28 September 2023

GPT4 Gets Massive Upgrade… You’ll Never Use Google Again (#160)

1Frameworks

10Insights

Frameworks in this episode

Marketing6 steps

AI Feedback and Acceleration Loop

Use AI before and after launch to improve marketing faster

Read →

Insights & moments

The myth-busts, hot takes, explainers, and tools worth keeping.

Myth Buster· 1

Myth Buster05:00

Why Personalized Tutorials Still Aren't Personal Enough

Traditional search and YouTube tutorials may be directionally useful while failing on model-specific details. Multimodal AI instead generates guidance for the exact bicycle, vacuum, slide, or other object in front of the user, creating what Bodner calls a one-of-one experience.

Generic tutorials often demonstrate the wrong product model
Small physical differences can make instructions unusable
Visual context enables advice tailored to one user's situation
One-of-one guidance differs from conventional personalization

“it's like oh that's actually not the exact vacuum model that I have”
Kit Bodner · 05:30

“it creates a oneofone personal experience not personalized experience”
Kit Bodner · 05:30

#personalization#youtube#search#multimodal ai

Hot Take· 4

Hot Take09:00

Personal Voice Models Could Make Everyone Multilingual

The episode predicts that individuals will eventually maintain AI voice models capable of speaking additional languages and producing content without repeated recording sessions. Bodner sees this changing how people create, teach, and communicate, although the transcript does not deeply address consent or authentication safeguards.

Professional actors helped develop OpenAI's voice models
A personal voice model could speak multiple languages
Creators may produce more content without rerecording
Voice synthesis could reshape teaching and communication

“one of the things that I believe that's happening in the future is that we're all going to have a voice model”
Kit Bodner · 09:30

“it is going to make creating teaching living very very different than it is today”
Kit Bodner · 09:30

#voice cloning#multilingual#content creation#text to speech

Hot Take12:00

Could Synthetic Voices Enable 'Deep Real' Communication?

Bodner acknowledges fears that synthetic multilingual speech will make impersonation and deepfakes easier. He contrasts those risks with an optimistic possibility: authenticated individuals controlling their own voice models to create more genuine cross-language interactions, while admitting the real outcome remains uncertain.

Synthetic speech raises impersonation and deepfake concerns
User control may distinguish authentic uses from deceptive ones
The same technology could broaden access to trusted speakers
The balance between risk and authenticity is unresolved

“people are now able to speak in languages that they don't speak”
Kit Bodner · 12:30

“I don't know how this is going to play out in practice”
Kit Bodner · 12:30

#deepfakes#trust#voice ai#identity

Hot Take13:00

AI Could Make New Businesses Global From Day One

Automatic language generation and translation could let businesses market and operate across languages from their inception. Bodner argues that this development will accelerate the internet's erosion of commercial borders and expose more people to previously inaccessible information and perspectives.

Language has limited access to creators and expertise
AI translation can expand a business's addressable market immediately
Marketing and operations may become multilingual by default
Cross-border commerce could accelerate beyond the internet's first wave

“businesses as I've said on this show in the past are going to be Global by default”
Kit Bodner · 13:30

“we are going to break down the borders between business and Commerce”
Kit Bodner · 13:30

#global business#localization#commerce#translation

Hot Take18:00

AI Speed Could Eliminate the Marketing Middle Ground

As inexpensive AI feedback helps everyone move faster, Bodner expects markets to react more quickly to news, memes, and emerging topics. Marketers who arrive slightly late may no longer capture residual value because competitors can iterate and publish at near-immediate speed.

AI compresses the time from idea to market
Faster production causes audiences and competitors to react sooner
Late participation in trends may become less valuable
Speed must accompany quality and relevance

“you're going to be able to get ideas to Market much faster”
Kit Bodner · 18:30

“if you are not fast you're going to be left behind there's going to be no middle ground”
Kit Bodner · 18:30

#marketing speed#competition#real-time marketing#ai

Explainer· 3

Explainer03:00

A Photo Turns GPT-4 Into a Personal Repair Assistant

Bodner explains how multimodal GPT-4 can inspect a photograph of a real object and provide situation-specific instructions. The bicycle demonstration shows the model identifying components, correcting misunderstandings, and helping the user find the correct tool instead of serving a generic tutorial.

A photograph can replace a lengthy written description
Follow-up images let the model verify each action
Instructions adapt to the user's exact object and situation
Interactive guidance can remove small knowledge barriers

“chat GPT is able to give stepbystep instructions for that bike based on a simple Snapshot from your phone”
Kit Bodner · 03:00

“you essentially have an expert who's there to walk you through step by step”
Kit Bodner · 04:30

#gpt-4#multimodal ai#computer vision#tutorials

Explainer07:30

Voice and Images Solve AI's Blank-Screen Problem

Bodner argues that multimodality makes ChatGPT easier for mainstream users because they can begin with a photograph or spoken question instead of composing a sophisticated prompt. Voice-to-voice interaction and suggested prompts further reduce the effort required to discover useful applications.

Smartphones make image input familiar and immediate
Voice interaction reduces dependence on typing
Suggested prompts help users discover possible tasks
Lower input friction could bring a new population into AI tools

“chat GPT just got 10x easier to use”
Kit Bodner · 07:30

“you don't have this like kind of the blank screen problem”
Kit Bodner · 08:30

#ai adoption#voice ai#user experience#chatgpt

Explainer10:00

Spotify Translates Podcasts Without Replacing the Host's Voice

Spotify's collaboration uses GPT-4-powered voice translation to create local-language podcast editions that retain the host's vocal identity. Bodner contrasts this with conventional dubbing, which requires other actors, costs more, takes longer, and can feel less authentic.

The test includes prominent podcasters such as Lex Fridman
Translated episodes preserve the creator's voice
The approach reduces reliance on traditional dubbing
The technology could eventually reach smaller creators

“what it's doing is that it is creating local language versions of your own voice and dubbing that over”
Kit Bodner · 11:00

“all creators will be having access to this technology over the next 12 to 18 months”
Kit Bodner · 11:30

#spotify#podcasting#translation#voice ai

Takeaway· 2

Takeaway14:00

Multimodality May Be AI's Mainstream Adoption Trigger

The host argues that ChatGPT's first wave consisted largely of curious early adopters, while image and voice interfaces make it practical for relatives, students, and other mainstream users. He expects broad rollout to drive another wave of accounts and paid usage over the following months.

Initial adoption favored technically curious users
Image and voice interfaces fit familiar everyday behavior
Mainstream adoption may expand over a three-to-nine-month rollout
Consumer use could prove more important than previously expected

“now there's a version of chat GPT that's for everybody”
Kit Bodner · 14:30

“we thought these models were going to be more used by developers than in consumers”
Kit Bodner · 15:00

#ai adoption#consumers#chatgpt#multimodal ai

Takeaway19:30

Follow the Money Behind the AI Platform Race

The episode closes by comparing Amazon and Google's investments in Anthropic with Microsoft's relationship with OpenAI. Bodner interprets these alliances as evidence that incumbent technology companies view large language models as a foundational strategic layer rather than a passing feature.

Google had invested $300 million in Anthropic
Amazon invested $1.25 billion with an option for more
Microsoft maintains its major relationship with OpenAI
Capital and partnerships reveal where incumbents expect growth

“Google had already invested 300 million in anthropic in may now Amazon has invested 1.25 billion”
Kit Bodner · 20:00

“follow the money Society always shows you where things are going”
Kit Bodner · 20:30

#anthropic#openai#big tech#ai investment