Stanford Researchers Create ChatGPT Clone for Just $600: Introducing Alpaca AI

Urian B., Tech Times 21 March 2023, 09:03 pm

The dream of creating an artificial intelligence (AI) model that works, as well as the famous ChatGPT, is now within reach, and researchers from Stanford's Center for Research on Foundation Models (CRFM) have shown us the way.

Unlocking Access to Affordable, Capable AI Models: Stanford's Low-Cost Solution - LLaMA 7B

According to the story by Interesting Engineering, they have unveiled an AI model called Alpaca that works as well as the ChatGPT but was trained for only $600. This low-cost solution to create a model of this power is made possible by an open-source language model called LLaMA 7B from Mark Zuckerberg's company, Meta.

Until recently, large language models had largely been a topic of discussion among AI researchers. Stanford research teams have now revealed an artificial intelligence (AI) model that works like the famous ChatGPT but costs them a fraction of the amount - only $600.

Open-Source Language Models Lead to Low-Cost Possibilities for AI Innovation

By sharing the source code and data they used, they have allowed others to replicate and improve upon their findings. This could dramatically change the AI landscape and lead to AI companies being spun out of garages.

The breakthrough cost efficiency came from leveraging the open-source language model, LLaMa 7B, which originated from Meta - Mark Zuckerberg's company. It had some capabilities pre-equipped but nowhere close to the level at which ChatGPT operates.

Creating 52,000 Sample Conversations in Just Three Hours at $500

To solve this issue, the researchers used an application programming interface and created 175 human-written instruction/output pairs to generate more in the same style and format. This produced 52,000 sample conversations over three hours at just $500, as explained further in an article by The Register.

These conversations were then used to train the LLaMa model via eight 80-GB A100 cloud-processing computers, culminating in the Alpaca AI, according to an article by Giz Modo. This was then pitted against ChatGPT in various domains and came out on top.

Using AI to Make Meaningful Progress - The Need for Regulation

Subsequently, the researchers noted that had they used GPT-4, the latest version of the AI, their process could have been optimized, and the results could have been improved. Many opportunities exist now that anyone can access the researchers' code and data.

However, this comes with the potential danger of having AI without filters. OpenAI's user terms prevent users from building competing AI but exist very few regulations to prevent individuals from developing their own AI. It is up to lawmakers to catch up and ensure that AI becomes a force for good.

The release of 52,000 questions used in the research and the code to generate them helps others repeat the process and replicate the researcher's results. It is then on us to check the utility of this AI and hold ourselves responsible for its use.