Ever since I started using #ChatGPT, one of my goals was to engage in meaningful conversations with the AI while leveraging my own personal information that isn’t readily available on the internet. Unfortunately, this wasn’t feasible by default. However, with the advent of ChatGPT version 4 and the introduction of extensions, it is now possible to achieve that functionality. It’s important to note that while these advanced features might not be accessible in the free version, they hold tremendous potential.
In-depth article will come later...
This realization prompted me to explore solutions like LLM (Large Language Model) that can operate with data from the content stored on my computer or at least within online databases, without compromising my data’s privacy by sending it to #OpenAI, the entity behind ChatGPT.
PrivateGPT was one of the early options I encountered and put to the test in my article “Testing the Latest ‘Private GPT’ Chat Program.” Although it seemed to be the solution I was seeking, it fell short in terms of speed. Simple queries took a staggering 15 minutes, even for relatively short documents. The bottleneck primarily stemmed from my computer’s CPU, and I couldn’t help but anticipate a faster alternative given the rapid advancements in GPT and artificial intelligence.
Then came LocalGPT, a recent release that caught my attention. Unfortunately, I couldn’t immediately dive into it due to other commitments. Finally, during the weekend, I dedicated an hour to set it up and address any coding issues.
#LocalGPT builds upon the foundation of PrivateGPT while incorporating GPU support, which significantly enhances its performance.
To put it to the test, I experimented with the Constitution of Latvia as my initial document. Remarkably, LocalGPT exhibited the ability to paraphrase and extrapolate information even from words that weren’t explicitly mentioned, yet conveyed a specific meaning. It also demonstrated proficiency in generating concise summaries.
Considering the reasonable response time of approximately 3 minutes (using an 8GB GPU), LocalGPT proved to be a viable option. Similar to PrivateGPT, it also provides contextual references to support its generated answers.
Next on the agenda is exploring the possibilities of leveraging GPT models, such as LocalGPT, for testing and applications in the Latvian language. It’s worth mentioning that I have yet to conduct tests with the Latvian language using either PrivateGPT or LocalGPT.
As I continue my journey of learning the Python programming language, which conveniently aligns with NLP (Natural Language Processing) and the exploration of artificial intelligence, I anticipate even more exciting advancements and discoveries in this field.