on
Random ComputerScience
AI and Anxiety
About two years ago, I wrote this post about some of my concerns with the AI hype. The points I raised as my concerns at the time are as the following:
- Context window and hallucination
- Interfacing with various systems
- Measuring model performance on production
Regarding context window and hallucination, I think the context window issue has been improved quite a lot today. As for the hallucination, it still happens from time to time hence we need to verify the output.
For interfacing with various systems, today’s MCP servers seem to have been working quite well. Now with AI agents becoming the trend, interfacing issues seem to have mostly been resolved.
Measuring model performance on production is still trickier than traditional machine learning approaches, but at this point I think we’re already figuring it out somewhat. With the scale of LLM deployments today, I think this is not as big of a deal than I thought it would be.
Generally, I think the capabilities of LLMs today exceeded my expectations, which is a good thing for technology development. But on the other side, some very good engineers I know seem to have developed some anxiety about AI since their sense of identity had always been tied to being the person who could debug and resolve issues nobody else could. Now with the current state of LLMs, everybody can debug and resolve hard issues relatively well with the help of LLMs and agentic coding AI.
So, what to do with that anxiety?
Honestly, that’s something I can’t really relate to because in my life and career journey up to this point, I never was the guy who could debug and resolve issues nobody else could. I think I have always been considered a pretty decent engineer by the standards of the organizations I have been at, but I don’t remember being considered as the guy with exceptional skills that nobody else can match aside from my ability to single-mindedly pursue a goal for a very long time (which when aligned with the company goals, I could be a pretty strong contributor but not necessarily technically the strongest).
So I have always been anxious about never being able to match the level of expertise of the people who are very dedicated towards their craft, considering that I see myself as a guy who happens to be able to do tech stuff instead of an absolute beast of a techie (I can name many people way stronger than me that I know personally, some are way younger). Hence, adding AI into the equation doesn’t really change anything there since I really have no pride in being the strongest expert at one thing.
I think most of the value I provide comes from my broad knowledge of different topics which helps me to connect with quite a lot of people with various interests and allows me to align their interests when working together with them. I was actually surprised when I came to the realization, considering that for a significant part of my life I was considered a loner kid who couldn’t really connect and communicate with people very well.
Can I give advice to people with that anxiety?
I don’t know. I’m not in their shoes so I can’t really say if my guess about how they’re feeling is truly accurate. Even if I’m in their shoes, I probably can only speak for people with cases very similar to mine, who probably aren’t the majority of the people affected with the AI-related anxiety.
But if there’s anything I can say, it’s that being anxious is human. Everyone is anxious about something at some point in their life, this time it just happens to be collectively about AI. And most of the time, being a human means having a lot more facets and aspects in our lives than the part that makes us anxious, so we probably can start from there and see how to realign everything in a way that can lead to an outcome we can be content with.