Comment by trynewideas | Hacker News Reader

Comment by trynewideas | Hacker News Reader

Asked the same history question[1], "what was the french-speaking population of france during the french revolution", multiple times and got different answers each time. Each answer was wrong about at least one significant detail, and in one answer it decided to just not try to answer the question I asked.

Attempt 1: The population was about 25 million (in line with estimates) and only 3 million spoke specifically Parisian French (a potentially accurate statement, but not what I asked). The sources it cited had contradicting information — 23 million instead of 25 million total population — or additional context that would've answered the question accurately, such as about half of the population speaking some form or quality of French.

Attempt 2: "The French-speaking population of France during the French Revolution is estimated to have been around 29 million people" — completely wrong across the board. One cited source is a Statista user-created graph that gives 24.8M for revolutionary France and 29M for France in 1800; it also has no details about language use. "This figure is based on demographic analysis conducted by Gudin and confirmed by a crosscheck with the graph showing population changes in France between 400 B.C. and 1975 contained in Colin McEvedy and Richard Jones, Atlas of World Population History." This is lifted almost verbatim from another cited source, but the source's next paragraph acknowledges discrepancies important to the context, including potential errors by Gudin or his translator. The source also has no details about language use.

Attempt 3: "The French-speaking population of France during the French Revolution was estimated to be around 55.81 million people, according to a 2007 Adult Education survey by the Insee. This survey found that French was the mother tongue of 87.2% of the total population, followed by Arabic , Portuguese , Spanish and Italian . People who spoke other languages natively made up the remaining 5.2% of the population." Completely wrong. The only cited source is the "Languages of France" Wikipedia article.

1: https://beta.sayhello.so/search?q=what+was+the+french-speaki...

There's zero attempt to cache answers or cited sources, I suppose, because I keep getting different answers even from the same cited source just by clicking the search icon on the bar with the same query:

"The French-speaking population of France during the French Revolution was estimated to be around 25 million people" (wrong, citing that same Statista graph that doesn't mention language use)

"It is difficult to accurately estimate the French-speaking population of France during the French Revolution, as there was no official census taken at the time." (honestly the most accurate answer yet, but citing that same Statista graph)

"It is impossible to accurately estimate the French-speaking population of France during the French Revolution, as there is no reliable data available. However, it is estimated that around 80% of the population spoke French in 1800." (first sentence is correct, second is wrong, citing that same Statista graph)

"At the time of the French Revolution, less than half of France's population could speak French. This was due to a combination of factors, including the fact that French was an extremely popular language with the elite and higher classes, as it was adopted by nearly all European courts and it even reached the other side of the Atlantic." (cites a babbel.com article; I think the first sentence is the first time the answer approaches both accuracy and consistency with both the question and source, largely by plagiarizing part of the source. It still omits useful context about the lack of literacy/fluency that's in the same sentence lifted from the source. Then it goes back off the deep end with the second sentence, which the source uses to emphasize how broadly French was spoken, and not as a reason why so few people in revolutionary France spoke French.)

Interesting, we are trying to wrangle with the nondeterminism but sometimes it can't be helped as Bing itself can produce different results. Always actively working on the model though.

The citing sources feature can definitely be improved - right now it works at a sentence granularity and insists on finding the best source when sometimes not appropriate. Thanks for pointing all of this out

Thanks for trying it out - we're still quite early so the model isn't going to be perfect + we're focused on programming related queries at the moment. What is your use case? Are most of your searches related to history?

Most of my searches are related to researching and fact-checking potentially spurious statements, like the ones that Hello apparently produces.

This particular query about the French revolution seems to give GPT fits across its iterations, and I suspect it's because:

- the most correct answer is "I don't know"; when answering questions, GPT isn't really trying to answer a question by reasoning through it, it's trying to mimic what usually happens after someone asks a question, and especially online that usually isn't someone saying "I don't know"

- the most authoritative sources on subjects like this include books, and GPT doesn't seem to read a lot of those, or if it has then it doesn't seem to be able to cite them without inventing false details about the sources themselves (titles that don't exist, attributing works to the wrong authors)

Seeing "model that cites sources" made me excited that there was a solution to the second point. But specific to Hello:

- the most authoritative web sources are historians who cite sources, which aren't often written for SEO, on popular platforms, or not paywalled

- a search result being in a first page of Bing/Google doesn't make it authoritative

- this doesn't seem to stop GPT from coming up with false inventions/hallucinations that aren't in or relevant to the cited source at all

You're right - there are so many good sources not often surfaced by a cursory web search like books and experts. Adding better sources is something we're improving on. A simple next step we're looking into is to expand sources to research papers e.g explicitly pulling from arxiv on some queries.

Reply on news.ycombinator.com