Gemini "duck" demo was not done in realtime or with voice
Google’s high-profile Gemini “duck” demo is being criticized for heavily staged and edited interactions that were presented as real-time, multimodal AI capabilities. Commenters argue that the video crosses from normal product polish into deceptive marketing, since the model actually responded to hand‑picked still images and detailed text prompts rather than live video and voice. The episode fuels broader concerns about trust in AI vendors, inflated benchmarks, and how aggressive hype from big tech distorts both investor expectations and developer adoption.
Nature of the Gemini “duck” demo
- Video strongly implied real‑time, open‑ended interaction with continuous video and voice.
- Later Google blog/press coverage show it was built from hand‑picked still images plus carefully written text prompts, with outputs edited and reordered.
- Audio, timing, and some capabilities (e.g., “generating” music vs just searching for it; the cup‑shuffling game from stills, not video) were dramatized beyond what the product currently does.
Disclaimers, marketing, and deception
- YouTube description and tiny on‑screen text noted “shortened sequences” and reduced latency, but did not clearly state that prompts, inputs, and interaction mode were substantially different.
- Many commenters argue this crosses the line from normal polished demo into misleading or “blatant false advertising.”
- Others see it as typical marketing “smoke and mirrors” or aspirational prototyping, not unusual in tech demos.
Trust, ethics, and potential legal issues
- Several say they were impressed at first, then feel duped; some say this permanently erodes trust in Google’s announcements and AI claims.
- Comparisons made to prior Google demos (e.g., Assistant/Duplex calls) and to industries that over‑promise (AAA games, EV startups).
- A minority raise the possibility of securities or consumer fraud, noting the stock bump, while others think it’s just aggressive puffery.
Comparison with OpenAI and product strategy
- Many contrast this with OpenAI’s GPT‑4/GPT‑4V demos and Dev Day, perceived as closer to live use and backed by immediately available products.
- Some say GPT‑4 has largely replaced Google Search and Stack Overflow for coding and many “simple question” queries; others still see classic search as essential for navigation, official docs, and business info.
- Several interpret Gemini’s launch as evidence Google is behind in deployed AI despite its research history (e.g., inventing transformers).
Hallucinations and correctness
- The demo itself contains at least one factual error (reasoning about why the rubber duck floats), used to illustrate that Gemini still “hallucinates” confidently.
- Long side discussion on how to judge LLM answers, the difficulty of verifying correctness, and whether imperfect tools are acceptable if used with care.
- Various techniques (testing code, RAG, chain‑of‑thought, external fact‑checking) are mentioned, but no consensus that hallucinations are close to being “solved.”
Broader implications
- Commenters worry that such staged AI demos normalize fakery in an era of deepfakes and generated content, further eroding trust in video, audio, and corporate claims.
- Some argue the real issue is corporate incentives: pressure to impress investors and “keep up” in AI leads to over‑hype and undermines genuine technical progress.