Remember: treat LLMs less like oracles and more like muses.
Remember: treat LLMs less like oracles and more like muses[sa].[sb][sc]
Topic: llms
1,598 mentions · 615 chunks · 117 episodes
169.8× 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.
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.
- llms appears in 615 chunks across 117 episodes, from 2023-11-06 to 2026-04-20.
- Its densest episode is Bits and Bobs 2/2/26 (2026-02-02), with 15 observations on this topic.
- Semantically it travels with ChatGPT, Claude, and prompt injection attack, while by chunk count it sits between Claude; its yearly rank moved from #3 in 2023 to #1 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.
Insights that "sound good" are what LLMs give.
Insights that "sound good" are what LLMs give. Which means you have to be careful, because it's giving you a consensus answer, which might be wrong. Performative rigor.
If you just YOLO writing of code, you get logarithmic value for exponential cost.
...s a poorly architected system. Always sprinting, never refactoring. That's what LLMs do when they write code. The "always firefighting" mode of software development.
Engineers now need to be more like engineering managers when working with LLMs.
Engineers now need to be more like engineering managers when working with LLMs. How to wrangle multiple agents to a coherent result with a reasonable architecture is very different from writing code by yourself.[se][sf] When it'...
I want a tool that detects when the stars have aligned.
...to has happened. To do so requires a level of human-like judgment. But now with LLMs you can set a high-level, human intent, and have an intern who never gets bored who can keep an eye on things for you. The rules for the kinds of thi...
LLM coding assistants are best thought of as interns.
...ject. But you would have them do the implementation. That's the best way to use LLMs for code, too.
I liked this article that posits that LLMs are best seen as cultural and social technologies.
I liked this article that posits that LLMs are best seen as cultural and social technologies. LLMs are more like a market or a bureaucracy than a person. A social technology, not an agent. Tha...
The danger scales with both the amount of data and tool use.
...both the amount of data and tool use. Lots of data, no tools, little danger for LLMs. There might be prompt injections, but they can't cause anything to happen. No data, lots of tools, little danger for LLMs. There's lots of things th...
Which will win, kino or slop?
...ventions in tasteful ways.[uk] Algorithms have pulled us more towards slop, and LLMs accelerate that. At some point though might the pendulum swing back? This section is riffing on some ideas from Robinson Eaton.
Assistance (ala Alexa, Google Assistant) used to be gated on reasoning and sensing.
...la Alexa, Google Assistant) used to be gated on reasoning and sensing. Now with LLMs, it's only sensing that's left. That is, what things might matter to you in your current context. That's both a sensor problem, but also a "user havi...
I like this cool combination of LLMs plus novelty search.
I like this cool combination of LLMs plus novelty search. LLMs pull us towards the centroid; this process uses LLMs in a systematic way to skate towards interesting outcomes, not average...
The main LLMs are improving fast enough that building an adjacent business is hard.
The main LLMs are improving fast enough that building an adjacent business is hard. For example, a business to finetune models to specific use cases, or a carefull...
I loved Sam Schillace's piece about the problems that arise from viewing LLMs as humans instead of viewing them as a tool.
I loved Sam Schillace's piece about the problems that arise from viewing LLMs as humans instead of viewing them as a tool. This is as far as I'm concerned the canonical take on the pitfalls of engineers using this incorrect men...
Ideas that people bring up in conversations are ones that they implicitly vote are interesting.
...l of humanity, something interesting comes out. This is one of the reasons that LLMs can be powerful; they're trained primarily on the most distilled collective votes for interestingness across society.
A startup idea: "Mechanical Turk but the tasks use LLMs instead of humans"
A startup idea: "Mechanical Turk but the tasks use LLMs instead of humans" Surely someone is building this?
When you're writing for other people, you have to meet your reader where they are as soon as you start affixing words down.
...d it front loads modelling how to reason about that topic with others. But with LLMs, you can write your notes in whatever form makes most sense to you. The LLM will be right there with you, understanding what you intend, adapt to you...
Someone should write a manifesto that is optimistic about technology and also fundamentally human-centered.
...e a vision for computing that is all of: Optimistic about technology (including LLMs). Centered around humans not corporations. Cozy and human-scale. Collaborative and prosocial. Aligned with our collective aspirations. The manifesto ...
Consensus always pulls towards mush.
...ntroid, the average. Notably, that centroid might not itself be a viable point. LLMs are inherently a kind of planetary-scale consensus mechanism. They can give outputs that sometimes are at the centroid of a phenomena but not themsel...
I thought the Stratechery interview with Ben Evans on AI was interesting.
...chery interview with Ben Evans on AI was interesting. A few of my highlights: "[LLMs] are good at things that don't have wrong answers."[vp] "With an intern, the power of the intern is, you can tell them why they did it wrong. … One o...
Hyper concentrated insight is now more valuable than before because it can be diluted by LLMs.
...centrated insight is now more valuable than before because it can be diluted by LLMs. Hyper-concentrated insight used to be hard to consume–too sickly sweet, too hard to gulp down. Just like concentrated orange juice; much cheaper to ...