Venice.ai

What is your opinion about venice.ai?

They claim to not store any of your queries and that your chat history is only stored on your device.

They also outlined their architecture here:

Venice doesnā€™t spy on you
Venice doesnā€™t censor the AI

In one of the requests, he refused to answer, citing the fact that I was using offensive language. He also stated, ā€˜I am not in a position to discuss the physiological characteristics of different groups of people.ā€™

Thus, they are lying about at least one thing; censorship is present after all. Therefore, I canā€™t trust them not to spy as well.

1 Like

Imo it wouldnā€™t make sense for them to use ChatGPT or Claude since they log and use your queries afaik. Also slightly worse LLMs might be a good tradeoff for a lot of people if this does really improve your privacy by as much as they claim.

lol, they can say whatever they want to AI. This is the purpose of being ā€œuncensoredā€.

2 Likes

It was featured by Naomi Brockwell.

The biggest issue is that the free models are only 7B models. Those models can already be run easily on your PC. (Check privacyguides. org/en/ai-chat)

The free tier is not very good but there appears to be a way to get premium without paying a monthly fee by purchasing a crypto token which isnā€™t too pricey.

This seems a bit sketchy honesty. venice.ai might be a nice concept, but if you want private AI just use duck.ai (by DuckDuckGo) or the AI now available at search.brave.com

2 Likes

This is not how this works. Venice does not train (create) models. They only provide private access to models with open weights (like LlaMa), running inference on their (rented) hardware and delivering inputs & outputs via a proxy.

Thus they have no influence on the biases and restrictions placed into models during their training.
They just use models that are less restricted like Dolphin and donā€™t censor prompts at the input level.
So they are not lying, you just donā€™t seem to understand how LLMs work.

That doesnā€™t seem true anymore. They have Llama 3.3
I would love feedback on this service as Iā€™m currently looking for a private AI that receives files/pics

1 Like

If you really want privacy you are better to self-host it.

I donā€™t have the hardware for Llama 3.3, or any good LLM

Understood. Tbf Qwq-32 is as good as Llama 3.3 70B.

If you use any online AI service always redact personal info and remove images metadata.

OP here:

Iā€™ve tried their premium through holding some crypto tier which they dropped soon afterwards.

Iā€™ve since switched to using Open WebUI connected to both local LLMs and openrouter.ai. For confidential stuff I send it to locally running models. For things that arenā€™t confidential I use openrouter.ai with the setting enabled that disallows providers to log queries. openrouter.ai makes it easy to switch between models fast and effortlessly.

Iā€™d probably recommend this setup over using venice.ai to almost everybody.

There is some telemetry, which is enabled by default, but you can disable it in the settings.