pull down to refresh

184 sats \ 3 replies \ @k00b 28 Jan

Congrats! wen deepseek?

reply

We're talking with our GPU provider now about this. I want this badly. For now it's Llama 3.3 70B.

reply
84 sats \ 1 reply \ @k00b 28 Jan

I must say it seems pretty bullish for the open source AI side of things. Supremacy by obscurity isn't working in AI at all.

reply

I'm also excited by the reproduction efforts underway to reproduce R1 with open data too. The whole stack being truly open source is gonna be wild.

reply

Just signed up. I'm gonna test it out.

reply

LFG, let me know what you think!

reply
100 sats \ 1 reply \ @NovaRift 28 Jan

I have tried to look for many private open source models, but I found that some don't have a free plan to test out. If they do, they provide the weakest model. Even in the pro plans, reviews say they hit chat limits faster. The closest thing I could find is https://duck.ai/ by DuckDuckGo.

They say:

Active Privacy Protection Your chats are private, anonymized by us, and not used to train AI models.

But I can't sync chats. Maple AI looks promising. Thanks

reply

Unfortunately I think other claims of "private" is pretty fast and loose too. What almost everyone else is doing is akin to being a "vpn that promises not to log", but you don't know that.

With us, we're hosting things in the secure enclave of a server, with proof that it's running the expected code. This expected code shows there's no privacy leakage occurring or sensitive logs.

reply
136 sats \ 0 replies \ @marks 28 Jan

It's the first app running on OpenSecret. We have more in the works from third party developers. Goal is to create a future where the apps we use everyday have privacy on by default.

reply

This looks great! Why does it cost more to pay in bitcoin though?

reply

The bitcoin plan is priced yearly, and the monthly plan amount is showing a discount of 50% for just 3 months.

reply

Lightning invoices are really simple to pay though. Why would that method of payment make it more desirable to commit to a year as opposed to a month/day or even a metered number of response tokens?

Lightning payments should enable less friction than credit card payments, not add more, right? (Sorry, not trying to be an ass, just trying to understand if there's a pain point I'm missing here - maybe we can do something about it.)

reply

I agree, LN is perfect for a "pay for what you use" Model... Paying for the whole year in bitcoin seems... Yeah, not optimal.

Have you tried https://venice.io?

I like https://brave.search for private inquiries. But still using https://poe.com for work stuff, because the prompt bots I make give me a lot of utility

Just added a new AI for my "to try" bucket

reply

Completely understand, though there's a completely free plan for 10 chats a week. For just one-offs and casual usage, just use it for free. No need to worry about micropayments. If it's of value to you and you use it a lot, I think that's when it makes sense for a subscription.

Completely understand not wanting to pay for a whole year though but when Zaprite gets better tooling around email notifications and subscriptions, we'll consider it again. Just needed to launch our product without worrying about all the variety of billing complexity and endless user preferences.

reply
Lightning payments should enable less friction than credit card payments, not add more, right?

I disagree. I don't think anyone wants to think about payments, especially if they have to do it every single month. We'll consider it in the future once zaprite integrates at least email-based notifications/subscriptions.

reply

Great Free Plan! Thanks

We are committed to making Maple AI accessible to everyone, so we offer a range of pricing plans to suit different needs and budgets. Our Free plan allows for 10 chats per week, while our Starter plan ($5.99/month) and Pro plan ($20/month) offer more comprehensive solutions for individuals and organizations with heavier workloads. We accept credit cards and Bitcoin (10% discount), allowing you to choose your preferred payment method.
reply

How does this site differ than PPQ or NVK unleashed chat?

reply

They are typically either proxies to other cloud AI providers or hosted on their own infra, unencrypted. It's pretty much a "we're a vpn that doesn't keep logs" kind of relationship.

Ours uses verifiable techniques to ensure that not only we can't inspect or tamper with it, but it also secures the AI chat all the way to the GPU model. So in no way is the chat viewed unencrypted anywhere except when you decrypt and view it yourself.

reply

I'm not ever paying a subscription for cloud services. Pay for what you use or gtfo. Especially in a world where you can HTTP 402 and LN zap per request.

reply
21 sats \ 1 reply \ @Diego 28 Jan

Is this an app as well or just web?

reply

Just web for now. If there's a lot of demand, we'll consider mobile apps.

reply

How is it better than running the model offline on personal hardware?

reply

ppq.ai lets you pay as you go using LN for cents at a time. They support multiple models including Deepseek too. No need for sign up, email or any of that.

reply

Just signed up to try this out.
How is it different from venice.ai? They provide access to the latest model while focusing on privacy. Leaving aside their shitcoinery, which was recently introduced, the latest LLMs work seamlessly there. If Maple is Bitcoin only, then that would be a plus.

reply

I had a similar question here too, same answer would apply to Venice: #867897

reply