The founder of Hinge raised $18M to build a new AI dating service, Overtone
THE SO WHAT
An $18M raise for Overtone’s AI- and audio-first dating app is another bet that AI curation can replace swiping as the core UX. Any marketplace or matching product should be re-evaluating whether AI-led introductions can increase liquidity and ARPU enough to justify the infra and trust costs.
READ THE SOURCE
MORE FROM THE WIRE
Startups & VentureA look at Founders Fund-backed State Affairs, which raised $70M and uses AI trained on reporting by its 76 staffers to power a Bloomberg Terminal-like product
Training an in-house model on 76 reporters’ work to build a Bloomberg-style terminal is a bet that proprietary, structured reporting can be monetized as a data product, not just ad inventory. If you own a niche data exhaust — legal, industrial, local, scientific — this is your prompt to price it as an AI-native terminal, not another newsletter.
Startups & VentureDeepSeek reportedly in talks to raise $1.5B, then IPO
A $1.5B raise at a reported $71B valuation for a Chinese LLM developer signals that capital markets are willing to fund multiple regional AI champions at hyperscaler scale. If you’re building on these models, vendor risk now includes geopolitics and listing venue, not just model quality.
Startups & VentureVieu launches AI-ready map of business relationships, challenging tech incumbents
A live "Business Graph" built from observed signals — shared work history, boards, co-investors — is a direct shot at self-reported networks like LinkedIn. GTM and BD teams should expect prospecting to shift from who-you-know to who-you’re-actually-connected-to in the data.
Startups & VentureLaunch HN: Agnost AI (YC S26) – Extract user feedback from agent conversations
Reading production chat and voice logs for rageprompting, repeated requests, and drop‑offs turns agent UX into a measurable funnel, not a black box. If you’re deploying customer‑facing agents, you need this style of behavioral analytics in place early—otherwise you’re flying blind on where the model is actually failing users.