I’ve never dreamed of writing blog posts two days in a row, but something massive dropped that people reading my previous post about the murky situation surrounding open-source language models should be aware of. A day earlier, Perplexity AI released an open-source large language model they’re calling pplx-70b. It is completely uncensored in its outputs and does not engage moral judgments or limitations. Anyone can go to
Creating LLMs will become easier and cheaper. You will be able to run one on your smart phone. This idea of regulating them or containing them will become moot soon enough. The Genie is out of the bottle.
I would much rather have an unfiltered large language model than a curated one. The big tech curators have shown they are less trustworthy with censorship than the rest of us are with open discourse.
I’m far more concerned about the powers that be using these tools to manipulate the public than I am of the public searching for forbidden information. Glad someone else feels the same way, and has put their money where their mouth is.
It's actually been eye opening using Perplexity Labs after having been a long time daily user of ChatGPT (free, paid subscriber for a while also). I don't know how to describe the difference, only to say that it feels "more approachable, more open, more free, less ableist, less plagued".... than ChatGPT. Sure, in the obvious ways as seen in the article above, but also in much more nuanced ways that I couldn't give an easy example of. In short, the interaction with uncensored feels much less "processed".... but with no clear and obvious evidence as to why it feels that way.
Ahhh ... got it. I used the default. I'm amazed this is still available though I suppose it's moot with the pace with which other different LLM's are being released.
Creating LLMs will become easier and cheaper. You will be able to run one on your smart phone. This idea of regulating them or containing them will become moot soon enough. The Genie is out of the bottle.
I would much rather have an unfiltered large language model than a curated one. The big tech curators have shown they are less trustworthy with censorship than the rest of us are with open discourse.
I’m far more concerned about the powers that be using these tools to manipulate the public than I am of the public searching for forbidden information. Glad someone else feels the same way, and has put their money where their mouth is.
It's no longer available in the model options, sad.
It's actually been eye opening using Perplexity Labs after having been a long time daily user of ChatGPT (free, paid subscriber for a while also). I don't know how to describe the difference, only to say that it feels "more approachable, more open, more free, less ableist, less plagued".... than ChatGPT. Sure, in the obvious ways as seen in the article above, but also in much more nuanced ways that I couldn't give an easy example of. In short, the interaction with uncensored feels much less "processed".... but with no clear and obvious evidence as to why it feels that way.
Not surprisingly, it seems like they have disabled the uncensored model described in the post.
I just accessed it. Use pplx 70b chat, not the online version from the drop down list.
Ahhh ... got it. I used the default. I'm amazed this is still available though I suppose it's moot with the pace with which other different LLM's are being released.
Oh the open source movement is accelerating quite rapidly. Mistral just trained a new model with no guardrails https://x.com/emollick/status/1734224029479731313?s=46&t=jwRH-lGeT8bRwO6m8ReEmw