Weekly Digest 8

AI

Personal Insights

AI-way of problem solving

When you do not know how to do things, you ask AI how to do it. You should take time here to really understand the underlying mechanism to your problems instead of memorize the procedures of how to do that things.

So next time you encounter that problem, just ask AI and you can swiftly validate and act upon the answers.

In case that you have encountered the same problem over and over again, try to build a workflow of it.

There is a underlying principle that do not try to build the structure prematurely until you really understand your needs.

Agent

introducing-chatgpt-agent

After seeing how it work in practice, I do not think it can do real work for a couple of reasons:

• It can make spread sheet, but this is no big deal.
• Cost too much time to perform a task.
• Plus user can only use 40 message per month.

With all this combined, I do not think it’s useful, especially from a engineering’s perspective.

But as shown in benchmark, this agent model is a really powerful model, this may worth giving it a try.

AI persona

The case for regulating AI companions

Why is 4o so much more sycophantic than o3? I don’t think it’s a mistake. It’s a consequence of the models’ different demographics. 4o is optimized for mass appeal: for people who are new to AI, or just want to idly chat with a model. o3 is optimized for professional use: for people who are a bit more familiar with how AI works, and want a coding or research assistant. Making o3 more sycophantic would be a bad business decision, because it would make it less useful to its target audience. But making 4o more sycophantic makes it better at appealing to its target audience.

So, this article explain why I should not use 4o – it is the product targeted for AI beginners, not targeted for professional use. For now, o3 is the best choice from OpenAI.