FrancesScott
Like a virgin
- Joined
- May 15, 2025
- Posts
- 541
I asked a friend:Don't be too hasty. Maybe O.P. hasn't seen any of your suggestions yet. Either way, you went out of your way to help, and the rest is on them.
On topic, we often debate that leaving grammar and stylistic errors helps? AI supposedly doesn't make them, and it shouldn't mix tenses the way the O.P. did? Or does the algorithm only take into account each sentence individually, so the mixing of tenses doesn't come into play?

- AI models (including me) are designed to avoid grammar and stylistic errors unless asked to mimic a particular voice.
- Humans often debate whether leaving “natural” errors makes writing more authentic. In practice, consistent grammar usually improves clarity, especially for factual or analytical writing.

- AI is not supposed to mix tenses arbitrarily the way human writers sometimes do.
- When AI does mix tenses unintentionally, it is typically a generation inconsistency that ideally should be edited out.
- That said, deliberate tense shifts (e.g., moving from past to present for immediacy) are acceptable if done intentionally.

- No, large language models (LLMs) like GPT do not process each sentence in isolation; they consider context windows (thousands of tokens) to maintain coherence, tense consistency, and logical flow.
- However, under certain conditions (prompt ambiguity, cut-off windows, or fast generation), models may lose track of tense consistency across sentences, leading to the kind of mixing you noticed.
In summary:
- Leaving errors for “authenticity” may be valuable in creative or conversational writing but typically degrades clarity in analytical work.
- AI should maintain tense consistency unless intentionally shifting.
- The algorithm takes context into account, but imperfect tracking can still cause tense mixing, which should be corrected during review.