Bilingual

I used to love Claude, but the latest models are slowly ruining it
10 days ago
- Author expresses initial preference for Claude over Gemini due to its strong memory and collaborative feel.
- Claude increasingly refuses or pushes back on creative and sensitive topics, showing inconsistency and overcaution.
- The problem spans multiple models (Sonnet 5, Opus 4.8) and may stem from heightened safety protocols post-government concerns.
- Clarifying prompts helps, but unpredictability remains, leading the author to consider Gemini for its consistency.
- Despite issues, Claude is still valued for its capabilities when it cooperates, and user experiences vary widely.
Automated Moderation Is Here to Stay–Accountability Must Keep Pace
7 days ago
- Automated content moderation systems frequently produce errors in low-resource languages like Arabic, with significant inaccuracies in detecting nonviolent content and hate speech.
- Moderation algorithms exhibit systemic biases, including suppressing content from specific regions like Palestine and misclassifying LGBTQ+ material as explicit, harming marginalized groups.
- The Santa Clara Principles 2.0 emphasize human rights and transparency, recommending automation as a supplement to human moderators rather than a replacement, with accountability measures such as appeals and audits.
GPT 5.6 solved all 6 problems from IMO 2026
5 days ago
- Users should report posts to moderators for review.
- Act in good faith and treat others with respect; avoid misinformation and straw-manning.
- No trashposts: low-effort posts without value may be removed; 'Is ChatGPT down?' posts are specifically prohibited.
- Self-advertising posts must relate to ChatGPT or LLMs; no sole focus on promoting other LLM services.
- Political content creation is fine, but general political debates are not allowed; discussions around AI politics are permitted.
- No NSFW content: all content must be SFW.
- GPT-4o/GPT-5 complaints should be directed to a dedicated megathread.
- AI art submissions must be unique and interesting, with insights into AI technology encouraged.
- Moderators have discretion to remove posts that do not meet the criteria.
xAI's first lawsuit against a user tests who is responsible for what Grok makes
2 days ago
- xAI sues a user for allegedly using Grok to generate child sexual abuse material by circumventing safeguards.
- The complaint highlights that Grok's guardrails engaged and refused prompts, but the user repeatedly altered prompts to bypass moderation.
- xAI reports suspending over 52,000 accounts and filing more than 73,000 reports to NCMEC, leading to nearly 250 arrests in 2026.
- The case contrasts with ongoing litigation where xAI is a defendant, accused of being responsible for Grok's output, not users.
- The lawsuit does not resolve whether general-purpose image models that can be manipulated to produce harmful content should be publicly available.

Hasty Briefsbeta

#content moderation

I used to love Claude, but the latest models are slowly ruining it

Automated Moderation Is Here to Stay–Accountability Must Keep Pace

GPT 5.6 solved all 6 problems from IMO 2026

xAI's first lawsuit against a user tests who is responsible for what Grok makes