Why I Stopped Following the Frontier

For most of 2023 I had a habit. Every morning I’d wake up, scroll through Twitter, and try to absorb whatever had broken in AI overnight. New model releases. New benchmarks. New papers. New companies pivoting into agents. New companies pivoting out of agents.

I told myself this was professional development. That I was staying sharp. That a person working in AI needs to know what’s happening in AI.

Around November I noticed something. I knew the names of about forty new model variants. I had not built anything substantial with any of them. The hours I had spent reading takes had not produced takes of my own — they had only produced opinions about other people’s takes.

The frontier moves faster than any individual can absorb. The dirty secret of working in this field is that almost everything you read about will be obsolete in six months, and the things that won’t be — the things that genuinely matter — are usually not the things that trend.

So I made a rule for 2024. I read papers on Saturday mornings. I do not check ML Twitter the rest of the week. When something genuinely important happens, it reaches me through people I trust within a few days, and when it reaches me I have the time and attention to actually engage with it.

I am building more. I am thinking more clearly. I no longer have informed opinions about the agent ecosystem of the third week of February. I think this trade is good.