I'm a few months in to agentic thinking, kind of cool to look at the source for Deep Search (open source replication of OpenAIs version). First 'others' project for agentic stuff I've personally seen -- nice to see others structure/thinking/way of doing. Do other people here have other projects with source/github of examples of agentic work that are worthy of a look? Article: https://arstechnica.com/ai/2025/02/after-24-hour-hackathon-hugging-faces-ai-research-agent-nearly-matches-openais-solution/ Repo: https://github.com/huggingface/smolagents/tree/gaia-submission-r1/examples/open_deep_research

Okay, smart people here with AI/data background. Please point me elsewhere if somewhere this is being discussed already. But I've been trying to understand what in the world the DOGE people could actually be doing in a matter of days with access to the US info systems? My thinking to date: - Perhaps since the election last November they've been doing some preliminary training or finding existing models to use for analysis. But with that they probably did not have access to the data, samples or perhaps even the format of such data. Not to mention all the departments being hit. I'm pretty damn sure there is not some unified code/format of systems in the US government. But perhaps I'm ignorant. - But in a matter of days or weeks, what could they be doing beyond fishing for amounts or payees/text they don't like? I mean... training models to find real fraud, that is not trivial. I mean, are we talking just feeding data into LLMs/agent structures? - Is there any world where any of what is being done could be accurate, responsible in so short time? If a truly unbiased source came in there's no way it would look anything like this. Even with highly trained and tested models. I'm curious if anyone else is wondering about this. I know there are some people that are amazing hackers and are "that good"... they can be young but are not necessarily young. Nor are they homogenous. The government like CIA, FBI, etc I'm sure has a number of them already. Lets keep this apolitical but a practical discussion of "what if" in terms of current tech... It blows my mind all the media talks about is that they are 'doing stuff' but there seems to be 0 focus on the tech/practical side. It honestly bothers me all the talk is "they're running AI on x department"... what does that mean people?!

Hi everyone! This post is my answer to a question someone asked me in one of my previous posts, and I thought it would be a good idea to talk about it. This was @Kevin Cranfield's message. Here’s what he asked: "Hi, what is your feeling about using ChatGPT/OpenAI to help you code? I started using it for little bits of help while building my Django projects. Then I figured out that I can get ChatGPT to write the whole project. I knew this was bad for my learning, but it helped me get the project done quicker." I’ll start by saying that I love ChatGPT! I think it’s an amazing tool, and I use it all the time. However, over-reliance on it can make you replaceable in the marketplace. If all you do is rely on ChatGPT to create, someone else can easily do the same. The real value comes from what you bring to the table—your ability to refine, adapt, and improve the outputs, making your work truly unique and personalized. The key is to leverage ChatGPT as a starting point while focusing on tweaking and enhancing its results to ensure your work stands out. One critical skill is learning to do things with ChatGPT that others can’t. Let me share an example from my own experience. A colleague and I were tasked with creating an automation in R. I had no idea how to do it, so I turned to ChatGPT. One thing I’m very good at is clearly expressing my ideas in writing, so I probably crafted an effective prompt. As a result, I had no trouble generating the script I needed and tweaking it, whereas the other person couldn’t get hers to work at all. Being able to write clear, effective prompts that deliver exactly what you need is a valuable skill. But there’s a downside to tools like ChatGPT: they can discourage us from learning deeply. Let me share another example. I heard a story about a wedding where the pianist didn’t actually know how to play the piano. She was there to press a button, and the music would play from the keyboard. But on that particular day, something went wrong, and the recorded music didn’t play. The violinist asked her to play manually, and she couldn’t do it! A similar situation happened to the same violinist on another occasion, but this time, the pianist knew how to play and immediately stepped in. The same principle applies to us: we need to understand the code we’re using and be able to recreate it if necessary. One way I approach this is by experimenting with changes to see their effects. I also try to write as much of the code myself as I can, and when I don’t understand something, I ask the AI to explain it, so I can better understand and learn.

Python environment management can sometimes give you that awful feeling in the pit of your stomach. So don't do it: just use Docker containers. https://www.kdnuggets.com/manage-python-environments-docker-containers

As luck would have it, I was watching a news show and the subject came up about AI hallucinations and the medical industry. I’m not sure why I hadn’t thought of the impact AI was having on medical professionals, but they too are using AI. https://pmc.ncbi.nlm.nih.gov/articles/PMC10552880/