The DLI has a different set of motivations: bring as many analysts to a working level of proficiency within a fixed time period, and has a lot of other things that confound this analysis: it's totally ok for the DLI to wash out some percentage of trainees in each flight, they can pre-screen candidates into whatever language they think is a best fit depending on candidates latent aptitudes, etc.
There is a universe where both pieces of advice can be right; surely you can imagine "don't worry about speaking" advice might be reasonable for someone who is learning casually, has no hard time frame to learn, wants to learn in dribbles and maximally passively, for whom a partial fluency might be a reasonable endpoint, and the greatest risk is burnout from negative reinforcement of failure.