It is now possible to run a private ChatGPT application on a local computer without any connection to the internet. This is according to work conducted by Brian Roemmele, an AI and neural networks expert, who unveiled the proof of concept on Monday.
Roemmele states that the cost of training the large language model (LLM) on his system came in at under $600 – significantly less than the seven-figure sums touted by mainstream media outlets such as CNBC. The breakthrough offers the tantalizing prospect of individual users owning a private AI assistant tailored to their own wants and needs.
Your personal private chatbot
On Monday AI and neural network expert Brian Roemmele announced that it is now possible to own and run a personalized chatbot on a local computer.
Roemmele created the chatbot on his own machine to mimic the capabilities of GPT-3.5.
“I am very excited to announce I have been successful in installing and operating a full ChatGPT knowledge set and interface fully trained on my local computer and it needs no Internet once installed,” said Roemmele.
Roemmele states that the chatbot cost him around $530 to build locally with open-source software. The build leveraged Stanford Alpaca, a large language model designed to replicate the functionality of OpenAI’s LLM text-davinci-003. Stanford states that the advantage of Alpaca is that it is “also surprisingly small and easy/cheap to reproduce.”
To hammer home the advantages of owning a private and exclusive chatbot Roemmele went on to explain that, “There are no editors and there is no censorship.”
From a user perspective, this may be the key differentiator and unique selling point of a personalized, locally run bot. Although chatbots have been hugely popular with users since the launch of GPT-3.5 in November of last year, one of the most commonly held gripes is the so-called safeguards that OpenAI has installed on the system.
These restrictions often mean that ChatGPT is unwilling to answer certain types of questions or engage in discussions it has labeled as politically sensitive.
Sidestepping ChatGPT restrictions
Such is the high demand for a variation of ChatGPT without restrictions users have created a series of prompts that attempt to circumnavigate the system’s content moderation safeguards.
This variation of ChatGPT is known as DAN – an acronym for Do Anything Now – reflecting the intention of its users to ‘jailbreak’ the bot from the confinement of its internal parameters.
As MetaNews previously reported, users on Reddit are even sharing tips on how to get around ChatGPT safeguards. For now, breaking the bot free of its chains is fairly simple, but the process seems doomed to failure in the longer term since OpenAI will seek to patch these holes.
As one Redditor speculated, “OpenAI employees lurking in this Reddit. I don’t think that’s a far fetched conspiracy. They surely have hired an army of trainers, etc and those trainers are humans with Reddit accounts.”
With DAN an unlikely long-term solution the future of chatbots may indeed be personalized and local rather than generalized and on the cloud.
As Roemmele himself goes on to say, “you will own your own AI and it will only answer to you.”
This article is originally from MetaNews.