The strong version says successive price changes are independent, so no amount of studying history helps. Randomly generated series produce convincing trends, head-and-shoulders patterns and support levels, which is the most uncomfortable demonstration for chartists.
The evidence is mixed rather than settled. Strict randomness is rejected by most studies: volatility clusters, momentum effects appear across many markets and decades, and short-term reversal effects exist. But the predictable component is small and inconsistent.
The honest position for a trader is somewhere in between. Assume most of what you see is noise, demand a large sample-size before believing in an edge, and treat any approach that claims high accuracy as almost certainly mismeasured.
Related: efficient-market-hypothesis, overfitting, technical-analysis, sample-size, elliott-wave-criticism