My partner is AI

Introduction

This is about ChatGPT.

Some time ago, I was skeptical about AI because ChatGPT handled data analysis, research, and sourcing primary references very poorly. Now, however, these skills have improved significantly. Competitors like DeepSeek lag behind on many fronts and constantly “hallucinate.”

Nonetheless, in narrow specialized fields ChatGPT still has issues, as many researchers note (I follow numerous chats and channels across different disciplines). It can make mistakes even when solving medium-difficulty problems in algebra or field theory. Hell, sometimes it even messes up basic algebra problems!

Practical Utility

These days, ChatGPT already saves a lot of time, including in work processes. Tasks it still doesn’t handle well:

• Researching and compiling dossiers on individuals or companies
• GEOINT (geospatial intelligence)
• Building phishing chains based on the latest trends
• Document analysis

Where it can help:

1. Large-scale data analysis (though it’s better not to upload entire files; send smaller fragments.)
2. Idea generation. Suppose I have a problem and need ideas or algorithms to solve it. I sketch out 5–6 options, then ask ChatGPT for help, and it offers several more—some I hadn’t even thought of.
3. Getting up to speed on new OSINT tools.
4. Writing parsers and other scripts. There’s a caveat here: a professional developer friend of mine can easily spot the difference between code written by ChatGPT and code written by, say, a mid-level engineer.

Important: Never share confidential details of your project.

What factors will influence the output, the quality of the information provided, and more effective interaction with AI (I won’t discuss different versions—that’s obvious)?

• Prompts. The more precisely the task is defined, the more useful the response. Ideally specify goals, types of sources, output format, and a description of the task. This also includes follow-up questions and refinements.
• Temperature setting. ChatGPT’s “temperature” controls randomness and creativity:
• Low (e.g., 0.2): favors common, predictable answers; best for tasks requiring accuracy and reliability, such as technical questions or code generation.
• Medium (e.g., 0.7): balances creativity and reliability.
• High (e.g., 1.0): increases creativity and explores less common possibilities, which can yield more unique (or hallucinated) results.
• Plugins or custom API integrations (e.g., web-search plugins).
• Structured data is generally processed more accurately than “raw” data.

Use Cases

Case 1

Problem: extract a list of employee names, positions, emails, and phone numbers from a company webpage that contains dozens or hundreds of entries. Manually it would be tedious, so we ask ChatGPT to help:

1. It fetches the page HTML.
2. Parses the DOM with BeautifulSoup.
3. Uses regex to find matches for our data patterns.
4. Outputs in the desired format.

One might argue, “We have plenty of parsing tools like Octoparse, ParseHub, WebScraper, Data Miner, etc.” True—but they aren’t always convenient to set up or use. It’s a matter of preference.

Case 2

Problem: you have 200 links discussing the scandal—80 in English, 50 on Reddit, and 70 on Russian-language forums. You need to:

1. Download the texts.
2. Translate everything into Russian.
3. Cluster the texts by theme.
4. Provide a brief summary for each cluster.

How ChatGPT can assist:

• Parsing and data collection: Suggest a combination of requests + BeautifulSoup for forums and Reddit, plus tweepy for the Twitter API.
• Translation: Use the DeepL API or ChatGPT’s built-in translation plugin to normalize the language.
• Clustering: Transform texts with a sentence-transformers model (e.g., all-MiniLM-L6-v2).
• Summarization: Ask ChatGPT to summarize each cluster. If a cluster is too large, sample a few representative texts and request an overview.

Critical Note

Every AI-generated step and output must be verified with critical thinking—so you’ll still need to engage intellectually. But the time savings are real.