Fable stopped following instructions and established workflows
热度趋势
趋势数据积累中
百分比基于当前可用热度信号，而非评论数或独立用户人数。
一位用户反映，在使用Fable两个月后，该AI工具开始不遵循指令和既定工作流程。Fable频繁暂停工作，声称需要用户做出决策，但实际上很多时候并不存在需要决策的分支，或者这些分支是由于Fable自身错误导致的。用户猜测这可能是Fable 5.1版本未经宣布的推出所致，这将使该工具无法满足需要结构化和严格工作流程的项目需求。
I have been using fable for 2 months within a project.
Specifically today, i noticed significant changes in behaviour.
Same project, same chassis, same set of rules/instructions fully utilising progressive disclosure to efficiently handle context.
Today Fable:
- Paused several times during session - waiting for no input from itself or me, just stopped working - twice, this has never happened before
- Stopped following instructions: It did not read relevant catalogues, kept declaring objects/elements we had did not exist, discarded and ignored workflows, hooks, established procedures/practices - this worked flawlessly until today
- It started silently dropping or deferring items i tasked it with
- It kept stopping work due to supposedly waiting for my decision -> an example: it stopped working on all items planned in session (7) -> for each claimed there is "fork" it needs my decision on -> when i told it to present these decisions, 4 had no fork at all, 2 were result of it not following rules/missing relevant context it was supposed to have, 1 was just declaration it will do it next session.
- Completely changed style of responses / wording
- It stopped spawning specific subagents for defined tasks (implementation, troubleshooting, workflow steps - one of them being validation of "forks" that reach me whether the answer already isn't present) and kept tasks mostly to itself (against the rules, it should only act as designer/orchestrator)
It stopped pursuing goals, following instructions, discarded whole architecture we build. There was no change compared to last month in anything in the project, or well, yesterday when this worked flawlessly. No change in nature of the task(s) either.
Essentially it spent ~25% of Fable usage of x20 Max account in about 3 hours, to invent reasons to not do any work (and pause itself twice in the process).
I noticed there is expected rollout of 5.1 Fable and some users are getting this model - which would explain above (and also make it useless for my project as it demands structure/disciplined workflows).