Semfont Demo
It’s been a while since my last post. Over the last months, I’ve been taking some time off here and there and spending a lot of time digging into data architecture. Some of the old insights still hold up, but AI is also turning parts of the field on its head. However, there are some posts in the pipeline again.
Recently, I came across a post on Hackernews that looks like a more friendly way to display text. Note that it is not based on AI, or embeddings, or some sort of emotionally weighing. It looks up words in small lexicon within the package, and performs a couple of clever local rules to account for negation or contrast.
Some LLM-generated text:
Great, another outage. Just what I needed today.
We fixed the crash and closed the security hole before anyone noticed.
The cluster recovered from the crash in under a minute, but the deploy probably still needs a manual check.
I would not go so far as to call the new editor great.
Less broken than last week, and far fewer complaints.
In that example:
| Visual cue | Meaning | What it helps you read | Example in this post |
|---|---|---|---|
| Red text | Negative or risky language | Problems, failures, warnings | outage, great in the hedged sentence |
| Green text | Positive or improving language | Fixes, recovery, success | fixed, recovered, broken |
| Bold text | High salience | Important terms, repeated ideas, topics | security |
| Yellow highlight | Surprise or contrast | Turns in the sentence, hedges, exceptions | deploy, probably, manual |
| Italic text | Uncertainty or hedging | Soft claims, probable statements, caveats | probably |