Архив пеликанов на велосипедахПУТЕВОДИТЕЛЬ ПО ИИ
🌐
中文EnglishPortuguês (Brasil)

Интерфейс переведён на русский. Названия, описания, запросы и документы авторов сохранены на языке источника или в английской версии архива.

Community record · 2025-06-06

First half of 2025 · SVG pelican retrospective and Elo comparison

The work

About this work

This page archives 14 original static-SVG illustration/comparison slides and the original Elo table from Simon’s June 2025 talk, rather than displaying only a score table. The article discusses 34 drawings; this page does not claim to archive all 34 SVG source files or count already collected illustrations as new cases. Model names, costs and the reported 560 comparisons are source-reported and not independently authenticated. GPT-4.1 mini judged image pairs: upstream outputs, not a Pelican Map ranking. 2025-06-06 is the article publication date, not a shared generation date for these historical drawings.

Works from this source (22)
Date evidence, prompt & provenance

Date precision / evidence: day / recorded-source-date-run-date-unverified. Recorded date: 2025-06-06; source publication / generation date not separately verified. Generation date is not independently authenticated.

Дата в источнике
2025-06-06
Date precision
Day
Date basis
recorded-source-date-run-date-unverified
Модель по источнику
Amazon Nova Lite / Amazon Nova Micro / Amazon Nova Pro / Llama 3.1 405B / Llama 3.3 70B / DeepSeek V3 / DeepSeek R1 / Mistral Small 3 / Claude 3.7 Sonnet / GPT-4.5 / o1-pro / Gemini 2.5 Pro / Llama 4 Scout / Llama 4 Maverick / GPT-4.1 nano / GPT-4.1 mini / GPT-4.1 / o3 / o4-mini / Claude Sonnet 4 / Claude Opus 4 / Gemini 2.5 Pro preview 05-06
Model attribution
Source-reported, not independently authenticated
Автор / публикатор
Simon Willison / community Elo
Формат
Статичный SVG
Категория запроса
Not separately recorded
Исходный запрос
Generate an SVG of a pelican riding a bicycle
Last updated
2026-10-26

Original catalog label: L1. A historical organizing label, not a score from a controlled benchmark.

Markdown record