 The use and mastery of phonetic transcription is important in many linguistic areas, in the description of the sound systems of languages, in the description of dialectal variation, in language-specific pronunciation training, in clinical linguistics, in forensic linguistics to name a few. These large language models yield amazing results in many areas, including the phonetic transcription of words and texts. But does this suffice for the mastery of transcription in these areas? No. Although large language models such as chat GPT can generate transcriptions, human expertise remains invaluable for refining such transcriptions and addressing limitations in accuracy. Even today, we still have to master the international phonetic alphabet and the principles of articulatory phonetics. We need to practice using various exercises and we need audio examples and ear training to encourage phonetic analysis and the comparison of transcriptions. But how can we realize phonetic training in the age of large language models? I recommend two central options, the use of language-independent and language-specific nonsense words and the analysis of language-specific erroneous material. With nonsense words, we take advantage of the fact that non-existing material cannot be transcribed at all or that an external control of the result, for example, by lexic lookup, is impossible. The use of nonsense words becomes even more challenging with audio input, either in face-to-face training situations or via audio files. Here are some language-independent examples for ear training. Zhiv, nup, Zhov-tri, Shan-ja. And now, let's present some nonsense words using present-day English phonotactics. The second option uses language-specific transcription passages with deliberately created problems. The following types of problems have shown to be very useful for this training purpose. The inconsistent use of phonetic symbols and the effects of connected speech. The learner's task is simple, identify and correct the problems. A comparison with AI-generated solutions is of little help here since their results are themselves error-prone. To summarize, the use of nonsense words and error analysis are reasonable options for practicing phonetic transcription even in the age of large language models. Beyond the introduction of the principles of articulatory phonetics and the use of the phonetic alphabet, the effective teaching of phonetic transcription necessitates exercises and practice materials to familiarize our students with these principles using audio input wherever possible. Large language models can be used for support. Common expertise, however, will always be the final authority.