 For teachers of English as a foreign language, the identification and mastery of connected speech effects is highly important for their own pronunciation and comprehension, in order to provide authentic language input, to address potential misunderstandings, or to enhance cultural awareness. A common teaching method uses texts and then asks the students to identify the connected speech effects, assimilation, elision, liaison and weakening. Since today's large language models yield amazing results in many areas, why not use systems like ChatGPT to solve this task for us? OK, let's see what ChatGPT can tell us. Here is a general part of ChatGPT's answer. Sounds great, but in connected speech we neither talk about sounds nor about syllables. The focus are phonemes and the question what phonemic effects can be identified in connected speech. So the answer is questionable. Let's see how ChatGPT continues. This is what ChatGPT has to say about assimilation, and the result is linguistic nonsense. Not only because of the sound versus phoneme problem, the analysis is wrong since long wide hallway is not an example of assimilation at all. A true example is NotMyRoom, where the alveola t in Not may change to bilabial b under the influence of m in My. And note, assimilation is optional. Let's continue with elision. Another optional effect where ChatGPT's answer is misleading. Again we have the sound syllable versus phoneme problem and then contractions are normally not listed under elision. A typical but not mentioned case is next door, where the middle of three consecutive consonants can be dropped. Again the answer is not very helpful. Finally, here is what ChatGPT has to say about linking, an effect that is normally subsumed under liaison. Again the answer is inadequate. Beyond the sound versus phoneme problem, the example is correct, but it is interpreted incorrectly. It is not the d that links to anything, but the r as in the door opened. To summarize, there is no need to change classical training methods to master connected speech effects. You need text material and have to find potential candidates of assimilation, elision and so on. Large language models are of little help. Their analysis results are by and large unacceptable and may confuse the learner rather than support him. You need well-trained linguists for the mastery of this topic. By the way, did you notice that ChatGPT did not even mention weakening effects? So you'd better not trust the AI text generator. Okay, that's it for now, thanks for listening and watching.