Exploring ideas in private with Venice.ai part 1
How do you value privacy when it comes to using LLMs?
"Privacy is not an option, and it shouldn't be the price we accept for just getting on the Internet." - Gary Kovacs
As powerful as ChatGPT and Claude.ai are, there is always one thing I wish were different: that they not use or keep anything I submit to their LLMs. The issue is the same when using Google’s Gemini LLM.
Said more directly:
“If you use ChatGPT or Claude all your documents are stored forever on their servers, and may be leaked, hacked, subpoenaed or spied on by staff or outside parties. Moreover, Claude does not have search capabilities, so the information will be outdated and limited to its training data.” How to Use an AI Research Assistant: Mastering Web Search & Document Analysis in Venice
For personal or sensitive topics, as well as research on subjects I'm interested in, I've started using Venice.ai. As you have heard me say, AI doesn't simply answer my questions; it guides me to new insights. This approach is a more empowering way to use language-based AI tools. When dealing with private matters, I prefer a secure environment to process what's revealed.
The goal of Venice.ai is to provide a permissionless and privacy-focused alternative to popular AI applications. Venice.ai aims to deliver uncensored, unbiased machine intelligence while preserving user privacy.
Here are the details:
Permissionless Access: this means users can access the platform without needing to create an account or provide personal information. Venice doesn't store (and can't access) your chat history. Instead, it is stored locally in your browser, and you can delete it here.
Uncensored AI: Unlike many other AI platforms, Venice.ai does not censor the AI's responses. This allows users to receive unfiltered information and interact with the AI without restrictions.
Open-Source Models: Venice.ai utilizes leading open-source AI models, such as Meta's Llama 3 and Stable Diffusion, to provide text, code, and image generation capabilities.
Real-Time Search: Venice.ai recently added real-time search capabilities using Brave's API, allowing the AI to provide up-to-date information and cite sources for transparency.
There is a generous free plan with no requirement for you to login to use it. This is what initially caught my attention. The free plan includes creating images with significantly fewer restrictions if that's your thing. It allows a maximum of fifteen (15) text prompts per day and a maximum of five (5) image prompts per day using this feature.
I paid for a year of use at the most reasonable price of $49 because I’m always testing new tools and I like the ability to test out the Open Source LLM models. More importantly, I like the privacy aspects of the product. Let me stress you do NOT need to pay if you’re fine with the free version. I used the free plan for a couple of months and found it extremely useful and I never logged.
Start with their free plan. Then go ahead and sign up for an account (still free). Then consider trying the LLM for a year. Explore those private thoughts or worries you feel need answers to settle the internal dialogue and concerns we all carry. Or maybe it’s a personal fantasy that has your curiosity fired up.
Update:
Just as I was about to publish this, a product update floated in from Venice plump with news and insights.
Upgrade: Larger PDF and TXT uploads
New feature: Download Chats
c now available on Venice
User Highlight | Tom Bombadil, Author and Illustrator
Hermes 3 8B by Nous Research now available on Venice
If you follow the latest news in image generation techniques, you’ll recognize FLUX image generation techniques have been hot in the news lately. Their image generation engine is becoming the El mejor (Spanish for the best) tool around so I was delighted to see it integrated into Venice.ai.
Even if you’re totally committed to using other LLMs, I recommend checking out Venice.ai to understand the power of Open Source AI solutions. I expect you’ll be impressed. I am and I find a sense of relief knowing what I input into their chatbox is not being kept or used for other purposes.
Appreciatively,
Tom
Thanks for sharing these insights about venice.ai, Tom - I’m going to look into it further. Permissionless and private are excellent properties, and I confirmed in their TOS that users ‘own’ what they generate.
I’d love to hear your thoughts on a few questions:
- The models they use, eg Llama, are open weights, not true open source. Is Venice.ai itself open source? The FAQ doesn’t address this. https://venice.ai/faqs
- Does the absence of guardrails mean that Venice could potentially be used unethically to generate images based on copyrighted base content, deepfakes, undressing, CSAM, etc? (The TOS say this isn’t permissible, but there doesn’t seem to be any control or enforcement mechanism.) Asking because Venice includes PixArt Sigma which was trained on Stable Diffusion's 1.5 model; that model was recently pulled from Hugging Face due to its inclusion of CSAM content. (https://spectrum.ieee.org/stable-diffusion).
- Do you have a venice.ai referral code I could use so that you get points when I sign up? :)
I have been using venice ai. It is really nice to see an AI that do not keep answering as an LLM....I can't do this and that. It keeps the creative juice flowing. And I can make image with Flux. Which is amazing