Even (very) noisy LLM evaluators are useful for improving AI agents.
... improving AI agents. Another example of noisy-but-biased signal still giving a gradient if you have enough of it.
Topic: gradient
114 mentions · 13 chunks · 12 episodes
21.4× distinctiveness vs baseline
?
How much more common this term is here than in ordinary English. Higher values mean the topic is more characteristic of this corpus.
4.2x burst in 2025 Q4
?
Peak quarter intensity across the topic's active span. Higher values mean attention was concentrated into a shorter stretch rather than spread evenly over time.
Topic summary
?
A short read on the topic's time range, peak episode, and strongest associations. Use it as the quick orientation before drilling into examples.
- gradient appears in 13 chunks across 12 episodes, from 2024-06-03 to 2026-06-01.
- Its densest episode is Bits and Bobs 10/27/25 (2025-10-27), with 2 observations on this topic.
- Semantically it travels with goodhart law, north star, and network effect, while by chunk count it sits between slop and combinatorial explosion; its yearly rank moved from #140 in 2024 to #95 in 2026.
Over time
?
Raw mentions over time. Use this to see absolute attention, not relative rank among all topics.
Observations
?
The primary evidence view for this topic. Sort it chronologically when you want concrete examples behind the larger pattern.
End-to-end encryption in messaging is now table stakes.
...s an abstract benefit, and has no concrete downside for users. That created the gradient where it could grow without bound. All it needed was the right seed crystal to kick it off. If it had concrete downside, then that would have dominat...
If a system is open but so different from everything else, it is effectively closed.
...e, it is effectively closed. It must be possible to adopt a new system across a gradient.
The gradient of competition can only give users what they want, not what they want to want.
The gradient of competition can only give users what they want, not what they want to want. Users get to decide what to purchase. They'll decide to use the one th...
If you follow the gradients of optimization you get what people "want" not what they "want to want."
If you follow the gradients of optimization you get what people "want" not what they "want to want." Don't drive something that matters off a cliff, or let them drive themselve...
There's a class of shortcuts that are good for the individual but bad for the collective.
...llective. Those are what will be taken, according to Goodhart's Law. That's the gradient that the swarm will follow. An individual who doesn't care about the collective (or feels that it doesn't make sense for them to do an action that's ...
Enshittification is inevitable for successful products.
...tification is inevitable for successful products. Over time they roll down that gradient. It's a one-way process, a ratchet. The only question is how quickly it rolls down.
It's not that Goodhart's Law just so happens to find shortcuts.
It's not that Goodhart's Law just so happens to find shortcuts. It's that the gradient the swarm descends fundamentally are shortcuts. The ideal vector is what all of the members of the collective, if they didn't know which member of th...
Data can only tell you the gradient, not where to go.
Data can only tell you the gradient, not where to go. If you just follow the gradient, then you'll random walk through the problem domain. If it's short-term data, (e.g. UXR for accepta...
Surprise is the gradient of improvement.
Surprise is the gradient of improvement. Surprise requires a mental model to be surprised in the first place. If you are incurious about the world around you you can't be sur...
Competition is the gradient of improvement.
Competition is the gradient of improvement. Once no one else can plausibly compete the drive for improvement is gone.[ahp][ahq][ahr] That's how you get stagnation.
If your thing looks superficially like existing alternatives, it's hard to get going.
...lly like existing alternatives, it's hard to get going. There isn't enough of a gradient to differentiate yourself. But if your thing has an internal network effect that the other things lack, then over time that network effect will get s...
Surprisal gives us a gradient we can choose to follow.
Surprisal gives us a gradient we can choose to follow.