It is interesting to see how research on the web has changed over the last two years. ChatGPT 3.5 was impressive but not a reliable source of information back then, but it improved with version 4, and now with o1, it becomes more and more convincing. However, LLMs - inherently - may produce incorrect or even nonsensical information.
Nevertheless, it has changed how I search for information, partly because Google's search page has become subjectively bloated. So, here's my current workflow for doing research on the web:
%%{init: {'theme':'neutral'}}%% flowchart TD Z[❓I have a question] --> Y[\🤔✔️ Will it be fast and easy to use Google?/] Y -- No --> A A[🤖❓Ask question to ChatGPT] --> B[\🤔✔️ Does the result look plausible?/] B -- Yes --> C[✅🧪Accept answer\nand possibly test it yourself] B -- No --> D[📚❓Ask for sources] D --> E[📚✅Check sources and compare\nwith answer] E --> F[\🤔✔️ Does it look plausible?/] Y -- Yes --> G C --> F F -- Yes --> H[ 🎉✔️ Done] F -- No --> G[🌐🔍 Search the web and find out yourself,\nlike in the good old days] G --> H
Here's a simple example (English translation below):
Click for an English translation
- Question: “Is the Fritz WLAN Repeater 310 mesh capable?”
- Answer: “No, the FRITZ!WLAN Repeater 310 does not support Mesh WLAN. For mesh
capabilities, you need a device like the Fritz!Repeater 1200, 2400, 3000, or similar
models that are mesh-compatible.”
- Follow-up Question: “Source?”
- Answer: “The FRITZ!WLAN Repeater 310 supports mesh functions. When used with
a FRITZ!Box running FRITZ!OS 7 or newer, the repeater is automatically
integrated into the mesh and adopts the WLAN settings of the FRITZ!Box. In
the overview of mesh-compatible FRITZ! products, the FRITZ!Repeater 310 is
listed as a supported device.”
- (Source: AVM)
In the example above, I started with ChatGPT, didn't trust the result and asked for sourced and corrected itself and provided the right link to the manufacturer's technical page. In this case, a simple web search would have been just as quick, but I must admit, the extra layer of abstraction and summarization functionality that ChatGPT provides compared to a search website offers a nice user experience.
Balancing Trust and Sustainability 🔗
Ultimately, it’s about deciding when to trust ChatGPT for accurate results during your research: for finding local shop hours, Google is more suitable. For specific technical questions that require context, I now start with ChatGPT. However, always remain cautious and verify the results along the way.
One consideration, naturally, is the environmental impact: new AI tools are undeniably power-hungry. Compared to a simple Google search query, this article suggests the increase ranges from multiple dozens to hundreds — an LLM query might consume between 0.002 kWh/query and 0.005 kWh/query. This is quite significant, but I don't believe this means we should avoid using AI entirely—the ship towards inevitability has sailed already. However, it's important to be mindful about the implicit costs and use it only when it makes sense - like a good engineer would do anyway.
Feedback 🔗
What are your thoughts? How do you use ChatGPT et al? Feel free to get in touch with me!