“rl from ai feedback”的搜索结果

Explore training workflows that reduce harmful model responses using a written set of principles. The skill covers self-critique, revised training examples, and RLAIF, reinforcement learning from AI feedback, with implementation examples and hardware guidance.

中文简介正在准备中。当前显示英语简介。

• Designing principles for model safety training
• Creating training examples through self-critique and revision
• Planning reinforcement learning with AI-rated responses

Explore training workflows that reduce harmful model responses using a written set of principles. The skill covers self-critique, revised training examples, and RLAIF, reinforcement learning from AI feedback, with implementation examples and hardware guidance.

中文简介正在准备中。当前显示英语简介。

• Designing principles for model safety training
• Creating training examples through self-critique and revision
• Planning reinforcement learning with AI-rated responses

Guide language model training with TRL, from learning to follow instructions to matching human preferences. Covers method selection, training data, reward models that score responses, evaluation, and adjustments for memory limits or unstable training.

中文简介正在准备中。当前显示英语简介。

• Train a model on instruction and response examples
• Align responses using preferred and rejected answer pairs
• Build a training pipeline using human feedback

Guide language model training with TRL, from learning to follow instructions to matching human preferences. Covers method selection, training data, reward models that score responses, evaluation, and adjustments for memory limits or unstable training.

中文简介正在准备中。当前显示英语简介。

• Train a model on instruction and response examples
• Align responses using preferred and rejected answer pairs
• Build a training pipeline using human feedback

Extracts full item detail from any open eBay item URL, returning JSON with url, itemNumber, title, subTitle, categories, price, priceWithCurrency, currency, wasPrice, available, availableText, sold, image, images, seller, sellerUrl, sellerFeedbackCount, sellerPositiveRating, itemLocation, brand, type, mpn, upc, ean, condition, shippingCost, shipping, whyToBuy, and itemSpecifics (full label/value map). Works across all eBay regional TLDs. Use when user mentions eBay item page, eBay listing detail, eBay product page, eBay itm, ebay.com/itm, eBay item scraper, extract eBay item details, eBay item specifics, scrape eBay item, eBay sold count, eBay availability, eBay brand MPN UPC EAN, eBay item location, eBay seller feedback, eBay wasPrice, eBay hi-res images, eBay itemNumber, eBay per-item enrichment. Also applies to enriching item URLs from a listing, monitoring price and stock, competitive product research, brand catalog audits, and pulling full field set from an individual eBay listing.

中文简介正在准备中。当前显示原文说明。

Método da Anthropic para treinar IA inofensiva através de auto-aprimoramento. Abordagem em duas fases - aprendizado supervisionado com auto-crítica/revisão, depois RLAIF (RL from AI Feedback). Use para alinhamento de segurança, redução de outputs prejudiciais sem rótulos humanos. Alimenta o sistema de segurança do Claude.

中文简介正在准备中。当前显示原文说明。

Choose a model training method and configuration from paired response preferences or individual positive and negative ratings. Covers DPO, ORPO, KTO, and SimPO, including pair construction, repeated training rounds, and troubleshooting learning settings.

中文简介正在准备中。当前显示英语简介。

• Choose a training method for paired response preferences
• Select a method for individual positive and negative ratings
• Review learning settings when DPO training loses general capability

Manage job searches through modes for evaluating vacancies, tailoring CVs, tracking applications, drafting outreach, and preparing for interviews. Start from a job description or URL, or choose a specific task to produce relevant documents and status summaries.

中文简介正在准备中。当前显示英语简介。

• Compare vacancies and prioritize applications
• Create a tailored CV and application email draft
• Track application status and overdue follow-ups

Fine-tune LLMs using reinforcement learning with TRL - SFT for instruction tuning, DPO for preference alignment, PPO/GRPO for reward optimization, and reward model training. Use when need RLHF, align model with preferences, or train from human feedback. Works with HuggingFace Transformers.

中文简介正在准备中。当前显示原文说明。

Check earlier Paperclip tasks for verified progress and actual deliverables, then report findings with links. When requested, add feedback as the connected person, with safeguards against posting the same comment twice.

中文简介正在准备中。当前显示英语简介。

• Check progress and deliverables from earlier tasks
• Add feedback to a task as the connected person
• Monitor task updates in an environment that supports Events

Open Plannotator’s browser-based code review interface from Kiro and address the returned feedback. Review the current directory, another working directory, or a pull request URL, and optionally choose a comparison base to focus the changes shown.

中文简介正在准备中。当前显示英语简介。

• Review code changes from Kiro
• Review a separate working directory
• Open a review for a pull request URL

Build a scheduled agent that collects public website or API data, summarizes or scores it with Gemini Flash, and stores results. Includes guidance for duplicate removal and using user feedback to inform future scoring.

中文简介正在准备中。当前显示英语简介。

• Collecting job listings and scoring their relevance
• Tracking product prices on a schedule
• Collecting news and classifying it by topic

Review code changes through a structured check of requirements, design, correctness, security, performance, and tests. Produce specific, constructive feedback that separates required fixes from optional suggestions, and build shared review guidelines for a development team.

中文简介正在准备中。当前显示英语简介。

• Review a pull request systematically
• Create a shared code review checklist
• Give actionable feedback to junior developers

Review a Figma design, screenshot, or written description for usability, visual hierarchy, consistency, and accessibility. Receive specific findings, explanations, and prioritized improvements, with feedback tailored to the design stage and intended users.

中文简介正在准备中。当前显示英语简介。

• Reviewing an early screen design
• Checking navigation and onboarding usability
• Finding inconsistent colors, spacing, and typography

Review a Figma design, screenshot, or written description for usability, visual hierarchy, consistency, and accessibility. Receive specific findings, explanations, and prioritized improvements, with feedback tailored to the design stage and intended users.

中文简介正在准备中。当前显示英语简介。

• Reviewing an early screen design
• Checking navigation and onboarding usability
• Finding inconsistent colors, spacing, and typography

Helps an agent take over an existing role after a planned handover. It guides checks of inherited records, identity, authority, and recurring duties, while explaining how to recover the predecessor’s reasoning and preserve lessons for the next agent.

中文简介正在准备中。当前显示英语简介。

• Getting oriented after a planned agent handover
• Checking inherited records and recurring duties
• Recovering the reasoning behind earlier decisions

Fine-tune LLMs using reinforcement learning with TRL - SFT for instruction tuning, DPO for preference alignment, PPO/GRPO for reward optimization, and reward model training. Use when need RLHF, align model with preferences, or train from human feedback. Works with HuggingFace Transformers.

中文简介正在准备中。当前显示原文说明。

Split a large pull request into smaller, reviewable changes. Plan branch dependencies, extract related code, verify each part, and track review changes while preserving the intent of the original work.

中文简介正在准备中。当前显示英语简介。

• Split a large pull request into reviewable parts
• Separate refactoring from behavior changes
• Plan dependent pull requests

Use Memori through MCP, a protocol connecting agents to tools, to retrieve past decisions and preferences, summarize ongoing work, and resume after conversation compression. It also guides selective long-term memory storage and usage limit checks.

中文简介正在准备中。当前显示英语简介。

• Retrieve decisions and preferences from earlier sessions
• Review current work when starting a session
• Resume work after conversation context is compressed

Help law clinic students prepare motions, letters, petitions, and declarations from case notes and templates. Depending on the supervisor’s teaching mode, provide structure, drafting feedback, or a first draft with missing facts and verification points clearly marked.

中文简介正在准备中。当前显示英语简介。

• Preparing an eviction answer from case notes
• Organizing an asylum declaration
• Drafting a demand letter for supervisor review

This skill helps users extract structured video list data and comment data from YouTube using the BrowserAct API. The Agent should proactively apply this skill when users request searching for YouTube videos and their comments, analyzing viewer sentiment for a specific video topic, gathering audience feedback on AI or automation, extracting a list of top videos and their viewer reactions, compiling YouTube video data along with user opinions, retrieving competitor video titles and related audience discussions, monitoring public response to specific YouTube search keywords, summarizing comments from search results for market research, tracking viewer engagement metrics and replies for trending topics, collecting YouTube video URLs and author details alongside community discussions, or automating the extraction of YouTube comments without manual scraping.

中文简介正在准备中。当前显示原文说明。

Extract and analyze writing improvements from GitHub PR review comments. Use when asked to show review feedback, style changes, or editorial improvements from a GitHub pull request URL. Handles both explicit suggestions and plain text feedback. Produces structured output comparing original phrasing with reviewer suggestions to help refine future writing.

中文简介正在准备中。当前显示原文说明。

使用统一标准审阅论文和研究计划，检查研究方法、证据、引用及文章结构，并说明评分依据、修改优先级和下一步需要修改的内容。

• 投稿前审阅论文
• 检查引用是否支持论点
• 评估研究方法与局限

Review code changes or documents through configured review lenses, each representing a different perspective. Findings identify the location, triggering condition, suggested fix, and potential consequence, with reports available as JSON, Markdown, or both.

中文简介正在准备中。当前显示英语简介。

• Reviewing a pull request
• Checking uncommitted changes
• Reviewing specifications or documents