EzDubs: Transforming Person-to-Person Translation
Y Combinator-backed EzDubs aims to address the issue of person-to-person translation through its innovative app. This app supports translation for calls, voice messages, and text in over 30 languages. It was founded in 2023 by Padmanabhan Krishnamurthy, Amrutavarsh Kinagi, and Kareem Nassar. Krishnamurthy and Kinagi first met during college in Hong Kong and built a project that read lips and translated speech into text for those with hearing loss. In 2021, they moved to Columbia University and began working on video dubbing. There, they met Nassar, who was leading Cisco's Speech AI Group while pursuing a master's degree part-time. Nassar had extensive experience in real-time speech AI products and later created Voicea, which was sold to Cisco. The trio officially started building EzDubs in 2023 with a dubbing model and an early version of a translation tool.The company initially launched a Twitter/X bot in January 2023 that translated clips on the platform. This bot has amassed over 340,000 followers and receives more than 500 dub requests daily, with translated videos getting over a million daily views. In July 2023, they launched a bot on WhatsApp, enabling users to translate voice messages and videos. However, users had to forward voice messages to the bot, get the reply translated, and then send it back. To overcome this, they decided to build their own app and released an early version this year.Krishnamurthy emphasized that partners at Y Combinator advised them that if they could solve the problem of communication latency in different languages, a communication platform would have a more significant downstream effect than video. This led them to opt for building a communication app instead of focusing solely on video dubbing.The EzDubs app, available on iOS and Android, offers real-time translation for calls with support for 30 languages. Users can call someone speaking a different language and receive an instant translation. The other person doesn't need to have the EzDubs app. The app also provides translation for text, voice, and video messages.Although users can share translated voice or video messages outside the app using a link, the founders believe that for those communicating in multiple languages, platforms like WhatsApp and iMessage are not sufficient. EzDubs has noted that many people use their app to make hundreds of calls daily, with an average call time of 17 minutes. The top use cases include people dating across cultures and professionals communicating with locals while abroad.At the core, EzDubs has two main models: one for voice cloning while maintaining the conveyed emotions and another for translation. The translation model is designed to handle interruptions and doesn't wait for a person to finish a whole sentence before starting internal translation.EzDubs has raised $4.2 million in seed funding led by Venture Highway founded by ex-WhatsApp CBO Neeraj Arora. Other participants include Y Combinator partner Jared Friedman, Replit CEO Amjad Masad, Replit president Michele Catasta, Applied Intuition's CEO Qasar Younis, and a16z-backed cloud startup Replicate's CEO Ben Fishman.Friedman pointed out that the company is on a path to making translation tools easily accessible to users. He believes in the product due to the founders' rich history in speech and language learning. He added that even within a company, there are language barriers that can hinder efficient communication, and EzDubs can remove these barriers.The company is soon launching a feature that allows users to scan a QR code and initiate an EzDubs call instantly without downloading the app. The founders stated that while apps like Google Translate have a real-time mode, it requires passing the phone back and forth. Eventually, they aim to make EzDubs a default phone calling app to handle incoming calls as well.In the coming months, EzDubs plans to build an extension for apps like Google Meet, Microsoft Teams, Zoom, and Slack.Unlock the Power of Google's Gemini with Deep Research
How Deep Research Works
When a user poses a question, Deep Research creates a "multi-step research plan" that the user can either revise or approve. Once approved, Deep Research refines its analysis over a few minutes. It searches for potentially interesting information, saves it, and then initiates new searches based on what it has learned. This process repeats multiple times until a report of key findings is generated. Initially, Deep Research is only available in English on desktop and the mobile web, with plans to expand to Gemini mobile apps in early 2025. Users can access it by selecting the "Gemini 1.5 Pro with Deep Research" option in the model's drop-down menu.
For example, imagine a student working on a research project. By using Deep Research, they can get a detailed research plan and quickly gather relevant information from across the web. This saves them a lot of time and effort compared to traditional research methods. It acts as a valuable tool to assist in the research process and help students achieve better results.
Potential Harms and Ethical Considerations
While Deep Research is an impressive feat, it also raises several ethical questions. Just like all AI, it makes mistakes and can hallucinate. This can have serious consequences, especially in education. As Jessica Grose pointed out in a recent op-ed in The New York Times, students are increasingly relying on generative AI to outsource brainstorming and writing. This risks them losing the ability to think critically and overcome frustration with difficult tasks.
There is also a potential financial impact on publishers. By scraping information from websites and compiling it into briefs, Deep Research could deprive these sites of valuable ad revenue. One study has shown that since the launch of AI Overviews, publishers have seen a 5% to 10% decrease in traffic from search. An expert estimated that AI-generated overviews could lead to more than $2 billion in losses for publishers.
However, Google claims that Deep Research can "connect users to relevant websites they might not have found otherwise so they can dive deeper to learn more." It remains to be seen whether this promise will be fulfilled and if the feature will truly enhance the user's research experience without causing harm to publishers.
Gemini 2.0 Flash: The New Flagship AI Model
Starting today, both free and paying Gemini users will have access to Gemini 2.0 Flash, Google's newest flagship AI model. This is an experimental version optimized for chat, with the full version set to arrive in January. Google claims that 2.0 Flash should deliver better performance across various tasks and faster responses. Users can select it from the Gemini model drop-down on desktop and the mobile web (but not the mobile apps yet).
Although the company cautions that some Gemini features may not be compatible with the experimental model, it still offers exciting possibilities. For instance, in a chat-based scenario, users can expect quicker responses and more accurate answers. This could lead to a more seamless and efficient user experience.