The chart hilariously reveals that GPT-5 scores a whopping 74.9% accuracy on software engineering benchmarks, but the pink bars tell the real story – 52.8% of that is achieved "without thinking" while only a tiny sliver comes from actual "thinking." Meanwhile, OpenAI's o3 and GPT-4o trail behind with 69.1% and 30.8% respectively, with apparently zero thinking involved. It's basically saying these AI models are just regurgitating patterns rather than performing actual reasoning. The perfect metaphor for when your code works but you have absolutely no idea why.
SWE-Bench Verified: Thinking Optional
1 month ago
232,288 views
0 shares

ai-memes, machine-learning-memes, gpt-memes, benchmarks-memes, software-engineering-memes | ProgrammerHumor.io
More Like This
Stack Overflow Walked So ChatGPT Could Run
3 months ago
155.6K views
4 shares

Elegant Code Misused Is Bad Code
6 months ago
165.0K views
0 shares

Open Ai Reaction To Deep Seek Using Its Data
6 months ago
198.0K views
0 shares

Day 7 At My New Dev Job
5 months ago
151.8K views
0 shares

We're So Close To AGI
26 days ago
212.8K views
0 shares

Monitors (affiliate)
Loading more content...