26 September 2023

Google’s Secret Weapon To Destroy OpenAI For Good (#159)

1Frameworks

9Insights

Frameworks in this episode

Strategy6 steps

Existing-Asset AI Accelerant

Use proprietary data, products, and distribution to make AI more valuable.

Read →

Insights & moments

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

Myth Buster· 1

Myth Buster18:00

Platform Access Does Not Settle AI Training Rights

The hosts distinguish technical access to content from permission to use it commercially for AI training. They discuss platform terms, reports about training practices, and the backlash faced by communities whose content owners seek to monetize user-generated data.

Possessing content does not eliminate contractual or ethical constraints.
Platform terms may prohibit commercial reuse by outside companies.
Communities can resist having their contributions sold for model training.
Reddit and X illustrate the strategic value and controversy of proprietary community data.

“YouTube's terms of services does not allow anyone to use their data for uh for uh business use”
Kieran Flanagan · 18:00

“we see basically what we're seeing is like all of these Community sites selling their communities data to build AI models”
Kieran Flanagan · 19:00

#data-rights#training-data#platforms#privacy

Hot Take· 3

Hot Take10:30

Why Major AI Companies May Welcome Regulation

The hosts argue that regulation often strengthens established companies because incumbents possess the resources and influence to shape and satisfy new rules. In AI, demanding regulation could make open-source alternatives less viable for enterprise adoption while protecting current leaders.

Compliance costs weigh more heavily on smaller challengers.
Incumbents may help shape rules that entrench their position.
Enterprise restrictions could weaken open-source model adoption.
Calls for safety regulation can also carry competitive incentives.

“when the government regulates you the bet it's good for the incumbent”
Kieran Flanagan · 12:00

“they could put some checks and balances in that make open source models unviable for Enterprise companies”
Kieran Flanagan · 12:30

#regulation#incumbents#open-source#competition

Hot Take13:00

Google and OpenAI Are Fighting Through Announcements

The hosts characterize the companies' competing multimodal announcements as public one-upmanship rather than proof of shipped superiority. Their practical conclusion is that product delivery and real-world performance matter more than strategic posturing.

Google promoted Gemini's forthcoming multimodal capabilities.
OpenAI rumors quickly emphasized similar or faster ambitions.
Announcements do not establish product quality or availability.
Users should judge working products rather than competitive messaging.

“don't posture publicly let's ship it let's let's use it let's see what happens”
Kit Bodner · 15:30

“they are doing something that I have not seen large brands do in a long time just like one up on each other on the…”
Kieran Flanagan · 03:30

#google#openai#public-relations#product-launches

Hot Take26:30

Why Microsoft Needed OpenAI to Keep Office Relevant

Kit argues that Microsoft's OpenAI investment was strategically necessary, not merely opportunistic. If Google can build conversational coordination across its productivity suite, Microsoft must deliver an equivalent experience across Office to defend its position.

AI is becoming central to front-office productivity.
Google's integrated experience pressures Microsoft Office.
OpenAI gives Microsoft a route to comparable capabilities.
Failure to respond could accelerate market-share erosion.

“this is also why Microsoft gave open AI so much money because they have to do this to have office be relevant”
Kit Bodner · 26:30

“it's just a strategic must must have”
Kit Bodner · 26:30

#microsoft#openai#office#strategy

Explainer· 2

Explainer04:30

Why Multimodal AI Is the Next Competitive Battlefield

The hosts explain multimodal AI as one model handling text, video, images, and audio rather than forcing users into separate point solutions. They position Gemini and the anticipated GPT-5 as competing attempts to collapse these capabilities into unified systems.

Most AI interactions were still primarily text-based.
Multimodal models combine several media types in one core system.
Google designed Gemini as multimodal from the ground up.
The hosts expect multimodality to define the next phase of AI competition.

“the game at Play here here is actually having one core llm that can just integrate all these things”
Kieran Flanagan · 06:00

“multimodal is like the battlefield for future of AI dominance because it collapses all of these use cases into a single platform”
Kieran Flanagan · 14:30

#multimodal#gemini#openai#llms

Explainer16:30

YouTube Data Could Teach Gemini More Than Facts

The episode discusses reports that Gemini may be trained on YouTube transcripts and emphasizes the scale and freshness of that corpus. Kit extends the idea beyond factual knowledge, arguing that creator videos contain personality, perspective, emotion, and communication patterns that could support richer models.

YouTube contains a vast corpus of creator knowledge.
Google has native and potentially real-time access to new videos.
Video transcripts contain viewpoints as well as factual information.
Personality modeling could become a distinct capability built from creator data.

“YouTube is the second largest search engine”
Kit Bodner · 17:00

“you also have what probably the best personality model in the world where you could actually build an llm that can replicate and predict personality…”
Kit Bodner · 20:00

#youtube#training-data#personality#gemini

Q&A· 1

Q&A06:30

Will AI Bundle Every Tool or Make Fragmentation Easier?

Kieran argues that natural-language interfaces could allow enormous numbers of capabilities to coexist inside unified platforms. Kit counters that the same interface could also coordinate hundreds of specialized models from different providers, making backend fragmentation invisible to users.

Traditional software platforms bundle point solutions over time.
Natural language may reduce the usability limits of large bundles.
A unified interface does not require a unified backend.
Specialized open-source models may remain valuable for distinct problems.

“there's only so much you can bundle to have people actually tolerate it before it gets unbundled again”
Kit Bodner · 07:30

“I'd argue it just as well supports fragmentation because it allows you to use very fragmented technology and a more unified way or connected way”
Kit Bodner · 09:30

#bundling#platforms#open-source#user-experience

Tool· 1

Tool21:00

What Early Bard Extensions Can Do Across Google Apps

Kieran reports using Bard Extensions to search Drive and Workspace, assemble tables, locate documents mentioning specific people, and analyze YouTube content. The hosts also discuss travel planning, email retrieval, comment categorization, and drafting replies as examples of conversational workflows.

Bard Extensions can access selected Google services.
Natural-language retrieval can outperform rigid keyword search.
The system can combine information into tables and plans.
Future workflows may monitor document changes and organize comments.

“this is a hundred times better finding an email than car Google's current gmail search”
Kieran Flanagan · 25:00

“this is a hundred times better to categorize and find things in your documents than gmail search”
Kieran Flanagan · 25:30

#bard#extensions#workspace#productivity

Takeaway· 1

Takeaway24:30

Most People May Adopt AI Without Realizing It

The hosts predict that mainstream adoption will happen when AI quietly improves familiar activities rather than asking users to learn a new technology category. Better search, trip planning, document retrieval, and inbox assistance can make AI feel like an ordinary product improvement.

Users care about completed tasks more than underlying model terminology.
Existing applications provide natural entry points for AI.
A magical improvement encourages repeated use without formal education.
Invisible integration may onboard more people than standalone AI products.

“they're just going to onboard the world to artificial intelligence and the world's just not even going to realize it”
Kit Bodner · 24:30

“it is the ultimate like user adoption Trojan Horse”
Kieran Flanagan · 24:30

#adoption#user-experience#consumer-ai#productivity