Recently ChatGPT is very hot. And I tried to use Notion about its AI feature. It is very useful. Will it be used in DEVONthink?
To do what?
No, there are no plans to integrate with ChatGPT. See…
PS: Corporately, we don’t jump on 'hot ideas. Technologies come; technologies go. Same goes for applications.
The problem with AI as a tool for creation or intuition is that AI doesn’t actually do anything. It feeds off what’s available. We’re told that media will become dominated by algorithmically generated output, which will surely feed back into AI learning.
As that Maypole dance intensifies, AI will turn coprophagic, consuming recursively digested information through accelerating orders of iteration.
AI has great utility and great limitations. I think it will have its greatest utility if we keep the limitations firmly in mind.
Searching and classification is probably going to be AI’s best contribution. I’m not sure if that’s appearing in applications yet or not. Maybe my “see also” tab will shed some light…
Craft uses GPT and I’ve been using it a bit. I should say I used it a bit until I lost interest. One problem with the originating services of GPT is that they charge users for each search, several pennies per. Craft has a limit per month on the number of searches for their users (since Craft must pay for user searches, this probably figures into what Craft charges for yearly subscriptions). I think the limit is 250 searches for the Pro plan and 50 for the basic plan.
If DT adds that to their service, someone will have to pay, it’s not free.
I think the limit is 250 searches for the Pro plan and 50 for the basic plan.
Not that I’m interested in messing with it, but those limitations would be easily reached just in trying to figure out what it may be able to do.
I share the concerns of many that creation of. spam-type marketing material is not the peak of software success. I also agree that this part of AI may be a fad.
But that said - no doubt AI is here to stay and we are about to embark on a wild ride as notable as the invention of the printing press.
I do believe there is a tremendous potential marriage between DT3 and OpenAI since fundamentally GPT3 is a language model - and language is at the heart of DT3.
My fantasy integration of DT3 and GPT3 would be the creation of an AI/GPT-3 Inspector. There you could define a prompt to be run on the document, with the output placed in the inspector panel. Of course in a perfect world the AI could be implemented via SmartRule or scripting and the output could optionally be placed in some other specified location in DT3.
Once implemented as such potential GPT3 prompts would be wide open to imagination including:
- Offering editing suggestions
- Adjusting the reading level of a document
- Adjusting the tone of a document
- Language translation
- Semantic search (within an entire group or database - not just at the document level)
- Summarizing documents
I believe coda.io has the competitive edge in this situation over Notion because Coda’s “Pack” system makes it easier to connect to the Internet and other applications than Notion’s API. These are super impressive videos regarding Coda’s current and future capabilities; many of these abilities arguably could be implemented in DT3 using smart rules and scripting:
The software industry appears to be adopting a usage model for integration with GPT where you create a Secret Key for the OpenAI API and provide that secret key to the host software application in secure form. Thus fees are the responsbility of each user and the host software company need not set up metering.
It appears to me that cost of OpenAI is only likely to become a limiting factor if you want to create a public-facing website or if you wish to train your own language model using embeddings or fine-tuning; only the most extreme hard-core AI users would do any of those.
I have been using OpenAI pretty compulsively over the last month experimenting with a host of ideas and very liberally running prompts on test documents. It took me 5 weeks to use up the initial free $18 credit given by OpenAI. I find that on a typical day my charges are under 50 cents and on days where I have done a large amount of testing I have never exceeded $1.50.
I began using OpenAI for entertainment, but now I find myself using it for a wider range of tasks. OpenAI’s API has become more affordable after the release of GPT 3.5. I noticed that incorporating AI into tools provides an obvious benefit over those that do not.
I look forward to a future where I no longer have to organize information using tags and folders. I hope it is a few weeks away.
I hope it is a few weeks away.
Should I let you down easily?
Hey @BLUEFROG just what is the problem? Surely one of those AI apps can write all the code and documentation and then do all the follow up support for you! I don’t get why you are not jumping at this opportunity!
We can pay for it with our DEVONthink NFT profits!
AI, AI, AI (pronounced "Ay, Ay, Ay”).
AI methods are maybe OK as long as they do not require Internet access. I don’t want to report my research to anyone. I swear, I’m that close (holding up fingers like Maxwell Smart) to perpetual motion!
I kid of course, but I have separate projects in separate DT databases. I wouldn’t want to poison the AI for one with learned cognition from another.
Or, maybe I would like to poison AI. It would be the perfect murder. No body makes for a difficult conviction, doesn’t it? Just asking for a friend, I wouldn’t really want to do anything like that.
You can write a ChatGPT integration with DEVONthink in a few hours.
Take a look at my example: Inspector Search in files with long lines - #10 by jsn
Whatever you want to do, it will help you write the code to do it.
If you want your documents to automatically be summarized, you can use my code.
If you want to have tags created for your document that are meaningful, you can create a prompt for that and have GPT send back the tags and you can apply them automatically.
I’m sure you can have it generate DT3 searches for you or do other tricks to automate your workflow.
Aren’t you a ray of sunshine @jsn !
This certainly looks interesting and I will indulge your good work this weekend. Thank you for sharing it here
I stopped bothering to read the other responses as I don’t relate to the sarcasm. And knowing that I am none to argue with such brilliant minds present here that dismiss new technologies!
I remember the same cycle when backlinks were catching fancy, and the response was “if it ain’t broke, don’t fix it.” Maybe the great DT team are the hipsters who insist on using their flip phones. That’s cool too!
I still love them (no other choice as I bought the license ). That counts for something, right?
It certainly is staggering how the DT development and community management in general is happy with the status quo of the DT3 product and the glacial pace of anything new or innovative to wrangle all the content in the DT libraries.
While it has become more powerful over the years, there are still core functions that feel like they’re stuck in 2014 - sync as an example.
Integration into LLMs to help find context in the content, or use local content as a source of information would be incredibly valuable. The question is, would it make more sense to wait for DT to build the capability/integration or would it make more sense to build a plugin framework and allow the community to build and maintain the plugins. Imagine the horror of having the community be allowed to build plugins - that would be way too forward thinking. Even worse, the plugins might even start adding value to a stagnant development cycle!
Do you have a specific problem with sync now that you need help with now?
I’m all for open source and community activity. And there is already sample code to integrate ChatGPT with DT, IIRC.
OTOH: Just a cursory glance at GitHub will show you tons of under-documented, never finished and abandoned projects. In general, community projects work well if the community is large enough. And if there’s someone taking the role of “benevolent dictator” (Torvald’s expression, not mine). The DT “community” (I don’t like this word because it tends to imply something that just is not there), or rather the DT users, have wildly different skills and wildly different interests/approaches to organization and wildly different data to work with. For example, I use DT nearly exclusively to handle my finances and travel planning – what use would I have for ChatGPT there?
The sheer number of people able to do what you suggest is, I believe, tiny. And then not all of them might be inclined to work on plugins in their spare time (if they have any).
But if you feel that such an effort is needed, why don’t you start it? DT is scriptable, and who would stop you writing scripts that integrate it with ChatGPT or any LLM? I frankly do not understand what makes you insinuate that you need anybody to “allow” you to do that.