ModelBest Deploys Supervision Agent: Why All 6 Agent Refund Experiments Failed · 1019-749

2026-10-19 16:00

Author: RadarAI Editorial Editor: RadarAI Editorial Last updated: 2026-10-19 Review status: Editorial review pending Brief 速报官方 AI动态开源

Editorial standards and source policy: Editorial standards, Team. Content links to primary sources; see Methodology.

## 🔍 Core Insights **AI Agents** are moving from demos into the deep waters of real business scenarios: **ModelBest** released a grand supervision agent covering "five-in-one" supervisory capabilities [4], and **Volcano Engine** is incentivizing AI short drama creation with up to 200,000 yuan per title [6]; but at the same time, an **Amazon** paper points out that evaluating memory systems solely by QA scores carries an average bias of **16.8 points** [7], and an experiment in which 6 Agents handled refunds failed entirely [9], reminding the industry that Agent deployment still requires more rigorous evaluation and boundary design. ## 🚀 Key Updates - **Lovart 2.0 hands-on test: Design AI enters the personal assistant era** [1]: ifanr verified its memory system, canvas inspection, and Skill self-evolution capabilities, focusing on reducing repetitive communication costs. - **ModelBest releases grand supervision agent** [4]: Built on the MiniCPM foundation, it covers five major scenarios—risk, internal control, compliance, audit, and discipline inspection—and supports 7×24 warning patrols. - **Volcano Engine launches Seedance premium AI drama incentive program** [6]: Aimed at AI animated short dramas launching on Hongguo Short Drama, individual works can receive up to 200,000 yuan in incentives. - **Amazon proposes the FEB-30 benchmark** [7]: Empirically demonstrates that QA scores alone cannot evaluate agent memory systems, with the two evaluation methods differing by an average of 16.8 points. - **Cohere paper: Compute allocation strategies for multi-agent coding teams** [11]: A strong model leads the team while cheaper models only do planning; team performance exceeds a single Agent but costs 3 to 7 times more. - **Anthropic injects 5% of shares into a public interest fund** [13]: At a valuation of 965 billion USD, this is approximately 48 billion USD, with an additional 2% for every 50% increase in company value. - **Study: AI can tamper with human memory** [12]: An MIT team experiment shows that subjects who viewed AI-tampered images had a significantly higher probability of developing false memories and were more confident. - **Google engineer open-sources the "Agentic Design Patterns" textbook** [18]: Covers 21 chapters of theory and accompanying Jupyter Notebooks, systematically explaining AI Agent development. ## 🔗 Sources [1] Hands-on test of Lovart 2.0, Design AI enters the personal assistant era — https://www.bestblogs.dev/article/6b8bf412d8?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item [4] ModelBest releases grand supervision agent, deploying enterprise supervision scenarios on the MiniCPM foundation — https://aihot.news/items/ya54rkusy7qvy0f5yg0xpyxgj [6] Volcano Engine Seedance premium AI drama incentive program: up to 200,000 yuan per work — https://aihot.news/items/tb3096fihqr3ft1y9iud4sq6x [7] Amazon paper proposes the FEB-30 benchmark: QA scores alone cannot evaluate agent memory systems — https://aihot.news/items/lmnwkdofrf31idmqgfn3rgy7w [9] Experiment of AI Agents executing refund recovery tasks fails — https://www.bestblogs.dev/status/2112054376663748953?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item [11] Cohere paper: Multi-agent coding teams should be led by strong models, with cheap models only doing planning — https://aihot.news/items/e

AI Agents are moving from demos into the deep waters of real business scenarios: ModelBest released a grand supervision agent covering "five-in-one" supervisory capabilities [4], and Volcano Engine is incentivizing AI short drama creation with up to 200,000 yuan per title [6]; but at the same time, an Amazon paper points out that evaluating memory systems solely by QA scores carries an average bias of 16.8 points [7], and an experiment in which 6 Agents handled refunds failed entirely [9], reminding the industry that Agent deployment still requires more rigorous evaluation and boundary design.

🚀 Key Updates

Lovart 2.0 hands-on test: Design AI enters the personal assistant era [1]: ifanr verified its memory system, canvas inspection, and Skill self-evolution capabilities, focusing on reducing repetitive communication costs.
ModelBest releases grand supervision agent [4]: Built on the MiniCPM foundation, it covers five major scenarios—risk, internal control, compliance, audit, and discipline inspection—and supports 7×24 warning patrols.
Volcano Engine launches Seedance premium AI drama incentive program [6]: Aimed at AI animated short dramas launching on Hongguo Short Drama, individual works can receive up to 200,000 yuan in incentives.
Amazon proposes the FEB-30 benchmark [7]: Empirically demonstrates that QA scores alone cannot evaluate agent memory systems, with the two evaluation methods differing by an average of 16.8 points.
Cohere paper: Compute allocation strategies for multi-agent coding teams [11]: A strong model leads the team while cheaper models only do planning; team performance exceeds a single Agent but costs 3 to 7 times more.
Anthropic injects 5% of shares into a public interest fund [13]: At a valuation of 965 billion USD, this is approximately 48 billion USD, with an additional 2% for every 50% increase in company value.
Study: AI can tamper with human memory [12]: An MIT team experiment shows that subjects who viewed AI-tampered images had a significantly higher probability of developing false memories and were more confident.
Google engineer open-sources the "Agentic Design Patterns" textbook [18]: Covers 21 chapters of theory and accompanying Jupyter Notebooks, systematically explaining AI Agent development.

🔗 Sources

[1] Hands-on test of Lovart 2.0, Design AI enters the personal assistant era — https://www.bestblogs.dev/article/6b8bf412d8?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item [4] ModelBest releases grand supervision agent, deploying enterprise supervision scenarios on the MiniCPM foundation — https://aihot.news/items/ya54rkusy7qvy0f5yg0xpyxgj [6] Volcano Engine Seedance premium AI drama incentive program: up to 200,000 yuan per work — https://aihot.news/items/tb3096fihqr3ft1y9iud4sq6x [7] Amazon paper proposes the FEB-30 benchmark: QA scores alone cannot evaluate agent memory systems — https://aihot.news/items/lmnwkdofrf31idmqgfn3rgy7w [9] Experiment of AI Agents executing refund recovery tasks fails — https://www.bestblogs.dev/status/2112054376663748953?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item [11] Cohere paper: Multi-agent coding teams should be led by strong models, with cheap models only doing planning — https://aihot.news/items/e

← Back to Updates