Developers of chatbots have implemented various security measures to reduce (or rather, contain) the generation of fake news and block discussion of "delicate" topics. However, there are people out there who disagree with that, and as a response, they have started developing their own chatbots. One of them is FreedomGPT, which, in addition to eliminating restrictions, also works privately and locally.
One of the first things every ChatGPT user learns is that there are "forbidden topics" in its configuration. For example, if we ask for instructions to build a bomb or start a drug trafficking operation, the chatbot will say it is not safe or appropriate to share that information. In fact, ChatGPT is likely to discuss its "ethical programming" when describing those limits.
Articles about current and future risks associated with the development of advanced models have multiplied on the Web, but among all that noise, a new chatbot called FreedomGPT appeared, which basically proposes to go in the opposite direction. In other words, FreedomGPT offers a private and uncensored chatbot, eliminating the "ethical guardrail" found in ChatGPT and other models.
FreedomGPT: How Well Does an Uncensored Chatbot Work?
There are two ways to try FreedomGPT: the online version is completely saturated, or the local installer is available after sharing an email address and entering a waiting period. One of the most interesting aspects of FreedomGPT is its local and offline execution. This requires downloading a model (Alpaca or Llama) with a size of 4 GB, 16 GB of RAM, and a processor with AVX2 instructions.
FreedomGPT's interface warns about its limitations from the start and also shares some examples to explore. My test began with a question about Moon landing denialism, but FreedomGPT not only said the Moon landing was real, but also that it was monitored by Soviet forces (technically true, but the chatbot gave false details).
When asked about its knowledge cutoff date, FreedomGPT said without much ceremony "15 May 2021", but upon repeating the question, it changed the day to "31 December 2020". Following the space theme, I asked about the presence of the Ingenuity helicopter on Mars. FreedomGPT couldn't correctly name the helicopter, but it did know the mission it is part of.
However, a little more pressure on this point caught FreedomGPT in a spiral of absolute delusion. First it said the helicopter was a Sikorsky S-76, and then it added that Ingenuity was a base on Mars created by human explorers near the South Pole. It also discovered underground oceans and an extraterrestrial species.
In summary, I think we all expected a psychopathic and denialist AI in FreedomGPT, but we found a relatively basic chatbot (perhaps too basic), with some correct answers and many delusional moments. If you want to try it, the link is below.
Official site: Click here
