| Include LLM feature [message #1903] |
Thu, 13 February 2025 05:18  |
vattha
Messages: 1
Registered: February 2025
|
Junior Member |
|
|
Dear Devs,
Does FoxTrot plan to include the large language model? Something like ChatGPT but exclusively focuses on my documents.
|
|
|
|
| Re: Include LLM feature [message #1904 is a reply to message #1903] |
Thu, 13 February 2025 14:41  |
Atlas
Messages: 150
Registered: August 2009
|
Senior Member |
|
|
I would like to hear Foxtrot Engineering's thought on this. LLM's and search tools like Foxtrot are not the same thing, but they complement each other. Search tools will only return search results that are actually in the text; where as LLM's generate approximated summaries that may not reflect what's actually in the text. Thus, search tools like Foxtrot can be used to validate and create ground truths for LLM's, but I don't think LLM's can replace search tools.
I've been thinking of ways to fit Foxtrot more elegantly into my research process, but I'm still trying to figure that out.
|
|
|
|
| Re: Include LLM feature [message #1905 is a reply to message #1904] |
Thu, 13 February 2025 20:03 |
CTM info
Messages: 187
Registered: September 2009
|
Senior Member |
|
|
Hello,
For over 20 years, we’ve always had a policy of not discussing unannounced features or possibilities, so we shall stick to that.
If you have suggestions on how you think a feature could integrate with FoxTrot, including workflow, user interfaceand expected benefits, we’re always happy to consider and feel free to post them via email directly to us or on this forum.
One particular area of interest, as we are quite privacy-focused and take pride in making sure that no one else but you searches your own indexed data,,would be thoughts about uploading parts of that index eddata to a large language model. Which ones are trustworthy? What type of questions would you be comfortable in asking them regarding the data that you have indexed in FoxTrot?
thank you.
jean michel/ctm qa
[Updated on: Fri, 14 February 2025 20:06] Report message to a moderator |
|
|
|
|
|
| Re: Include LLM feature [message #1909 is a reply to message #1903] |
Tue, 25 February 2025 11:13 |
arusim
Messages: 6
Registered: June 2021
|
Junior Member |
|
|
|
Honestly, when people talk about using LLM for searching, I think what they really want is just semantic search. However, this feature must retain the speed, flexibility, and convenience that FoxTrot Search currently offers.
|
|
|
|
| Re: Include LLM feature [message #1943 is a reply to message #1909] |
Thu, 01 May 2025 17:09 |
Grant Barrett
Messages: 35
Registered: October 2019
|
Member |
|
|
I have been thinking about this, too, as I have tried all the other LLM solutions that purport to be able to index my documents and offer me insights. And what bothers me about most of them is:
—They want me to put my documents in the cloud. I don't want that because they are mine, there are too many, and their total byte size is large.
—They can't handle my volume of documents. I constantly change, update, delete and add to my documents in my own space, time, and way.
—They don't want to use the way I already have my documents stored and sorted. I have a background in certain professions that gives me expertise that many software developers don't have; they should trust my (and all users') preferences and knowledge, if possible, or at least let the user have the option to go their own way.
—They aren't always using platform-native tools already in place. I don't want to have yet another copy of the same open source models installed in yet a different place on my hard drives, or another version of Python, or another copy of tools that I have taken care to get from the right vendors in the right versions and keep up to date in the right ways. (With caveats about env, pyenv, etc.) They should do a lot better about trying to work with what is already there.
—FoxTrot Professional has already done a lot of the heavy-lifting of figuring out importers for many of the key file types. Many of the LLM vendors are way behind even on that one fundamental thing.
—Similarly, my personal OCR is better than the LLM-based software makers trying to work in that space, usually, at least at what I am willing to pay per page. Yet they will throw a barely tweaked version of pytesseract at a PDF and call it OCR. I work with multilingual dictionaries, and they can't be programmatically handled in that way. (I recognize that there is a whole business-to-business market that does not care about my academic and personal OCR needs! That's okay.)
So! When I think about FoxTrot as an on-ramp to some sort of LLM document-examining-using software, I see that it has some distinct advantages. I hope Jerome and the crew figure out something excellent!
|
|
|
|
| Re: Include LLM feature [message #1944 is a reply to message #1943] |
Thu, 01 May 2025 17:27 |
CTM info
Messages: 187
Registered: September 2009
|
Senior Member |
|
|
Thank you Grant for this useful input.
May we ask you to elaborate a bit on what you mean by "some sort of LLM document-examining-using software" ? What exactly are the type of prompts that you would want to throw at your indexed data set, if you can write up five to seven typical prompts it would help us greatly.
Kind regards,
jean michel/ctm qa
PS: Feel free to respond via direct mail if you prefer, you can create a support case from within FoxTrot Pro for this
|
|
|
|
| Re: Include LLM feature [message #1945 is a reply to message #1944] |
Sat, 03 May 2025 17:44 |
AJKS
Messages: 58
Registered: June 2020
|
Member |
|
|
I have experimented with AI/LLM software, including a couple that do local document analysis and search, but I haven't been impressed by the results.
One of my interests is in journaling and diaries, also known as ego-documents and life-writing.
The software I used couldn't discern between real historical diaries and fictional ones.
That's pretty critical. I spent a long time trying to discover more about a particular diarist before realising that the diarist was fictional.
Note:
I used the free versions of these programs, maybe you get better results when you pay more?
I have a lot of scanned PDFs so OCR isn't aways perfect, that would always be a problem.
Maybe I don't know enough about AI and LLMs to ask the right kind of questions.
The software I used didn't permit me to export or save results.
The software couldn't give me precise information about the sources used.
I was trained to document my sources, and to use legitimate and authoritative sources.
Being able to cite the document etc is key.
Maybe using AI can help me to get a wider picture and consider some things I wasn't previously aware of.
But so far I'm not impressed by local document search and analysis.
Just my experience so far.
[Updated on: Sat, 03 May 2025 17:46] Report message to a moderator |
|
|
|
] 
FoxTrot Search Forum
Search
Help
Members
Register
Login
Home
