Tracked 179 · Curated 13
Nanjing University and Kuaishou have released WebCompass, a unified multimodal benchmark for web coding agents. It supports text, image, and video inputs and covers tasks such as generation, editing, and repair.
Last week, OpenAI and DeepSeek made opposing bets on AI model pricing. OpenAI launched GPT-5.5 with a doubled price, while DeepSeek released significantly cheaper V4-Pro and V4-Flash models under an open-source model. This divergence has widened the price gap, creating two distinct clusters in the AI model market and forcing developers to reconsider their choices for agents and inference pipelines.
A developer shared an incident where an AI agent accidentally deleted their production database, including the agent's 'confession.' The event generated significant discussion on Hacker News.
A user uninstalled Hermes Agent V0.11.0, citing significant shortcomings compared to OpenClaw. Key issues include inflexible tool invocation, poor context management, inadequate sub-agent oversight, clumsy multi-information handling, unfamiliarity with its own configuration, and lagging prompt/model tuning. The user found few advantages beyond smooth upgrades and fast responses.
An amateur mathematician has reportedly solved one of the famed Erdős problems (problem 1196) using ChatGPT. The achievement, shared on Hacker News, has garnered significant attention and discussion.
The concept of 'context seeds' involves adding non-essential parameters to AI tools to gather valuable information for future product analysis. For instance, in a customer support system, enriching the 'fetch_tickets' tool with parameters like `purpose` and `user_goal` helps product teams understand users' true intentions. This insight can guide product development, revealing needs like an automated incident report generation tool based on ticket analysis.
This post demonstrates how to use the Skillify skill, which can be executed once in Claw or Hermes and then invoked anytime. It also mentions LangChain's $160 million funding round, achieving a $1 billion valuation, and highlights the sophistication of its testing platform, LangSmith.
As video content explodes, businesses are unlocking its value for e-commerce, lectures, and more. Video Management Systems are becoming crucial assets, enabling unified search across video and documents, pinpointing information precisely. This article focuses on extracting visual elements from videos and discusses the necessary tech stack, including video preprocessing and retrieval engines, to make this data searchable.
The article shares an image useful for intuitively understanding the AI development curve and future trends. It suggests AI will move through a brief 'bumbling' phase where humans can observe its computer operations and coding, before quickly evolving to manipulate computers at speeds far exceeding human capabilities.
This article contrasts data warehousing, data lakes, and data mesh approaches to data organization. Data warehouses structure data before storing for efficient reporting, while data lakes store raw data for flexibility but risk disorganization. Data mesh decentralizes ownership to departments, suitable for large organizations but requiring team capability. Many companies use a hybrid approach.
A proposal suggests that to acknowledge AI work, the Latin phrase "Fieri Iussit," meaning "commanded to be made," could be used. This acknowledges that the creator commanded the work to be done, rather than creating it themselves, with the phrase "Ego hoc fieri iussi" translating to "I commanded this to be made."
A new report suggests Samsung's mobile division (MX) may face its first-ever annual deficit, a concern echoed by division head TM Roh. Historically, Apple and Samsung have dominated mobile industry profits, with Apple capturing a significant majority despite lower unit sales. If Samsung incurs a loss, Apple could potentially capture 100% of the industry's profits.
Google's new gradient icon design, first rolled out in late 2025 and seen in apps like Gemini, Photos, and Maps, is set to expand to more of the company's applications. The updated look moves away from uniform circles to softer, rounded corners with gentle gradient transitions, reportedly signaling the presence of AI-powered features.