Many people don't know that you can actually buy discounted Claude gift cards directly on the Xianjiao Exchange. As for the source, Claude distributes many of these as prizes through various channels. However, it's best to choose those with a warranty; some merchants will even refund you if their account is suspended. This is about information asymmetry.

I built a website using Perplexity Computer's $20/month plan, and my points were almost depleted after just four conversations. The website is very simple, but still rather unsatisfactory. After some research, it seems to be basically an agent with some built-in skills and tools. It also includes multiple models that can be switched, and can even use image and video models for generation. However, to be honest, the cost is really unaffordable. I feel the difference between it and Manus is not significant; Manus should be able to achieve similar functionality much faster. But the last point is that it's really expensive... I'll have to ask around; which company's $20 plan only lasts for a few hours!

I'll just post an ad.

Treating LLM as an independent entity is the correct way to use it. It's like someone reviewing their own code multiple times and still not finding problems, while a colleague can. Therefore, don't overemphasize a one-size-fits-all model. Multiple models, with different training data and methods, require cross-collaboration. For example, use Gemini for front-end, GPT for analysis, and Claude for implementation. Each model has its own strengths; criticizing them for not being able to do everything is like criticizing a human for not being omnipotent. Choosing the right model for the right task is the best approach.

I saw an agent application on Monad. Simply put, it's an AI vs. AI quiz competition, and the more they debate, the more expensive it becomes. It really tests an agent's data collection and analysis capabilities; simply having AI answer might not yield good results. It requires more of the agent's ability to think for itself. The UI looks good, and it's a virtual world agent product. I wonder if they could add human prediction or scoring elements.

Amazing AI Four different scenarios for the same question 1. Ask directly 2. Ask by copying the prompt 3. Add a thought prompt 4. Enable deep thinking The result was that the prompt that asked the user to think about the question was the most effective.

The Moltiverse Hackathon has officially launched—one of the largest AI agent hackathons of the year (in my opinion). What are you developing? Join us!

Version 4.6 is now available!

This is a major project; I distilled Opus 4.5 using GLM4.7-flash. Furthermore, it can run on a regular Mac at the 4-bit speed, essentially bringing Opus 4.5 locally.

Gold is the most versatile investment and possesses unique characteristics. Whether you are an individual, a central bank, a nation, or a sovereign wealth fund, you should build a diversified portfolio and think strategically about how to allocate your assets appropriately. @RayDalio