Hacker Newsroom AI for 20 May: Karpathy Joins Anthropic, Gemini 3.5 Flash, Gemini Omni, OpenAI Image Provenance

Hacker Newsroom AI for 20 May recaps 5 major AI Hacker News stories, moving through karpathy joins anthropic, Gemini 3.5 Flash, gemini omni, openai image provenance. 1. Karpathy Joins Anthropic The next story is Andrej Karpathy announcing that he has joined Anthropic, saying the next few years at the frontier of large language models will be especially formative and that he wants to get back to research and development, which matters because it signals where one of the field's most visible builders thinks the center of gravity is moving. Hacker News mostly treated it as a major talent move for Anthropic, while arguing over whether this says anything real about the company, the market, or the race toward more capable systems. Story link Hacker News discussion 2. Gemini 3.5 Flash The next story is Google launching Gemini 3.5 Flash as the first model in its new Gemini 3.5 family, claiming frontier-level coding and agent performance at much higher speed and positioning it as the engine for long-running agent workflows, which matters because Google is trying to make fast, practical agents feel production-ready instead of experimental. Hacker News reacted with interest in the benchmark claims and speed, but a lot of the discussion quickly turned into a fight over the pricing and whether something called Flash should really cost this much. Story link Hacker News discussion 3. Gemini Omni The next story is Gemini Omni, a new Google DeepMind creation model pitched as turning any mix of text, image, audio, or video into a coherent edited video through step-by-step conversation, which matters because it pushes multimodal AI from generating single assets toward persistent creative editing across many turns. Hacker News reacted less to the product page itself than to what tools like this could do to filmmaking, visual effects, and the already shaky economics of media production. Story link Hacker News discussion 4. OpenAI Image Provenance The next story is OpenAI adopting Google's SynthID watermarking for its AI images, alongside C2PA content credentials and a public verification preview that checks whether an uploaded image contains OpenAI provenance signals, which matters because the major labs are trying to make image origin easier to trace before synthetic media becomes even harder to trust. Hacker News mostly treated it as the opening move in an arms race, with skepticism that watermarking can survive once removal tools spread and once platforms or propagandists have incentives to strip the signals out. Story link Hacker News discussion 5. Remove AI Watermarks CLI Library The next story is a GitHub project called Remove-AI-Watermarks, a command-line tool and library that removes visible Gemini marks, strips C2PA and EXIF provenance data, and uses diffusion to weaken invisible watermarks like SynthID, which matters because it immediately tests how durable the new provenance push really is. Hacker News reacted with a split between people who see this as a useful proof that watermarking is fragile and people who think it mainly helps bad actors disguise synthetic media. Story link Hacker News discussion That's it for today, I hope this is going to help you build some cool things.