Antonym: The (um... sorry, er) Disfluency Edition

I need an AI to make me sound more human.

Dear Reader

This week has been, as they say, a lot. So I’m going to lean on AI and have some of the brilliant apps, bots and systems I work with help me out.

I’ll be clear where AI has helped out, by reporting what these systems said rather than just copying the outputs. Hopefully, this will also be a good way to share some of the tools we’re using in our daily work at Brilliant Noise.

Prepared Minds launch

First of all, we held our online launch event for the Prepared Minds paper on AI literacy. The AI literacy framework is this bit…

…and here is a link to the full video if you’d like to check it out. The paper will be released later this week for general circulation, so please reply to this email if you’d like a copy.

I used Opus, a tool that analyses your long-form videos to spot clips that will be interesting to viewers. Then it lets you edit them into whatever format you like – TikTok, Insta square or 16:9 - and add your brand look and feel. See the screenshot below and note the scoring on this number one (of thirty) suggested clips.

Here’s the clip it liked from the webinar:

That’s actually an apt clip for this newsletter. I am more cognitively constrained than a typical day. After an extremely exciting and demanding week at work, on Saturday we helped my daughter move into her university halls. Take an emotional whirl-y-gig of a day, layer on a road-trip and there’s not much left in the tank the following day for writing newsletters.

Podcast has existential crisis

But I have to talk a bit about NotebookLM’s podcast feature, which we mentioned in Antonym: The Competence Porn Edition a couple of weeks ago.

The buzz that started then has turned into a massive sonic boom, as more people have heard its incredible and uncanny ability to turn anything into a conversation between two upbeat podcast hosts.

I say upbeat. But some joker gave it a text about how the product was made, which caused the “hosts” to discuss the fact that they weren’t real.

The effect is a little disturbing. What is actually happening is this: an AI has written the script that has been voice acted by another AI. So no intelligences were harmed in the making of it!

What’s REALLY interesting about NotebookLM

On a more serious level, here are the two (probably) real hosts of the NY Times Hard Fork podcast talking to Steve Johnson of Google, who led the development of NotebookLM.

Essentially, it’s a series of bots, feedback loops for content improvement, and a crucial phase where the AI introduces disfluencies into the text. Disfluency is definitely my word of the week, and if I were starting a new newsletter today, it might very well be the new name.

This interview sparked a lot of thoughts about how products are being developed and how easy it would be for anyone to create something like NotebookLM’s audio feature.

I discussed it with a lot of people this week, so I asked the AI chatbot built into Otter, a transcription tool we use to record meetings to summarise what my thoughts were:

You find several aspects of Google's NotebookLM podcast generation feature noteworthy:

Conversational Format: The feature generates a podcast where two AI voices have a conversation about the uploaded content, making it more engaging and natural-sounding than a simple text-to-speech conversion.

Rapid Production: The entire process of creating a podcast from a document takes only about five minutes, which is remarkably efficient.

Multi-step AI Process: The podcast creation involves several AI-driven steps:

• Initial script generation

• Script critique and improvement

• Addition of disfluencies (ums, ahs, stutters)

• Emotion and intonation synthesis

Iterative Improvement: The system uses feedback loops to refine the script, with one bot critiquing the output and another improving it based on that feedback.

Human-like Elements: The addition of disfluencies and emotional intonation makes the AI voices sound more natural and less robotic.

Versatility: You've used this feature to create podcasts from various documents, including business reports and wealth reports, demonstrating its adaptability.

Accessibility: The feature makes dense written content more accessible by transforming it into an easily digestible audio format.

Potential for New Content Forms: You see this as opening up possibilities for entirely new forms of content creation and dissemination.

You've described this feature as "mind-blowing" and "the most amazing AI product I've seen this week," highlighting how impressed you are with its capabilities and potential implications for content creation.

Yep. That. Thanks, bot.

The Crowd + Lab strategy

Next, I wouldn’t call this a recommendation—it’s more of a demand that you read Ethan Mollick’s newsletter this week about AI and organisations. It aligns closely with what we’ve been discussing in Antonym.

He emphasises the need to tap into the wisdom and innovation of the crowd and to establish a lab for developing more advanced uses of generative AI within businesses. You can’t achieve this by simply buying in solutions; he argues that you have to educate everyone, give them space to experiment, and celebrate those experiments.

That’s exactly what we’re doing at Brilliant Noise. Please take the time to read it, or listen to the podcast version that, naturally, he generated using Notebook LM.

And finally, some AI fireworks

Lastly, for some fun this week: Pika. It’s a video generation app that’s been around for a while and is great for feeding in photos or images to create five-second animations. Its latest generation of features lets you turn the subject of any image into a squeezy, squishy toy, inflate it like a balloon and watch it float away, explode, or get crushed under a press.

Give it a try! Here’s Jupiter getting squished:

The Antonym logo getting blown up:

And here’s my favourite: Pika animated one of the illustrations from an edition of Antonym—a room full of monkeys on typewriters.

That’s all for this week.

Thank you for reading, as always.

‘Bye.

Antony