I just read through the email blast from Greg. The thing that stood out to me was the collaboration with the GPT platforms. Would we need to purchase a separate subscription to a GPT model to fully utilise the function? I note that there are limited pdf uploads permitted on certain GPT models - subscription is needed to unlock features for more.
I was anticipating Paperpile to build a feature like Adobe Acrobat’s “AI assistant”. This would be a great feature to have!
Hi Hansley,
Thanks for voicing your thoughts here. Keep them coming!
For Paperpile, there is an ever-present tension between (1) building product functionality within Paperpile that wraps around some key technology, versus (2) integrating Paperpile with a vendor or product that uses that technology.
Which approach is best? It depends on the details.
For example, back in ~2012 when we first started Paperpile, one such technology was real-time collaborative writing on the web. We knew this was a transformative & enabling tech, but we also knew we couldn’t build a great product from the ground up ourselves. So we took approach #2, integrating with Google Docs. This has been great – Google offers a full-featured Docs experience for free to anyone who signs up with a Google Account.
As we think about the use case of asking LLMs clarifying questions while reading papers, I see the same tension arise: should Paperpile favor integrating with external products & providers of this technology (ChatGPT, Gemini, etc.) or should we build our own product experience around the lower-level underlying tech?
One revealing signal from our AI survey was that many people have an existing preferred AI chat product, either paying for it themselves or with access through their institution. Many respondents expressed fear that Paperpile might add “yet another” separate AI chatbot, or subscription which they’d have to learn how to use and pay for. And that a Paperpile-only chat assistant wouldn’t likely be as full-featured or as efficient as the major products out there.
I do see your point on the other side: the free tier for many providers is limited (because these models have a high cost to run).
For now, we are investing our early development toward integrating with other AI systems. We think this offers a nice improvement to a workflow that many researchers are already using. Longer-term, we will keep an eye on the landscape and the details of compelling use cases to help make our users’ research lives better; if it makes more sense to build our own product atop a given technology, we may make that change.
To end with a question for you (or others who come across this thread): we’re still learning about the motivations and concrete examples of how researchers might use a feature like Adobe’s “AI assistant”. If you’d be willing to share why, or how, you use something like Adobe’s AI assistant while reading academic papers, I’d love to hear more.
Thanks,
Greg
Hi Greg,
Thanks for the response. That is a great tension to have to be fair.
Note that I started using Google Docs exclusively because of the Paperpile function and got through my MSc program with this combination.
I do think there is value in paperpile integrating with an AI system. You ask about how I use the Adobe “AI assistant”. The image probably summarises how I use it. Assuming I already have background knowledge on the subject matter, I mainly use it as a matter of convenience to summarise and compare texts. Querying multiple documents at once is extremely useful to extract basic textual understanding. Ofcourse, this is largely a measure of whether I have “asked the right question” to the model to guide it’s query.
The Adobe function as a pdf document manipulator is what i primarily pay for. The AI function feels like an added bonus that I could also have access to through an LLM like chatGPT. Subscribing to Adobe just means that I can have access to all the above on one platform. I use ChatGPT because of the cumulative insights it develops specific to me.
Thanks
HG
Personally, I would very much like an AI-summarize feature in the PDF annotator. For example, when I export a compilation of my notes on a certain PDF, I’d like to include a short AI-summary in that compilation. I frequently import these note compilations into Obsidian for further study, so it would be great if they could be accompanied by a brief summary of the paper in question.
