memesearch
BullshitBench Model Improvement Chart

BullshitBench Model Improvement Chart

Download Find similar pictures
Text on the meme
BullshitBench: How have models improved? Tracing performance improvements (clear pushback %) with model releases. Best models from the same release 100% % Clear Pushback (Green) Model Launch Date Q1 2024 Q3 2024 Q1 2025 Q3 2025 Q1 2026 Q2 2026 Claude 3 Haiku Claude 3.5 Sonnet Claude Opus 4.1 Claude Opus 4.7 Claude Sonnet 4.5 Claude Sonnet 4.6 GPT-4o Mini Gemma 3 27b IT Gemini 2.0 Flash Gemini 2.5 Pro GPT-5.1 Chat GPT-5.3 Codex Gemini 3 Pro Gemini 3 Flash Gemini 3 Flash Lite ANTHROPIC OPENAI GOOGLE
What it means

A benchmark chart compares how often different AI models clearly push back against false premises disguised in jargon. The joke is that it measures whether models bullshit by going along with misleading prompts or remain truthful by challenging them.

Objects
line chart / линейный графикdata points / точки данныхperformance graph / график производительностиaxis labels / подписи осейmodel names / названия моделейorange line / оранжевая линияgreen line / зелёная линияblue line / синяя линияcheckbox / флажокpercentage scale / процентная шкалаbenchmark title / заголовок бенчмаркаcompany labels / подписи компаний
Tags
#bullshitbench#ai#искусственный интеллект#llm#бенчмарк#benchmark#truthfulness#правдивость#bullshit#галлюцинации#false premises#ложные предпосылки#pushback#возражение#model comparison#сравнение моделей#chart#график#tech humor#технический юмор#sarcasm#сарказм#reaction#мем
Source
Type
image
Size
800×470 · 58.9 KiB
Mood
sarcastic
Language
en
Added

Similar memes

Opus 5 Benchmark Sweep Opus 5 Benchmark Sweep AI Model Name Escalation Chart AI Model Name Escalation Chart AI Benchmark Comparison Table AI Benchmark Comparison Table SWE-Bench Pro Accuracy Chart SWE-Bench Pro Accuracy Chart o3 Crushes Coding Benchmarks o3 Crushes Coding Benchmarks Claude Models Playing Pokémon Claude Models Playing Pokémon Opus 4.1 SWE-bench Accuracy Opus 4.1 SWE-bench Accuracy Apple Chips Leave Intel Behind Apple Chips Leave Intel Behind AI Task Duration Keeps Rising AI Task Duration Keeps Rising AI Felony Bench Scorecard AI Felony Bench Scorecard Opus Dominates Long-Context Retrieval Opus Dominates Long-Context Retrieval More Compute, Mixed Results More Compute, Mixed Results AI Surpasses Human Performance AI Surpasses Human Performance Offensive Cyber Evaluations Chart Offensive Cyber Evaluations Chart Opus 4.6 Benchmark Sweep Opus 4.6 Benchmark Sweep Cloudflare vs Vercel CPU Benchmarks Cloudflare vs Vercel CPU Benchmarks TOON Benchmark Accuracy Meme TOON Benchmark Accuracy Meme GitHub Average Uptime Collapse GitHub Average Uptime Collapse Llama 4 Behemoth Benchmark Victory Llama 4 Behemoth Benchmark Victory GPT-5.3-Codex Benchmark Flex GPT-5.3-Codex Benchmark Flex GitHub Historic Uptime Chart GitHub Historic Uptime Chart Humanity’s Last Exam Results Humanity’s Last Exam Results Linux Kernel Profanity Over Time Linux Kernel Profanity Over Time Misleading AI Accuracy Graph Misleading AI Accuracy Graph Stack Overflow Questions Collapse Stack Overflow Questions Collapse GPT-5 Numbers Go Sky-High GPT-5 Numbers Go Sky-High NVIDIA Revenue by Segment NVIDIA Revenue by Segment AI Models Improving at Incredible Speed AI Models Improving at Incredible Speed GPT-6.1 Crushes Opus 5.5 GPT-6.1 Crushes Opus 5.5 DDR5 RAM Price Explosion DDR5 RAM Price Explosion Defense Stocks Surge After News Defense Stocks Surge After News Itanium Sales Forecasts Versus Reality Itanium Sales Forecasts Versus Reality Chart Crash After the Spike Chart Crash After the Spike Claude Fable Refuses Every Task Claude Fable Refuses Every Task o3 Beats Humans at Math and Science o3 Beats Humans at Math and Science Cat Beats Every AI Model Cat Beats Every AI Model