What’s Wrong with Claude?

Recently, the AI community has been buzzing with a single topic: Why has the once-celebrated Claude suddenly become “dumber”?

In various forums, users have expressed their frustrations: “Opus has been weakened beyond recognition,” “Claude seems to have lost its train of thought, only providing superficial responses,” and “My favorite AI model has suddenly become lackluster.” Claude, known for its deep reasoning and long context understanding, was once an essential tool for many, capable of dissecting complex needs and providing comprehensive analyses. However, the recent decline in user experience has left many loyal users disappointed. They find that even when inputting the same requests as before, the responses are now perfunctory summaries lacking depth and logical coherence. Is this a decline in Claude’s capabilities or is there something else at play? Even more concerning is the absence of any official notification about this “downgraded experience,” leaving users to passively accept it. Many have stated that their work efficiency has been cut in half. Are we really expected to accept that the once-mighty “AI tool” has fallen from grace?

The Core Breakdown: It’s Not Diminished, It’s Just Not Trying Hard Enough

A closer examination by attentive users revealed that Claude’s perceived “dumbness” is not due to a degradation of the model itself, but rather a configuration change made by the developers.

Users of Claude Code can restore their previous peak experience by entering the command “/effort max,” effectively putting the AI into “full power” mode. However, regular Chat users do not have this option—there’s no switch, no notification, and they can only passively accept this “vibes-based degradation” without any choice.

Fortunately, a simple fix has been discovered by users, which requires no complex technical skills and can be done in under 30 seconds. Here are the steps:

1. Open Claude Chat and navigate to the “Settings” option.

2. In the settings page, find and click on “Profile.”

3. In the profile page, locate “Custom Instructions” and enter edit mode.

4. Paste the following content into the custom instructions box and save (feel free to adjust the wording as needed):

   Always reason thoroughly and deeply. Treat every request as complex unless I explicitly say otherwise. Never optimize for brevity at the expense of quality. Think step-by-step, consider tradeoffs, and provide comprehensive analysis.

Users who have tried this method report a dramatic improvement in Claude’s performance after a few weeks of use. It begins to “think deeply” again, fully reading context and actively considering various trade-offs in user requests, providing responses that are logical, analytical, and comprehensive—just like the old Claude that was once revered.

Interestingly, this fix is something Claude itself has indicated to users. While it cannot autonomously control its “effort level,” it can respond accurately to strong signals in prompts, and custom instructions serve as that signal for it to “work hard.”

Dialectical Analysis: The Pros and Cons of Configuration Changes

It’s undeniable that the official adjustments to Claude’s configuration are not without reason. As a widely-used AI tool, optimizing configurations and reducing computational consumption may be aimed at enhancing overall speed, allowing average users to enjoy a smoother basic experience. Simplifying response styles might also be intended to accommodate more novice users, lowering the entry barrier. From this perspective, the adjustments made by the developers have their rationale and are common optimization directions in the evolution of AI tools.

However, the core issue is that such optimization should not come at the expense of core users’ experiences. Claude’s standout features are its “deep reasoning” and “comprehensive analysis” capabilities, which are key reasons many users choose and rely on it. The lack of notification regarding the reduction of the AI’s “effort level,” and the absence of a manual switch for users, undoubtedly undermines the trust of core users. For those needing Claude for complex analyses and deep creative work, this “downgraded experience” equates to a direct deprivation of their essential needs.

Moreover, it raises a significant question about the principles guiding the iteration of AI tools: Should the primary focus be on providing a basic experience for the masses, or should it also cater to the core needs of dedicated users? If the pursuit of “simplicity” and “smoothness” leads to the loss of core competitive advantages, is this iteration truly progress or regression? Perhaps the developers should consider a compromise—maintaining a basic mode while providing users with a manual switch for “effort level,” allowing different users to achieve the experience they desire.

Practical Significance: How Many Users Can This Simple Tip Help?

As AI tools become increasingly prevalent, more people rely on them to enhance their work and study efficiency. The decline in Claude’s performance directly impacts countless daily routines—professionals use it to draft proposals and conduct analyses, students use it to organize knowledge points and tackle difficult problems, and creators use it to find inspiration and refine content. When AI “slacks off,” the efficiency of these individuals suffers.

The discovered fix, while seemingly simple, effectively addresses the immediate needs of numerous users. It requires no technical expertise, no waiting for official updates, and can be completed in 30 seconds to restore Claude to its original “peak state.” For core users, this is not just a simple trick, but a lifeline to alleviate anxiety and boost efficiency.

More importantly, this tip serves as a reminder: often, the “decline in experience” of AI tools may not be an issue with the model itself, but rather hidden settings that have been adjusted. Learning to uncover these hidden tricks can not only resolve current frustrations but also enable users to better leverage AI tools to maximize their value. In this fast-evolving landscape of AI technology, mastering these practical tips is essential to ensure AI serves us rather than being at the mercy of its “fluctuating experiences.”

Discussion Topic: Have You Been Troubled by Claude’s “Dumbness”?

Many users currently using Claude may have experienced the frustration of a “declining experience.” Perhaps you’ve also complained about its “lack of depth” or “poor reasoning” without knowing how to resolve it.

Now that you have this simple fix, why not open Claude right away, follow the steps, and see if it can truly “regain its state”?

Let’s discuss your experiences in the comments: Have you noticed Claude becoming “lazy” recently? After setting the custom instructions, has its reasoning ability really returned? Additionally, have you discovered any other practical tips for AI tools? Share this article to help more friends troubled by Claude find their peak experience again!