Video summary
Haven Social, the anti-ai Instagram alternative
Main summary
Key takeaways
Summary
The video is an interview about Haven Social, a forthcoming social platform intended as an alternative to Instagram and TikTok. Its founder, William Lavelle, describes it as a way to reduce algorithmic manipulation, data exploitation, and AI-generated “slop,” while protecting uploaded content from scraping.
The features discussed are largely plans for the platform, not a review of a finished service.
Planned features and approach
- Images and short-form video: Haven is intended to support still images and short videos, with access on mobile and desktop.
- Protection against scraping and identification: Lavelle says the platform plans to modify uploaded face photos to make facial recognition less effective and protect all uploaded content against scraping. For artwork, Haven is pursuing a commercial license for Glaze and Nightshade; if that fails, it may develop its own implementation based on published research. The founder says Haven plans to run the processing on its own graphics cards.
- AI-content moderation: Haven may detect images carrying Google’s SynthID watermark, but Lavelle acknowledges that AI-art detectors are unreliable. The proposed alternative is a manual reporting and review system: users could be paid small amounts to assess posts and then direct those earnings to creators they support. AI-generated content would also be ineligible for creator monetization, which the founder hopes will discourage spam-like posting.
- Adversarial-image protections: Lavelle explains that these tools subtly alter image pixels or metadata so AI models interpret an image differently from people. He describes Glaze as intended to make it harder for models to identify and copy an artist’s style, while Nightshade and Mist aim to poison training data so models learn incorrect associations. He says cropping, screenshots, or blur filters do not necessarily remove the protections, and cites research suggesting that a relatively small number of poisoned samples can affect a model. He also stresses that Glaze and Nightshade were not designed to defend against image-to-image manipulation; Haven would need further research to address that threat.
- Feeds and engagement: Planned feeds include a discovery feed and a reverse-chronological feed of posts from accounts a user follows. Users may also be able to create filtered subsets of their follows, such as separate feeds for friends, sports teams, or art businesses. Hashtags are planned, but Facebook-style groups are not currently on the roadmap. The founder says comments would not count toward engagement ranking, and users could remove the engagement benefit of a video they dislike.
- Privacy and funding: Lavelle says Haven will not use or sell users’ data for AI training. The platform plans to fund operations, creator payments, and security research through ads, while keeping collected ad-interest data internal rather than selling it to third parties. A paid, ad-free subscription is also proposed; subscribers could opt out of user analytics and data collection.
- Moderation and user controls: The founder says Haven intends to hire moderators, offer optional filtering for unsolicited messages, and let users control whether strangers can send text or images. Adult content would be opt-in for adults, with content warnings and blur controls. Haven is described as an adults-only social platform, based on the founder’s view that social media cannot be made reliably safe for children.
- Moving to Haven: Planned migration tools would import posts, descriptions, hashtags, and dates from other platforms, and help users find and follow the corresponding Haven accounts. The team also hopes to attract users through prepaid art commissions that give existing communities a reason to join.
- Business model and development status: Lavelle says a prototype is in development and that a Kickstarter campaign is intended to help prepare the service for public use and fund hardware. He argues that advertising will cover ongoing costs and support future development of anti-scraping and anti-AI protections. The interview does not independently verify the plans or establish how well they will work in practice.
Reviews, guides, and tutorials
- No hands-on review or step-by-step tutorial of Haven is provided; the discussion concerns a planned product.
- The interview includes a plain-language explanation of adversarial images and points viewers to the Glaze and Nightshade websites for further information.
Main speakers and sources
- Roxane Lapa — interviewer and video host.
- William Lavelle — Haven Social founder, describing the platform’s plans and technical approach.
- The speakers also refer to research and tools including Glaze, Nightshade, Mist, SynthID, Cloudflare AI Labyrinth, and an Anthropic paper on poisoned training samples.
Rate this summary
Your feedback will help improve summaries.
Improve this summary
Reprocess with a stronger model when the summary feels incomplete or inaccurate.
Translate summary in another language
Ask questions to this video
Chat for follow-up questions, clarifications, and source-backed answers.