All kinds of this.
Real world tests in real world situations against humans with the same inputs.
This is part of what I was looking for — not necessarily summaries, but the ability to synthesize information from unrelated sources. I will check out Claude Sonnet, thanks. The very small sample I used to test Msty with generated some good answers but also made some simple blunders (e.g. the year some event happened in).
I’ll see how things go with Claude Sonnet but at this point I’ll be doing it out of curiosity — I realise there isn’t yet a simple, local solution for what I asked about above.
EDIT: Claude Sonnet asks the user for a mobile telephone number in order to login, which is more than what I’m willing to give them. I’ll wait.
Today it seems pretty far away, yes. But at the rate things are advancing, I don’t know. Real-time 3D for the masses seemed quite impossible at some point in time. And then, poof, 3dfx Voodoo.
We also have evidence that points at the other end. 10 years ago Tesla claimed that cars which actually drive themselves were imminent. It’s still essentially the same rhetoric now. The technology appears to have progressed in the meantime, however it will take only one major crash to annul much of this progress.
Automobiles crash on the road. 20th-century space exploration crashed in the finance-minded congress. Generative AI as a whole has not crashed yet, but the past few years of AI saga has demonstrated the extraordinary number of ways in which a crash may happen.
2 Likes
If you want your privacy assured you need to sign up for a paid account and access Claude through their API. API queries are not used for training purposes; Anthropic will even sign a HIPAA BAA which is needed for confidential medical documents.
Annul I don’t know. Slow down probably. I agree about the hype of course, but there’s a lot of real-world examples of what LLM can do well and it’s difficult to imagine all of it vanishing into self-driving cars territory. We just don’t know, that’s why I was curious as to the current state of the technology for private users.
Thank you for pointing this out.
This point is IMO as valid as its counterpoint. Historical languages and scripts — those used by literally every single member of their respective communities — have vanished over a short span (several years or a couple of decades) due to changing times. If a social tool as fundamental as a language can vanish, then there’s reason to believe that such a possibility, not a certainty, exists for any other social tool.
Concerns about energy, privacy, sci-fi-ish eschatology, impact of a potential Eurasian war upon the manufacturers of AI hardware, ethics, and technocracy are as real as the optimistic visions. Both are based upon the impressive capabilities offered by the silicon today.
1 Like
Can’t argue with that. Although I will, for the sake of argument. As I understand it, the scripts and languages you refer to disappeared when the social groups using them were either exterminated or conquered by other social groups. But LLM are not a language, it’s a technology. And technologies seem to persist over time – transformed, adapted and developed – rather than destroyed, with the exception of obsolete ones which are replaced (e.g. building a castle with 8th century techniques, although even this is being revived for research’s sake nowadays).
Hardly the place for such a complex and interesting debate, though 
I see your point, although historically it’s not only the “obsolete” technologies that gets replaced. The history of alchemy before the Scientific Revolution provides many examples of highly sophisticated technologies being lost without replacement, because their intended customers simply turned their interest somewhere else. There are, by the way, apparent similarities between alchemy and AIGC in their strongly individualistic orientation and reappropriation of precious resources.
It’s not necessarily a bad idea to have a fire temple within an unsuspecting housing district 
This is what I had in mind. I find it hard to believe that LLM could be lost or disappear without replacement, considering their potential. Therefore I believe there will always be customers for them in the foreseeable future. As to what might replace LLM in the future, this is way past my pay grade and life expectancy, I’m afraid.
The parallel you draw between LLM and alchemy is a surprising and interesting one. Haven’t seen that before. The first difference that comes to mind is scale and global impact. I don’t have to draw you a picture of the influence of tech giants on society. Although they are indeed champions of individualism and reappropriation of resources, their impact is far more global and far-reaching, I believe, than the one alchemists could have. Another reason to doubt that this technology will disappear any time soon.
Until the unsuspecting DT dev team kicks us out, sure, why not
1 Like
Our forums are a place for free and open discussion provided there’s no flaming, trolling, personal attacks, or just strays wayyyy outside the lines.
1 Like
Taoist alchemy dominated all sectors of medieval China for a couple of centuries. Most non-Buddhist Chinese families in the Tang dynasty with some money to spare were interested in consuming Taoist “elixir pills” as a means to achieve eternal life. The “elixir pills” were alchemy products, some of which poisoned emperors on their thrones. In terms of scale and impact, alchemy in ancient China was definitely greater than Generative AI of the present, which is reserved for the tech-savvy few in the foreseeable future.
AI is not truly special within the history of human technology, when you look at what it has achieved in the past five decades rather than think about what it may achieve in the future.
1 Like
I agree. But I’m focusing on what it may achieve, which seems to be quite a lot and which was the reason for my post — I thought we were already there, or at least half way, for local PDF library analysis.
I know the analogy is getting old already, but photography was very disruptive in the 19e century and I think the impact was easy to foresee even 10 or 20 years after its invention, when color emulsion (or moving film) hadn’t even been invented yet. People could rant about how the art of painting was going down the drain, but no one could possibly foresee 4K sensors in everybody’s pocket. I don’t believe considering the past impact of a technological breakthrough can tell us much about its future.
I don’t know if AI is “special” in the history of technology, but it certainly seems to be on par with other important inventions (penicillin, radio, transistors, atomic fission, just a few examples). It might be even end up being more important than some of those.
As to China, you’ve piqued my curiosity and I’ll go read up on that. Unfortunately I have zero knowledge of Chinese history, so I really can’t comment on the subject.
As you see I try to have some mention of PDF collections here and there, just to stay focused Thanks.
1 Like
Just a comment on that, because I don’t believe it’s true. I write for a living, and I’m already getting positive feedback from people who say they are tired of reading material which is obviously, to them at least, AI-generated. And these are not tech-savvy people. The impact on school papers is already non-negligible as well.
Completely agree there
I think a key point many miss is that the value of present AI is searching/finding relevant content, not creating it.
Yes, of course, at least that was what I was asking about (and why I didn’t like the “Read with AI” blurb). Searching, finding and synthesizing was what I had in mind.
Say Document A has: “Mr. Smith was born on August 1st, 1940, in Greece.”
Say Document B has: “Mrs. Jones died on January 1st, 1945. She had never left her home country of Finland.”
Either document doesn’t mention the person from the other document.
I was hoping to be able to ask “Did Mr. Smith meet Mrs. Jones?” and have an answer e.g. “Mr. Smith was five-years old when Mrs. Jones died, and since she never travelled it seems unlikely. Unless Mr. Smith’s parents had left Greece with him to Finland when he was very young.”
Of course, the idea is to have the LLM run over hundreds of PDFs, not two. But the questions I’m interested in are of the kind mentioned above. I’m not looking to have AI synthesize long documents or have them read for me, and certainly not create prose or commentary using it. As I said, I’ll wait…
This is my view too. AI will become useful when you can speak a search and it turns this into appropriate search results. I am less bothered by summarisation as the act of summarising helps me understand the subject more deeply. Where AI can help is the search/filter of information I have curated (and know it is good and correct).
If you are interesterd in searching, finding, and synthesizing you may want to look at coda.io
It’s somewhat like Notion but much more capable and much easier to integrate with other sources.
I am working on an academic authoring system. In one “Doc” I can search multiple sources of peer-reviewed literature, curate both manually and by AI, annotate manually or by AI, write my own content which also references the external citations, and save the results to Zotero, Devonthink, or its own website. All totally user configured/customizable.
Perhaps your Future has arrived
https://www.perplexity.ai/search/mr-smith-was-born-on-august-1s-6KeCGUjBS..KJavydtYxSA
Nice I knew a LLM could do that. Now just find a software that can do that with a collection of 500 PDF documents (locally, with no subscription fees) and this thread will close…
It’s important to note that the early 1940s coincided with World War II, which significantly restricted international travel and communication, especially between countries like Greece and Finland.
Hadn’t even thought of that!
Thanks for coda.io, I’ll have a look.
With no subscription fees - LOL
But for a price I do believe Coda Brain may be the closest to what you are seeking