Hacker News Reader

Comment by mmooss | Hacker News Reader

Britannica has a uniquely high-quality collection of general knowledge. Many of their current articles are written by domain experts, sometimes leading experts. (In the past, people like Albert Einstein and Sigmund Freud wrote articles for Britannica.) They also own Merriam-Webster, a highly-respected dictionary.

I wonder if that content trains their AI and how much difference it makes.

Also, is there value in training an AI on a dictionary? It would lead to poor writing style.

View on HN

Errors in the Encyclopædia Britannica that have been corrected in Wikipedia:

https://en.wikipedia.org/wiki/Wikipedia%3AErrors_in_the_Ency...

This is a great approach if an encyclopedia represents a very large proof about the world, but is that what it is?

Every model is trained in Wiktionary, a larger and higher quality dictionary. Every model is trained on Wikipedia, a larger and higher quality encyclopedia.

"Train an AI" is a bit of a red flag, fyi, as a phrase. What is an "AI"? Do you mean "large language model"? Something else?

> Wiktionary, a larger and higher quality dictionary

> Wikipedia, a larger and higher quality encyclopedia.

What is the basis for those claims?

In my experience with them, both have clearly less accuracy than Merriam-Webster and Britannica, respectively. But that's only anecodotal.

This is a large subject of course but the consensus I have always seen is that wikipedia is better because it has more eyes on each subject that can find errors.

https://en.wikipedia.org/wiki/Reliability_of_Wikipedia

>Opinions on accuracy were almost equal between the two encyclopedias (6 favoring Britannica, 7 favoring Wikipedia, 5 stating they were equal), and eleven of the eighteen (61%) found Wikipedia somewhat or substantially more complete, compared to seven of the eighteen (39%) for Britannica.

> the consensus I have always seen is that wikipedia is better because it has more eyes on each subject that can find errors.

More eyes can help, but aren't necessary, they are not sufficient, and they have a downside or two

* In the post-truth age of automated mass disinformation, majorities can be completely wrong (probably they could do that previously, also). Imagine trying to write about climate change or vaccines or trans health for certain populations; the many eyes often bully and abuse people into silence.

* The nuance of an expert is often lost in editing by a mass of non-experts, who don't understand it enough to know what they are missing or to state it in other ways and contexts.

* Look at sports: The masses sometimes strongly disagree with decisions and actions that are (nearly) provably correct. The coach, the expert, knows better.

On the upside, Wikipedia has much greater breadth, but personally, I never know what I can trust.

Do you have an anecdote? I'm sure you're right and there are Britannica articles that are more accurate than their Wiki counterparts, but it'd be interesting to see an example.

It would be an interesting use of AI tools to try to flag differences between the two as a means of identifying gaps in coverage or factual mistakes (in either one).

And I specifically mean flag. As in, for human review. The last thing Wikipedia needs is spam AI edit submissions (it's probably already happening).

I don't see the value:

If you mean the anecdote is evidence, it's much too small a sample to be representative of anything.

If you don't believe any Britannica article is more accurate than a Wikipedia counterpart, you have other issues. :)

Reply on news.ycombinator.com