The Voicebot Podcast

Chandra Khatri is CTO and co-founder of Got-It AI, a company that built an AI that builds conversational AI solutions. It can ingest existing conversation data and automatically generate an intent model and conversation flows that designers can edit in a no-code platform. That same technology was more recently applied to checking the output of GPT-3. Known as CheckGPT or Truth Checker, it verifies the truthfulness of large language model outputs, one of the key concerns of enterprise users of generative AI.  Khatri earned a master's degree in machine learning in 2015 and took that knowledge to eBay, where he implemented a generative AI solution for automatically creating product listings. He then went to work in Amazon's Lab126 where he was a founding team member that launched the Alexa Prize. 

We had another big week in generative AI news. The testers of the new Bing Chat Mode made some disturbing discoveries, but Microsoft also made some changes and revealed a 71% approval rate by early users. Google is pulling out all of the stops to get Bard tested and ready for launch, while Jasper AI and Vertione introduced new generative AI enterprise solutions. And we had Opera and Yext provide new evidence that web browsing and SEO are about to change.  Voicebot.ai's Bret and Kinsella, and Eric Schwartz break down the news, provide updates, and put the developments in context, all while answering questions from the live audience.  Links to stories: Bing Chat Mode: https://synthedia.substack.com/p/bing-chat-goes-wild-with-hallucinations Google Bard Code Red: https://www.cnbc.com/2023/02/15/google-asks-employees-to-rewrite-bards-incorrect-responses-to-queries.html Jasper AI goes enterprise: https://voicebot.ai/2023/02/14/jasper-introduces-generative-ai-api-and-enterprise-tools/ Veritone's new generative AI enterprise applications: https://voicebot.ai/2023/02/16/veritone-releases-generative-ai-features-to-fuel-entertainment-and-advertising/ Opera's GPT-3 features: https://synthedia.substack.com/p/how-llms-will-change-web-browsing You can also watch the show's live recording on Voicebot's YouTube Channel. Follow Bret Kinsella on LinkedIn to get notified of future live recordings. 

Andrei Papancea co-founded NLX in 2018 to solve some of the problems he faced as a software engineer working on natural language understanding at American Express. He worked extensively with designers and analysts that could not make improvements to customer self-service and conversational support channels without engaging software engineers to hardcode the changes. He was confident that you could build conversational systems that enabled non-technical users to make these changes.  In addition, he was particularly interested in how conversational systems could be married with visual channels on the web and mobile to deliver even better customer self-service solutions. These ideas led to NLX, which is used by companies ranging from Copa Airlines to Red Bull. More recently, NLX integrated with GPT-3 to provide users with generative AI solutions to augment conversational customer experiences. He stopped by the Voicebot Podcast to discuss the origins of the company and how things have changed with NLX clients since the introduction of ChatGPT. He also breaks down how NLX's new GPT-3-powered features work and how he expects adoption to play out. 

A lot happened this week in the generative AI and synthetic media. Today introduces a new weekly (or when appropriate) addition to the Voicebot Podcast. The GAIN Rundown is the generative AI news of the week. So much is happening in this space and it is so important to the conversational AI industry, we thought that a short weekly rundown of the top headlines would be useful. Let us know what you think.  The big news for this episode was Google's ChatGPT competitor Bard and Microsoft's debut of what we like to call BingGPT. We also saw schools banning ChatGPT and David Guetta show off an Eminem deepfake. The show starts off looking at some OpenAI data that you are likely to find interesting. If you would like to view the videos that we included in the discussion, you can see those segments on YouTube through the links below. 5:02 - Microsoft https://lnkd.in/gid_Gq4v 14:00 - Google https://lnkd.in/gZ6P8kCq 29:40 - David Guetta: https://lnkd.in/ghtNjsns Also, we are publishing these recorded videos on Voicebot's YouTube channel. If you would prefer to watch the discussion, subscribe to the channel and watch here: https://www.youtube.com/@voicebotai

Karen Kaushansky is a conversation designer at Google that led the Google Assistant UX design for WearOS and, more recently, for the Pixel Watch. While there has been a lot of attention around conversational UX on smart speakers and mobile phones, wearables introduce new variables and different mental models. Kaushansky goes into detail about designing voice experiences for the watch, what it's like to be an API or embedded in the software, how it's different when you also control the hardware or run software on the device, and more.  The interview also discusses how conversation design has changed over the past 25 years. Kaushansky started in the industry in the 1990s and has seen many technology shifts over the years. This also enables us to update our discussion on multimodal interfaces, which was the focus of her appearance on episode 40 of the Voicebot Podcast five years ago. We finish up with a discussion about large language models and the role of conversation designers in applications built on generative AI technologies. She also offers a great tip for designers on navigating this change that is the center of so much discussion today.  Kaushansky began her career as a speech technology designer at Nortel, then spent time at Nuance, Microsoft, and Jawbone. At Microsoft, she was part of the team that created Cortana and deployed it on the Windows phone. She joined Google in 2019 and has led user experience design for Google Assistant on a number of products. 

My guest is D-ID co-founder and CEO Gil Perry. We talk about how the company logically evolved into tools for creating talking digital people and how its capabilities in GANs and protecting consumers from facial recognition technology were the ingredients for a unique AI-based video solution. The company is well known for powering MyHeritage's Deep Nostalgia product, which has animated over 100 million photographs for consumers. D-ID was also instrumental in helping Jean-Baptiste Martinoli win two film festival awards for his AI-generated short film in 2022.   Last fall, the company introduced Creative Reality Studio. That solution enables anyone to upload someone's picture, add some text, and quickly create a scripted video with an avatar in the likeness of the photo. In December, D-ID added the ability to create the script using a prompt to GPT-3 and upload images created by Stable Diffusion.   This is a great example of how synthetic media is often enhanced by layering several generative AI solutions together. The new use cases are also why these markets are the hottest in tech today. Perry, a former software developer that worked on the viral hit mobile apps Meerkat and HouseParty, offers an insider's view of the rapid rise and current trajectory of generative AI and synthetic media.

The launch of ChatGPT on November 30, 2022, spurred new interest in conversational search. For the first time in over a decade, many people are beginning to think about what comes after the Google search model that has become so familiar. Dustin Coates knows a lot about search. He is the principal product manager the implemented Algolia's voice search products and worked on the integration with OpenAI's GPT-3 in 2021.  Algolia is a search giant in its own right, with over 17,000 customers using its website search capabilities instead of Google technology for 1.75 trillion annual searches. Coates walks through different types of search such as keyword, semantic, concept, and conversational. He breaks down how machine learning and AI are changing search models and performance. This includes a comparison between how Algolia, Google, ChatGPT, and other services handle search today. Coates also offers insights into where GPT-3 powered search does and does not work for its clients and why concept search has become so popular.     

"The adoption of large language models and generative AI is booming, and I think it began with creativity use cases. And now we are seeing as it slowly moving toward productivity use cases.... and that's is going to be the most valuable trend over the next couple of years," says AI21 Labs CEO Ori Goshen.  AI21 Labs is known for developing a large language model and using it to develop products such as Wordtune and Wordtune Read. The company is focused on productivity gains for professionals, changing the way we write and consume written text, and providing the means for other companies to build new applications using LLMs.  Prior to AI21, Ori Goshen was the founder and VP of technology for Crowdx which was acquired by Cellwize. He was the founder and VP of R&D for Tawkon and an entrepreneur-in-residence at Cisco. He has a background in cybersecurity, software development, and big data analytics.

Bret Kinsella wrote this: Jasper AI quietly built one of the top AI-based writing assistants atop OpenAI's GPT-3 large language model (LLM). Then, suddenly its growth and recognition exploded ... in a good way. That led to a $125 million series A funding round that was well-timed ahead of the new interest in the AI-writing assistant space after the introduction of ChatGPT.  Shane Orlick is president of Jasper AI and walks through the company's origins, the product, and how customers use these tools today. He even mentions some product features that have not yet been announced (breaking news on the Voicebot Podcast once again) and how users are applying the new Jasper Chat (a ChatGPT-like interface) versus the templates that Jasper has created and refined for specific use cases. For those of you interested in the technical stack behind Jasper and the company's move to build an internal NLU and new AI models that supplement the OpenAI APIs, this may be the only conversation out there with that insight.  Shane also discusses the broader market news, such as the impact of ChatGPT, the emergence of text-to-image models, and the rumors about OpenAI's big valuation and potential new investment from Microsoft. We cover a lot of ground around products, user behavior, generative AI, and the broader synthetic media market.  Jasper AI rewrote the above to this: Jasper AI is quickly becoming a leader in the AI-writing assistant space. Founded atop OpenAI's GPT-3 large language model (LLM), the company has seen tremendous growth and recognition, so much so that it was able to secure a $125 million series A funding round just before the rise of interest in AI-writing assistants with ChatGPT. Shane Orlick, president of Jasper AI, outlines the company's origins, product features (including some exclusive news!), and how customers are using these tools. In addition, he dives into the technical stack behind Jasper, which includes an internal NLU and new AI models beyond what OpenAI APIs offer. He also shares his thoughts on the broader synthetic media market, including ChatGPT's influence, text-to-image models, OpenAI's potential big valuation, and Microsoft potentially investing. This podcast offers an insightful look at what users can expect from Jasper as well as trends in generative AI and user behavior when it comes to writing and creating content with these tools. ***** Let me know which one you like better!

Taylan Kamis was inspired by the movie Her to pursue AI technologies that could make synthetic characters and voices more lifelike. After several years with Microsoft, including time on the media and applications team and serving as a CFO for some venture-stage startups, Kamis co-founded DeepZen in 2017. The first problem the DeepZen team sought to address was one of the harder ones in the industry: creating synthetic voices that were high enough quality to be used as narrators for audiobooks. A key element of this problem is the length of the content. The synthetic voice or voices must be pleasing enough to be suitable for long passages and hours of listening at a time. Another important element is the emotive quality of the synthetic voices.  DeepZen today provides audiobook production services and enables voice actors to create custom voices and monetize them without having to be in the studio for every project. We talk at length about the audiobook solution and how it works. That is followed by a discussion around new applications that are taking DeepZen into even larger markets.