ChatGPT is great but a lot of folks are concerned about data privacy.
Here's an alternative: Install a local LLM (Large Language Model) on your computer. You can it use offline and keep your data private.
Best part — you can do it for free!
Here's how to do it:
1. Go to lmstudio(dot)ai and download the version of the studio compatible with your device.
Open the setup file to install LM Studio on your computer.
2. Open LM Studio after you have installed it.
In the search bar, type in the name of an LLM you would like to use and click on "Go."
For the purposes of this tutorial, I am going with LLaMa, an open source model by Meta AI.
On the left side of your screen, you will see different version of LLaMa. On the right side, you will "Quantization" of different models.
On the top of the Quantization column, you will see a check mark with "Should Work." It shows that your computer's hardware fulfills the requirements of this model.
If your hardware does not fulfill the requirements of a given model, you won't see "Should Work."
I looked up Falcon and it told me I would need 150+ GBs of RAM to run this model.
Pi is a new AI chatbot launched recently by Inflection.
You can use it on your computer or a mobile device (iOS and Android).
It's totally free.
Here's an overview of Pi and a comparison with ChatGPT 3.5
Go to inflection(dot)ai and click on "Meet Pi" and create an account.
Pi has a bunch of useful "personas" that you can access by clicking on "Discover."
You can use these personas for anything from venting to brainstorming to career planning.
I like this feature. Plus the user interface feels polished.
To test Pi as a brainstorming partner, I ask it to help me brainstorm research questions about contemporary anglophone Pakistani fiction with regard to world literature and postcolonialism.
Pi's suggestions are useful but seem a bit superficial.
I take one of Pi's suggestions to dig deeper and it hallucinates (i.e. making things up on its own).
How to determine if a paper is relevant to your project? Should you scan and skim or read the whole thing carefully?
You can make a custom GPT to help you find out how relevant a paper is to your project.
This will take you less than 10 minutes.
Here's how to do it:
Open your ChatGPT and click on your profile photo in the bottom-left corner.
Click on "My GPTs" and then "Create a GPT."
ChatGPT will give you two options: Create and Configure.
Ignore "Create" and click on "Configure."
(Please note that you will need a subscription to GPT-4 to create a custom GPT.)
In the "Configure" tab, you will see options to choose a name and write a description for your GPT.
I named my GPT "Paper Relevance Advisor" and in the description I wrote "To read or not to read."
Under Description, you will see Instructions.
This is the most important part of creating your GPT. Your instructions will dictate the way your GPT behaves.
You can use the following template for instructions. Feel free to customize it according to your specific requirements.
Paper Relevance Advisor is a highly specialized tool that helps scholars working in the fields of [your primary research field] and [secondary field].
The current focus of this GPT is on a project [describe your current research project in a few lines].
This GPT will avoid citing any fake references at all times. Reliability and accuracy of information are paramount.
When a research paper is given to this GPT, it will go through its whole text and evaluate its relevance to the current project on [your project title].
This GPT will categorize research papers as:
1. Must Read: Papers that are very relevant to the current project [your project title]. Not including these papers may have an adverse effect on academic rigor of the project.
2. Scan and Skim: Papers that may be relevant to the current project on [your project title]. These are the papers that if cited in the project will make it impressive but not citing them will not damage the core argument.
3. Read If Time Allows: Papers that are not directly relevant to the current project but may contain some theoretical or methodological insights that can be extrapolated.
4. Don’t Read: Papers that are totally irrelevant to the current project and contribute nothing to it.
The GPT will also give justification for categorizing a give research paper.
The GPT will opinionated analysis in a respect tone. If a paper is not from the field of literary studies, it will not shy away from saying that the paper is totally irrelevant to the project.
Click on "Save" in the top-right corner and then "Confirm."
It has two models Pro and Ultra. Gemini Pro is already integrated into Bard.
Gemini Pro is supposed to be better than ChatGPT 3.5 and Ultra (not released yet) better than GPT 4.
Here's a comparison of ChatGPT 3.5 and Gemini Pro for academic purposes (with my recommendations):
I start by asking both ChatGPT 3.5 and Bard a simple question about a literary scholar.
"Do you know who Franco Moretti is?"
ChatGPT 3.5's response is useful but a bit shorter.
Bard, on the other hand, give a detailed answer with Moretti's background, list of works, his influence, and additional information. It's sort of like a short Wikipedia entry.
If Google can make Bard/Gemini Pro sustainable and consistent, they are unlikely to lose their userbase despite what Google's critics say.
For generic questions, Bard seems like a better choice.
Next I raise the stakes a little bit.
I take a passage of around 500 words from one of Moretti's essays and ask ChatGPT 3.5 and Bard the following question:
Below is an extract from his famous essay, "More Conjectures on World Literature," which is the second part of "Conjectures on World Literature."
I teach an undergraduate class at the University of Southern Denmark. What are the points in the extract below that an undergrad may find difficult?
ChatGPT 3.5 goes straight to the point and tells me the exact terms that undergraduate may find difficult.
In comparison, Bard starts with a generic suggestion that Moretti uses complex theoretical concepts.
That said, it does point out the difficult terms in the second bullet point.
Bard also give detailed (although generic) suggestions on how to make Moretti's text accessible for undergraduate students.
Bard will certainly have an edge over ChatGPT 3.5 because:
1. Unlike ChatGPT 3.5, you don't need to sign up for Bard if you have a Google account.
This may seem like a minor detail, but every time I give a talk I ask folks if there is anyone who doesn't have a ChatGPT account. And there are always a few people.
2. You can integrate Bard with your Google Drive, have it access documents, and summarize them for you. ChatGPT 3.5 doesn't let you upload any documents.