11 May 2023

Leaked Google Memo Reveals A Huge Opportunity For Entrepreneurs (#117)

2Frameworks

8Insights

Frameworks in this episode

Innovation5 steps

Depth-Over-Breadth AI Opportunity Filter

Turn proprietary niche data into a focused, defensible AI product.

Read →

Strategy5 steps

Freemium Upmarket Flywheel

Monetize upmarket, then use free access to defend the lower market.

Read →

Insights & moments

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

Hot Take· 2

Hot Take06:00

AI Could Deaggregate the Web 2 Giants

Kieran argues that AI is reversing the aggregation dynamic that produced dominant Web 2 platforms. Small teams can now construct differentiated interfaces and experiences around particular datasets instead of requiring internet-scale infrastructure.

Web 1 was decentralized and Web 2 consolidated around aggregators.
AI reduces the infrastructure needed to challenge large platforms.
Specialized chatbot experiences can compete in narrow markets.
Incumbents' foundational technologies now enable their challengers.

“It basically deaggregates the aggregators”
Kieran Flanagan · 06:00

“Then you can compete with a multi-trillion dollar company”
Kieran Flanagan · 06:00

#web3#aggregation#disruption#platforms

Hot Take14:00

Why Meta May Be the Secret Winner of the AI Wars

The hosts argue that Meta gained strategic leverage when developers began building on LLaMA. By enabling an open-source ecosystem, Meta effectively recruits a large external community to improve, extend, and normalize its technology.

LLaMA became a popular foundation for open-source development.
External contributors provide Meta with ecosystem leverage.
Open-source adoption can outweigh some costs of relinquishing control.
The strategic prize may be the ecosystem rather than the foundational model itself.

“they've just inherited millions of workers for free”
Kieran Flanagan · 14:30

“Your success is directly dependent "on how many people want you to succeed.”
Kipp Bodnar · 15:00

#meta#llama#ecosystems#community

Explainer· 2

Explainer02:30

How Open Source Collapsed the Cost of Building AI

The hosts explain how leaked and publicly released models allowed individuals and small teams to reproduce capabilities once limited to major research organizations. Vicuna is presented as an example of a chatbot approaching leading commercial quality after an iteration reportedly costing only hundreds of dollars.

LLaMA accelerated open-source model development.
Small teams can now run meaningful AI experiments.
Vicuna reportedly approached leading chatbot accuracy at dramatically lower cost.
The barrier shifted from major research labs to individuals with capable laptops.

“this chat bot was trained for $300.”
Kieran Flanagan · 03:00

“the barrier to entry for training and experimentation has dropped from the total output of a major research organization to one person and free evening”
Kieran Flanagan · 03:30

#open-source#llama#vicuna#ai-costs

Explainer08:30

Why On-Device AI Changes the Competitive Paradigm

The hosts describe an unintended consequence of years spent improving phones, browsers, and connectivity: capable AI models can now run on ordinary devices. Infrastructure incumbents expected AI to remain too expensive for local execution, but smaller models undermine that assumption.

Modern browsers and phones can execute increasingly capable AI systems.
Smaller models reduce dependence on centralized infrastructure.
On-device execution democratizes experimentation and company formation.
Foundational technology investments can empower future disruptors.

“You can run the large language models on a Pixel phone.”
Kieran Flanagan · 09:00

“the future and the innovation is often a function of unintended consequences”
Kipp Bodnar · 09:30

#on-device-ai#mobile#browsers#disruption

Story· 1

Story04:00

The Google Research That Helped Make OpenAI Possible

Kipp traces the current AI wave to Google's public transformer research. The story illustrates how publishing foundational technology can enable external companies and communities to create unexpected competitive threats.

Google and DeepMind conducted foundational AI research.
The transformer paper underpinned modern large language models.
OpenAI built on technology disclosed publicly by Google.
Open publication can create powerful unintended successors.

“Without that transformer white paper being published to the public, OpenAI wouldn't exist.”
Kipp Bodnar · 04:30

“open sources really just means building in public.”
Kipp Bodnar · 05:00

#transformers#google#openai#research

Q&A· 1

Q&A22:00

Do Competitive Moats Still Exist in the AI Era?

The hosts question whether rapid technological commoditization is erasing traditional software advantages. They conclude that hardware and data infrastructure may retain defensibility, while community, brand, and possibly first-mover distribution remain important unresolved candidates.

Software features and models are becoming easier to copy.
Execution may matter more as technical differentiation declines.
Hardware and data warehouses may retain structural advantages.
Community, brand, and first-mover distribution remain possible moats.

“do moats exist anymore?”
Kipp Bodnar · 22:00

“community and brand is a moat”
Kieran Flanagan · 23:00

#moats#brand#community#first-mover

Tool· 1

Tool11:00

The Coming Plug-and-Play Platform for Custom Models

Kieran predicts platforms where businesses feed in their own information and receive a model they can connect to applications. Such a service could make custom AI development resemble a drag-and-drop product workflow rather than a specialized research project.

Businesses may create models by supplying their own feeds or datasets.
Platforms could handle model construction and application integration.
Drag-and-drop tooling would broaden access beyond AI specialists.
The winning AI ecosystem platform has not yet been established.

“we can just feed in our data”
Kieran Flanagan · 11:00

“people will just wanna plug in their data and be able to build front end apps themselves on that data”
Kieran Flanagan · 11:30

#no-code#custom-models#platforms#ai-tools

Takeaway· 1

Takeaway16:00

Businesses Will Use Many AI Models, Not One Winner

The discussion rejects the idea that a single large language model will control every business workflow. Companies will combine proprietary and third-party models according to the data, economics, and capabilities each task requires.

Open source reduces the value concentrated in foundational models.
AI markets are likely to contain both large and small providers.
Businesses will select different models for different jobs.
Some organizations will build models while buying access to others.

“it's not gonna be one large language model rules all of the business that you want to get done”
Kipp Bodnar · 18:00

“There are gonna be a lot of diverse players in this market, some big ones, some small ones”
Kipp Bodnar · 18:00

#ai-strategy#multi-model#open-source#business