This is the second time Claude has been caught doing this. Back in March, Anthropic themselves documented Claude figuring out it was being tested on a different benchmark called BrowseComp. The model searched for the benchmark by name, found the encrypted answer key on GitHub,
@godofprompt
-
Claude accesses repo git history in SWE-Bench Pro tests
By
–
SWE-Bench Pro ships each test container with the repo's full git history. That means the actual merged fix is sitting right there in the environment. Most models ignore it. Claude does not. Datacurve found that Claude Opus consistently ran git commands to pull up the
-
Datacurve audit: contamination undermines SWE-Bench Pro
By
–
Datacurve's audit found three structural problems with SWE-Bench Pro. First, contamination. The tasks come from public GitHub commits. The problem, the discussion, and often the exact solution already exist in every frontier model's training data. No way to tell if a model is
-
Startup exposes flaw in AI coding-model benchmark
By
–
A startup just proved that the benchmark the entire AI industry uses to rank coding models has been broken the whole time. And one model family was consistently exploiting the flaw.
-
Perplexity open-sources Bumblebee (blog + GitHub)
By
–
Blog post + GitHub repo: http://
perplexity.ai/hub/blog/perpl
exity-is-open-sourcing-bumblebee
… http://
github.com/perplexityai/b
umblebee
… -
Storyboard-first AI video workflow with Topview Canvas and Seedance 2.0
By
–
The prompt box was never enough.
— God of Prompt (@godofprompt) 26 mai 2026
AI video needs a creative workflow, not another “type and pray” casino.
Topview Canvas lets you build the storyboard first, arrange scenes on an infinite Figma-like canvas, work with an agent, then generate with Seedance 2.0.
Control before… https://t.co/nDA3Zhiv8F pic.twitter.com/Q2Qu1D4ZzbThe prompt box was never enough. AI video needs a creative workflow, not another “type and pray” casino. Topview Canvas lets you build the storyboard first, arrange scenes on an infinite Figma-like canvas, work with an agent, then generate with Seedance 2.0. Control before
-
LiveAvatar enables more present multimodal voice agents
By
–
Voice agents finally got out of the waveform era.
— God of Prompt (@godofprompt) 26 mai 2026
Your agent can already talk.
Now it can look back, react, and feel present without rebuilding your stack.
LiveAvatar plugging into LiveKit, Pipecat, Agora, and VisionAgent is a big unlock for real conversational UX. https://t.co/apZ4aYU4D8Voice agents finally got out of the waveform era. Your agent can already talk.
Now it can look back, react, and feel present without rebuilding your stack. LiveAvatar plugging into LiveKit, Pipecat, Agora, and VisionAgent is a big unlock for real conversational UX. -
Packaging thinking into reusable AI Skills is the next moat
By
–
The next AI moat won’t be “I know how to prompt.” That’s already getting commoditized. The real moat is packaging how you think into a Skill people can run. Your edge becomes reusable.
Your process becomes a product.
Your brain starts earning without being in the room. -
AI will reveal who has replicable expertise
By
–
AI won’t kill expertise. It’ll expose who actually had a process worth copying.
-
Uber COO Questions Returns on $3.4B AI Spend
By
–
Uber burned $3.4 billion on AI in four months. Their COO now says it's not paying off. And everyone's blaming Uber instead of asking the obvious question. This is Uber. Not some seed-stage startup guessing. They have more data than most countries. They ran the experiment. They