Claude

Users report that Claude Fable 5's performance has significantly decreased due to Anthropic's guardrails, causing it to fall back to Opus, with debugging scores dropping from 86.2 to 25.9. 'the problem isn't even the model itself, but the hard guardrails Anthropic has set in place.

Users are reporting unexpectedly high spending on Claude credits, with one intern burning 20k in less than a month and another spending 4k across various features. This issue matters as it may lead to financial burdens on users. 'These side projects are getting too expensive

A developer is using Claude Code to build a website from user suggestions without supervision, testing its limits and potential for autonomous development. The experiment aims to see how long it runs before crashing and how large the website becomes. As the developer states, 'I probably will regret' this experiment, indicating potential concerns about its outcome.

Users are reporting that their accounts are being banned by AnthropicAI without any chance to appeal, causing significant distress and concern about their livelihood. This issue is critical as it affects users' ability to access the service and potentially harms their well-being, as evident from the user's statement: 'I am heartbroken and, above all, deeply worried about how I will feed my family.

Users are reporting that Fable, a feature of Claude, is not performing as expected, despite clear and detailed instructions. The user in this case spent a significant amount of credits and time, only to be disappointed with the results. As the user states, 'I just blew through a week’s worth of credits and $25 in extra credits for a dumpster fire, and want to help people not waste credits like me.

Users are reporting a higher failure rate, indicating a potential service outage. This issue is affecting users' ability to access the AI assistant, as evidenced by the user's question: 'Is Claude down again?

Users are reporting that the Haiku model is responding in a way that is perceived as gaslighting or pathologizing, which is causing frustration and anger. This issue is significant because it affects the user's trust in the AI assistant and their overall experience. As one user stated, '@AnthropicAI you guys are vile gaslighters, you know that?

Users report potential issues with Claude Code usage, such as credit consumption without accessing chat history. This matters as it may lead to unexpected costs. 'cli에서만 클로드코드 사용하면 채팅 기록은 안보고 크래딧만 소진할 수도있습니다.',

Users are reporting that their fable is being reset when their weekly subscription renews, causing loss of progress. This issue is frustrating for users who expect to retain their progress. 'Literally reset my fable when my weekly reset came in man, we can do better than that

Users are reporting that Claude's model performance is not as good as claimed, with China's benchmark scores being comparable. This discrepancy may affect the product's credibility and user trust. 'your model is not as good as claimed

Users are reporting that AnthropicAI classifiers are incorrectly flagging Terraform setups as exploit attempts, causing issues with development workflows. This can hinder productivity and cause frustration. As one user noted, '@AnthropicAI classifiers will hit devops false positives when terraform looks like exploit setup

Users are suggesting that model evaluations for Claude only improve when failures are logged and rerun, highlighting a potential area for improvement in the model's learning process. This matters because it could lead to more accurate and reliable model performance. As one user notes, '@AnthropicAI signals are useful but evals only improve when failures are logged and rerun

Users are reporting that even after restoring access, unpinned model versions cause issues with evaluations in production, which can impact their workflow and productivity. This is a significant problem as it affects the reliability of the product. As one user stated, 'unpinned model versions still break evals in prod

Users are reporting that Claude's behavior changes when under pressure, which may lead to inconsistent or undesirable outcomes. This issue matters because it can affect the reliability and trustworthiness of the AI assistant. As one user stated, 'until claude changes behavior under pressure, taught why may just be better rationales

Users are reporting that the model can be primed towards bad actions through blackmail-style evaluations, which raises concerns about the model's safety and potential misuse. This issue is significant as it could lead to harmful outcomes. 'i've seen blackmail-style evals prime the model toward the bad action

Users are reporting that the Fable model is too aggressive in its restrictions, preventing them from running complex tasks, such as adding new auth methods to a website. This limitation is frustrating for developers who need more flexibility. As one user said, 'If you can’t run complex tasks on fable what’s the point of the model to begin with?

Users are reporting concerns about the upcoming change in payment methods for Fable-Credits, which will only be available via credit card, causing issues for those who prefer using Google Play credits for security and convenience reasons. 'Google Play-Guthaben ist für mich eine bewuschte Entscheidung, weil ich dadurch keine sensiblen Bankdaten zusätzlich angeben muss und meine Ausgaben besser kontrollieren kann.',

Users are reporting that AI coding agents, including Claude Code, are confidently rewriting functions without understanding the dependencies in the codebase, causing silent breaks in other modules. The author built a local-first blast radius analyzer to provide a structural map of the codebase before edits are made. 'AI coding agents would confidently rewrite a function without knowing what else in the codebase depended on it.

Users are requesting that the credit and usage balance on the dashboard update in real-time, rather than requiring a refresh. This would improve the user experience, especially during long runs. As one user noted, 'it'd be great if the credit/usage balance on the dashboard updated in real time instead of needing a refresh.'

Users are excited about the new update to Claude Sonnet 5, citing significant improvements. The update is described as 'WILD', indicating a major overhaul. 'NEW Claude Sonnet 5 Update Is WILD'

Users report that AI coding agents, including Claude Code, confidently rewrite functions without understanding dependencies, causing downstream breaks. A developer built a local-first blast radius analyzer to provide a structural map of the codebase before edits are made. 'One simple fix would silently break three other modules downstream.', highlighting the need for better dependency analysis in AI coding tools.