Report VC Community drinks November 9th 2023 – Plenary session
VEECEE AI-event: ‘Hemmingway can now talk Mandarin’
A very cool VC-event took place at Google at 9 November 2023 on AI in action with experts Diederik Veelo and Sohrab Hosseini.
Photography by Robert Tjalondo
Coming into the latter half of 2023, there is little doubt as to what is the Big Theme in tech at the moment: artificial intelligence. Big things are afoot. With new and better iterations of the big AI-models being released almost every week and the recent meeting of world leaders at Bletchley Park to discuss the impact of machine minds on freedom and democracy, it seems we are reaching an important turning point in the trajectory of the technology.
On a smaller scale, investors are facing the questions of what AI means for their existing portfolio companies. Many VC’s in the ecosystem are scrambling to pull together the cash to build the startups in the implementation layer that translate the large, cloud based models into practical applications for consumers and businesses.
Gathering at the Amsterdam office of Google the VEECEE-community was welcomed by Raphael Neuhaus, who manages the Google services delivered to Dutch startups and scaleups and is obviously well positioned to speak to the crowd as to what the latest generation of Google’s AI can contribute to their portfolio companies.
The community welcomed Diederik Veelo on stage. Diederik is Head of Innovation at creative production agency Ambassadors and founder of creative production platform Cube and Vocoda.ai, a voice synthesis studio that produces realistic voices for advertising use with the help of AI. Customers include Apple TV, HBO, Adidas and G-Star.
He gave the VC-professionals a glimpse of how AI is transforming advertising at the moment. With the current pace of developments in AI, it’s currently more questions than answers, but there’s one thing we no longer have to ask ourselves: we now know what the famous American writer Ernest Hemingway sounds like in Mandarin.
The Amsterdam ad agency used speech-to-speech AI to synthesize a voice model that mimics Hemingway’s voice in Mandarin and have it read ‘the old man and the sea’, as part of an ad campaign for the Chinese branch of the Penguin publishing house. It was one of the many applications Veelo and his agency have developed to integrate generative AI into the creative process.
Another interesting case was the campaign they developed around Lionel Messi – the footballer is a reluctant actor with very little time, so in order to save on the development time of multiple campaigns they used body doubles and AI morphing to get him into many more ads than would have been possible with traditional filming.
Maybe even more astounding than the impersonating tricks shown was the incredible speed with which the generative AI is developing. Leaps and bounds are measured in weeks rather than years or months. Veelo showed a couple of iterations of the prompt ‘women in front of the statue of liberty’ spaced a couple of weeks apart over the summer of 2022, which went all the way from slightly creepy and definitely surreal pictures at the start to pictures that would be perfectly acceptable in an advertising campaign, with important aspects like picture resolution and color depth rapidly catching up with traditional standards.
Next on stage was Sohrab Hosseini of Orquesta. Orquesta was recently backed by Curiosity VC and Adriaan Mol and provides SaaS companies with a unified gateway for direct integration of various LLM models, including those from OpenAI, Azure, Google, Hugging Face, and Cohere.
Hosseini explained how companies like his in the middleware are providing the plumbing that connects the AI ecosystem. The purpose of this middleware layer is twofold. The first is to form a bridge between the large models provided by the big AI companies like Google and OpenAI and the practical applications at the consumer and entrepreneur level. Data management, load switching, connecting multiple AI modes and models, cost control – these are all tasks that are done in the middleware layer.
The second purpose of the middleware is to build a bridge between the users with engineering chops and non-technical personnel: by providing no-code workflow the power of AI-tools are put in the hands of a much larger group of domain experts, integrating the large AI-models with the practical needs of users executing their specific business needs.
Hosseini provided the audience with a couple of pointers: don’t try to do everything at once but start with a couple of small, clearly defined use cases; don’t rely on one model but try to incorporate multiple models and modalities (with the added note that in addition to the large, general models a new crop of smaller and more specialized models is coming up). It’s also essential to not get bogged down too much by the friction in the current generation of models – with the speed of development most problems will be ironed out pretty fast.
Inspired by the creative possibilities of AI and provided with an overview of how to turn large models into practical applications with the help of middleware the community had a lot to talk about over the ensuing drinks at Google.
This VEECEE Community event was sponsored by Google Cloud and Houthoff.
Google Cloud is committed to supporting VC’s in their mission to invest in high-potential companies. Portfolio startups can use their cloud platform free of charge, through Google Cloud’s startup program. And even though they build their products for free with Google’s cloud, they receive help from Google’s engineers. This is explained in this case study about an Amsterdam portfolio company from Antler.
Startup programma
https://cloud.google.com/startup
Case study
https://cloud.google.com/customers/peckish
Click here to view the photos of the VeeCee Community drinks. Hope to see you next time!