.com/claims/c_dc1c714f5a9f