A week in Generative AI: Toys R Us, Gemma & Perplexity
A week in Generative AI: Toys R Us, Gemma & Perplexity
News for the week ending 30th June 2024
Lots going on in the advertising industry this week with the launch of Toys R Us’s new ad, the first to be made with OpenAI’s SORA, a GenAI ad from Motorola and Meta saying that AI will increase the need for advertising agencies. We also had the release of Google’s second generation open model, Gemma 2, and continued controversy around Perplexity and their practices.
There are also some great articles about the power usage of training and using GenAI here and here and good long reads from John Gruber on Apple Intelligence, Sequoia on AI’s $600b Question and a great interview in Time with Dario Amodei.
First Ever Brand Film Created with Sora
So, we have the first ad made with OpenAI’s Sora, which has impressed many in the previews when the announced it back in February this year. Only a few professional film makers and studios have had access to Sora, so it’s difficult to judge how good it is in real life, but we now have a 60-second ad made with the model for Toys R Us.
The ad depicts the history of Toys R Us and its mascot Geoffrey the Giraffe and was made by Toys R Us studios, the retailer’s entertainment arm and the agency Native Foreign. Sora helped condense “hundreds of iterative shots“ down to a couple of dozen shots which were then edited together.
There has been some backlash in the advertising industry about the ad with creators criticising the quality of the ad and accusing OpenAI of training Sora on plagiarised work. This visual consistency of the young boy in the advert has also been criticised.
I’ll let you watch the ad and make up your own minds on what you think of it. Regardless, using GenAI in advertising will become the norm over the coming years as the technology improves and the industry figures out how best to use it. We also saw this GenAI ad from Motorola this week too.
Meta: AI will increase the 'need for agencies'
There are some interesting points of view here from Derya Matras, Meta’s vice-president of the UK, Northern Europe, Middle East and Africa. She believes that technology (AI) will replace many tedious tasks, especially in advertising agencies, so that people can prioritise brainstorming, strategic conversations, ideation etc. Derya also believes that this shift will require re-skilling people in the advertising industry and that we’ll need to create agile organisations and able to adapt to these new technologies.
I couldn’t have put it better myself!
OpenAI delays ChatGPT’s new Voice Mode
The new voice mode that OpenAI announced in May alongside GPT-4o has been delayed and now won’t be available to most people until the autumn. OpenAI say they need more time to improve the user experience, scale the infrastructure required and ensure guardrails prevent the model from generating certain content. Real-time voice interaction is coming, we’re just going to have to wait a few more months to get our hands on it!
Google announce Gemma 2, their latest open models
Google have launched the second version of their open Gemma models. Gemma 2 comes in 9B and 27B sizes and compete with Meta (Llama) and Mistral’s open models. The first Gemma models were only launched in February at 2B and 7B sizes, so this is a big step up.
Gemma 2 models currently beat the equivalent sized models from their competitors at various benchmarks, so it will be interest to see how they perform in the real world as more people get to test them.
How to stop Perplexity and save the web from bad AI
There has been a bit of controversy flying around about Perplexity, a new GenAI based approach to searching the web recently. They have been disregarding some standard online practices and scraping/recreating content from websites without permission, and doing so when those websites had specifically tried to stop their content from being scraped.
This is not a good look for Perplexity, throws shade at GenAI as a whole and goes against some of the fundamental principles of how the internet should work.
AI Ethics News
Researchers upend AI status quo by eliminating matrix multiplication in LLMs
Record labels sue AI music generator startups Suno, Udio for copyright infringement
NBC to use AI version of announcer Al Michaels’ voice for Olympics recaps
Hackers ‘jailbreak’ powerful AI models in global effort to highlight flaws
Researchers fool university markers with AI-generated exam papers
Long Reads
Daring Fireball - Apple Intelligence
Sequoia - AI’s $600B Question
Time - Anthropic CEO Dario Amodei on Being an Underdog, AI Safety, and Economic Inequality
TED - Ray Kurzweil: The Last 6 Decades of AI - and What Comes Next
The Guardian - AI scientist Ray Kurzweil: ‘We are going to expand intelligence a millionfold by 2045’
The Guardian - The most pleasant AI yet?
“The future is already here, it’s just not evenly distributed.“
William Gibson