08 June 2023

Sam Altman’s Leaked Meeting Notes Reveals OpenAi’s Future (#127)

0Frameworks

8Insights

Insights & moments

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

Myth Buster· 1

Myth Buster19:30

Users Want ChatGPT Inside Their Apps, Not Apps Inside ChatGPT

OpenAI reportedly found weak product-market fit for most ChatGPT plugins. The more promising direction is to embed natural-language capabilities into existing products, combining conversational commands with the visual interfaces users already understand.

ChatGPT plugins reportedly lacked broad product-market fit
Users prefer enhancing familiar apps with natural language
Visual and conversational interfaces solve different interaction problems
Chat can replace complex menus for clearly expressed tasks
Most major software platforms are likely to add a natural-language layer

“a lot of people thought that they wanted their apps to be inside ChatGPT but what they really wanted was ChatGPT inside of their apps.”
Kieran Flanagan · 19:30

“humans basically want both a visual user interface and a chat user interface”
Kip Bodner · 20:30

#chatgpt#plugins#product-market-fit#user-interface#natural-language

Hot Take· 2

Hot Take09:30

Custom AI Models Will Become as Routine as CRM Objects

Businesses may eventually operate many specialized models rather than relying on one universal copilot. Kip compares these models to custom CRM objects: ordinary building blocks configured for specific workflows, customer activities, and automation needs.

Companies will use multiple narrow copilots
Custom models can support distinct customer workflows
AI models may become ordinary business configuration objects
Vertical specialization can improve product stickiness
Ubiquity may turn sophisticated models into commodities

“we will have custom models, like custom objects.”
Kip Bodner · 10:00

“they're going to be as small and as normal as like a custom object in a CRM.”
Kip Bodner · 10:30

#custom-models#copilots#crm#business-automation

Hot Take24:30

OpenAI’s Future May Look More Like AWS Than a Consumer App Company

Sam Altman reportedly said OpenAI did not plan to build additional consumer products beyond ChatGPT. The hosts interpret this as a commitment to becoming a dependable platform on which developers and businesses can build without constantly fearing competition from their infrastructure provider.

Competing with customers undermines platform trust
OpenAI reportedly plans no additional consumer products beyond ChatGPT
A platform strategy creates space for founders to build complementary products
The comparison is closer to AWS or Azure than a consumer software portfolio
Developer ecosystems can become the primary source of durable platform value

“outside of ChatGPT, they are not going to build any more consumer products.”
Kieran Flanagan · 25:00

“what that really means is he's going to build a true platform.”
Kip Bodner · 25:00

#openai#platform-strategy#developers#aws#entrepreneurship

Explainer· 4

Explainer02:00

AI’s Biggest Constraint Is Hardware, Not Software

OpenAI and other AI companies face a physical constraint on progress: scarce GPU capacity. The hosts connect this bottleneck to NVIDIA’s valuation and explain how concentrated demand for computing hardware can constrain the entire AI supply chain.

GPU shortages restrict AI model development and deployment
Hardware capacity affects model depth, speed, and scale
NVIDIA benefits from demand spanning cryptocurrency and AI
Supply-chain bottlenecks can determine which companies advance fastest

“the biggest limitation all these companies have right now is GPU shortages.”
Kieran Flanagan · 02:00

“AI's biggest problem isn't software, it's hardware.”
Kip Bodner · 02:30

#gpu#nvidia#ai-infrastructure#supply-chain

Explainer03:30

Collapsing AI Costs Could Create a Wave of Niche Businesses

The cost of training and operating sophisticated language models is falling rapidly, making previously uneconomic products viable. The hosts predict that inexpensive intelligence will support specialized search engines, smaller companies, and narrowly focused AI products serving individual verticals.

GPT-3-level training costs have fallen sharply
Lower inference costs make large-scale chatbot services viable
Vertical search products can succeed with smaller revenue bases
Cheaper intelligence lowers the barrier to starting focused businesses
Economic viability matters as much as technical capability

“they just want to drive the cost of intelligence down.”
Kieran Flanagan · 03:30

“we are gonna have more smaller businesses because costs are gonna come down so low that you can have a more niche and more focused…”
Kip Bodner · 08:00

#ai-economics#vertical-ai#startups#llm-costs

Explainer11:00

Why Larger Context Windows Change What AI Can Understand

Context windows determine how much material an AI session can hold and manipulate at once. Expanding them from tens of thousands toward a million tokens would let users analyze long videos, contracts, lectures, strategies, and potentially entire media franchises in a single session.

Token limits determine how much information a session can process
GPT-4 could handle roughly a short video transcript at the time
Claude’s 100,000-token window could accommodate a two-hour transcript
Million-token windows would support much larger bodies of source material
More context enables deeper summarization, querying, and repackaging

“token limits is basically how much data can that AI session hold so you can manipulate that data and do something with it.”
Kieran Flanagan · 11:30

“He wants to get to a million tokens”
Kieran Flanagan · 14:00

#context-windows#tokens#claude#gpt-4#summarization

Explainer15:30

Multimodal AI Could Personalize Every Marketing Format

Multimodal systems can ingest and generate text, images, and video within one interface. This would let marketers evaluate complete customer experiences, regenerate weak creative assets, and personalize content across formats rather than optimizing text in isolation.

Multimodal AI connects text, image, and video workflows
Marketing audits could evaluate complete emails and onboarding experiences
Systems could generate improved versions of weak creative assets
Visual behavior data could drive real-time website changes
Cross-format understanding enables personalization at unprecedented scale

“Multi-modality is the ability for AI to understand text, video, images and give you back text, video, images.”
Kieran Flanagan · 15:30

“I think that is going to lead to mass personalization for our customers on a scale that we have not seen to date”
Kieran Flanagan · 17:30

#multimodal-ai#personalization#marketing#creative-ai

Takeaway· 1

Takeaway25:30

The Money Is Usually in the Boring Layer

The hosts argue that infrastructure and business-facing platforms often monetize more effectively than glamorous consumer products. Consumers expect flashy products to be cheap or free, while businesses reliably pay for foundational tools that create measurable operational value.

Boring infrastructure can have stronger monetization
Consumer excitement does not guarantee attractive economics
Business customers pay for useful foundational capabilities
Platform revenue can be more durable than consumer-product revenue

“The money is in platform, not consumer.”
Kip Bodner · 25:30

“if you want to make money, follow boring things, the more boring the thing is, the more there is money in it”
Kip Bodner · 26:00

#business-models#b2b#monetization#platforms