Researcher at Unknown
Identified similar glitch tokens in GPT-2 and GPT-3, including the famous 'SolidGoldMagikarp' token
How media typically covers Jessica Rumbelow
Research or work cited
Analysis of OpenAI's open-weights GPT-oss model reveals that GPT-5 was trained on data from adult websites, discovered through examining token embedding norms and tokenizer composition.
“Identified similar glitch tokens in GPT-2 and GPT-3, including the famous 'SolidGoldMagikarp' token”