When using OpenAI Deep Research for history topics I usually ask it to also include contemporary quotes and articles. It can do this but somehow the extremely rich archive of English newspaper articles on nearly every subject is never used or returned.
This is awful to me. We went to all the trouble of producing maybe 50 detailed journalistic day by day evaluations of the world for hundreds of years running now, but their present day owners aren't making it at all easy to use them to study history!
It seems to me that a second tier paper, if it licensed its reporting to be used in a top LLM, could leapfrog the NY times as the "paper or record" (a place of respect which the paper forfeited when it became clear it was a puppet for the regime during Iraq war two about WMD)
Since then its glory has faded and it simply awaits the coup de grace.
Judgment
If history queries to a top LLM in the US start being able to return quotes and articles from a wide span of history, from a single (or many) US based English language major newspaper or news magazine (back to 150 years prior or earlier) such that even relatively insignificant (at the time) articles in say the want ads prove to be available to it, then YES. This should also be as a result of that media source somehow opting in to this role.
Deadline: June 30 2029
Otherwise NO at deadline. Can be any public LLM or agent or other new type of research report generating system.
Sample queries which should hit:
Give a week by week summary of any mention of "hot dog stands" in NYC in (the local newspaper which satisfied the claim), from 1890-1990, listing connotation, class, average price of any, and any other useful info
Identify ten soldiers featured as returning from WW2 in 1945 to towns in the Midwest. (If we got a Chicago paper for example) Note down their names and ages. Then trace any mention of them through their lives giving a life timeline for each, continuing through any descendants, covering civic activity, family, achieve, influence, created works and life etc
Or, this query about bullet types in WW2:
https://chatgpt.com/share/67cb71fe-2b58-8003-9c72-c16afad3b5f9
The report is great but I think lacks lots of probably very interesting journalistic, media, interview and science tech based articles and quotes about its subject.
1,000People are also trading
Media companies, like other publishers (especially academic publishers), want to know if their content is going to make another product more valuable, that they get a cut. They're not going to make the "freemium" mistake again. (This is my somewhat informed guess as to their thinking, not a judgement on my part.)
@WilliamGunn Yeah, it's just surprising that of all of them, none is going this route. i.e. If I were always reading historical summaries from a certain paper, in every single historical novel, history report etc, which were from the paper on that day, it seems to me it'd raise their profile in my eyes. Is the washington post (founded 1877) really making much off its back catalog?
Hmm, my 150 year time limit knocks out a few of the top papers which are right on the cusp. Still, by 2029 the Wash Post would count. The market is only about USA papers but here FYI is a list of large papers including the UK.
And here is a list generated before I filtered for papers which also include a large international reporting staff - there are probably way more of these super old papers still in business, at least doing some kind of job reporting local or national news for 200+ years.
@Ernie Some companies are signing licensing deals with LLM providers. It's just a slow process because both side are negotiating & OpenAI dramatically undervalues curated and quality-controlled info relative to stuff like reddit posts and synthetic data. They cynically believe that subscribers won't know or care enough to notice the difference.
@WilliamGunn yeah, I mean it really is early days. There are a lot of ways things can go, and the basic structure of rights is pretty important.
Noble effort holding the line of assumed subscriber preferences but, how many times has that side been defeated (tiktok, instagram, etc.) I don't care what the masses do, I just want to get access of some type to all that data