Original video in Chinese.
Key Takeaway
- I use Python scripts to let an Agent automatically collect AI-related news and discussions from Reddit and global media, then generate a daily briefing called “AI Insider.”
- Valuable AI content is mainly overseas, so mastering English and learning directly from the source is essential.
- Reddit’s LocalLlaMa channel, Medium, and YouTube are important sources for AI news and for learning technologies like RAG and LangChain.
- CrewAI and GPT Researcher are practical tools for building Agent systems. CrewAI is suitable for freely building multi-Agent workflows, while GPT Researcher excels at information gathering and report generation.
- Agent technology can help users handle information efficiently and improve the efficiency of information retrieval and organization.
Full Content
The first thing I do every morning when I wake up is run two Python scripts.
An Agent powered by GPT-4 helps me collect all the AI-related news and discussions.
There are two information sources:
- Discussions in Reddit’s LocalLlaMa channel.
- Hot news reported by major media outlets around the world.
The biggest advantage of doing this is that it lets me process everything I need to know within one concentrated, complete block of time.
Once that’s done, unless there’s some especially major event today, like the release of Sora or GPT-5, I won’t be distracted by all those random articles anymore.
This AI-generated daily briefing about AI, I call AI Insider.
Hello everyone, welcome back to my channel. I share the Why and How of AI. If you really want to get involved and seize this once-in-a-lifetime opportunity in AI, make sure to follow. Let’s explore it together.
Back to today’s topic: sources for and access to information related to AI and large language models.
I know everyone is extremely hungry for AI-related news and knowledge. But there is one very harsh fact:
The valuable content is overseas.
If you don’t understand English and can only read secondhand news from within China, it’s really hard not to get ripped off.
Let me share a few sources I absolutely have to check every day.
Reddit/LocalLlaMa
If you work in the field, or already have a pretty good understanding of large language model technology, then the LocalLlaMa channel is a must-read. The discussion quality is quite high.
Let me casually open two posts:
If you want to learn technology, there’s a tutorial here: this guy fine-tuned Mistral-7B and used it to build an Agent, and the result was better than Gemini Pro. He gave the specific method.
If you want to read news opinions, there are also knowledgeable people here: Google is using Reddit data to train AI—what do you think of this news? Just like on Zhihu, the top-liked comment is a joke: garbage in, garbage out. But there are still serious answers: the reason they use the Reddit dataset isn’t to let the AI learn facts, but to let it learn how to converse—that immediately makes sense.
I check the content in the LocalLlaMa channel every day. Of course, I first use the Python scripts I mentioned at the beginning to screen and summarize everything for me, and I’ll introduce that later.
Medium
If you want to learn about a certain technology, you can go directly to Medium and search for it. Basically, there are articles that explain things in detail. And the authors on this platform are very patient; they write very thoroughly.
For example, this article: Build a Personal AI Tech News Agent. The author teaches everything step by step, from the principles to AWS setup and so on. It can be considered a nanny-level tutorial.
Another thing worth mentioning is that Medium’s recommendation mechanism is pretty impressive. It recommends exactly the kind of content I want to see. So on this platform, besides searching for something I need, I probably go there proactively every three days to take a look, and if I come across a good article, I bookmark it and save it.
YouTube
The first time I learned how to use RAG, LangChain, and so on was from these two guys on YouTube.
The first one is Sam. His Advanced RAG series and LangChain series are both especially good, and they were extremely, extremely helpful to me. Also, whenever a new popular large language model comes out, he puts out a tutorial.
The second one is James. He also created RAG and LangChain tutorial series, and he especially focuses on how to connect to vector databases like Pinecone.
For beginners, once you’ve watched their videos, you’ll definitely know how to do it. And the code in the videos is shared through Google Colab, so you can run it directly in the cloud and get a feel for it.
AI Insider
The two Python scripts I mentioned at the beginning weren’t written by me; both came from GitHub.
The first script came from this young lady. I found it through her YouTube channel.
In the script, she used CrewAI to build a workflow containing three Agents. CrewAI is an Agent framework that’s especially simple and intuitive. I strongly recommend everyone try it.
The key to this script is actually giving the Agent a tool for reading Reddit data. The author used PRAW, which is the Python Reddit API Wrapper. It’s a Python package that allows you to access and manipulate Reddit data programmatically.
As for Medium mentioned earlier, I also tried using an Agent to crawl data there, but the result wasn’t good. Whether there is an open data interface makes a huge difference.
The second script is also ready-made, called GPT Researcher. After cloning the GitHub repository and running it, it provides a local link, and when you open it, you get a visual interface like this. Then you just enter the information you want it to gather for you. After about 10 seconds, a pretty decent briefing is generated.
Finally, I put the results generated by these two scripts into a Word document, adjusted the formatting a little, and that was it. Actually, I could also have the Agent generate a document in a specified format, but I couldn’t be bothered to spend any more time on that. Anyway, it’s just for me to read, so I don’t have to be that particular.
CrewAI and GPT Researcher, which I just mentioned, were both introduced in my previous videos. You can go to my homepage to find them. Agent technology is very important, so I recommend everyone understand it in detail.
OK, that’s all for today. I’ll keep iterating on this AI Insider going forward, and I’ll share more with everyone then. See you in the next episode!