21 March 2023

GPT-4 Update: EVERYTHING You Need To Know (#102)

2Frameworks

13Insights

Frameworks in this episode

Marketing6 steps

Freemium Value-to-Upgrade Sequence

Prove recurring value for free, then reserve the major capability leap for paid users.

Read →

Marketing5 steps

The WTF Moment Product Launch

Anchor every launch around one simple, surprising, shareable demonstration.

Read →

Insights & moments

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

Hot Take· 4

Hot Take16:00

AI Is a Platform Shift Because It Breaks Existing Systems

Kipp characterizes AI as a new platform rather than another isolated application. Its significance comes from forcing institutions and workflows—including education, productivity, and software—to be rebuilt around newly available capabilities.

Platform shifts alter multiple systems at once.
Education and examination systems face immediate pressure.
Note-taking and productivity workflows will be redesigned.
Businesses must treat AI as foundational rather than optional.

“AI is a new platform and it's a new platform, because it's breaking existing systems, right?”
Kipp Bodnar · 16:30

“And that's how you know that there's a new platform shift in the world”
Kipp Bodnar · 16:30

#platform-shift#business-models#systems-change#ai-adoption

Hot Take17:00

Yesterday's Funded Startup Can Become Today's Free Twitter Tool

A single developer connected GPT-4 to Google Sheets to generate sales-prospecting emails at scale. The hosts observe that functionality which previously might have required a funded engineering team can now be built by one person and released free for attention or distribution.

Individuals can create useful software with fewer engineering resources.
AI lowers the cost of building narrow product features.
Some former startup opportunities may become free audience-building tools.
Rapid commoditization changes where businesses can capture value.

“That is the commoditization of value.”
Kieran Flanagan · 17:30

#startups#commoditization#solo-builders#sales-tech

Hot Take29:00

AI Search May Make Brand More Defensible Than Traditional SEO

The hosts argue that AI-generated answers will commoditize much educational content and make conventional optimization less predictable. Because marketers may not know how a language model selects recommendations, broad brand awareness and trusted mentions could become more valuable than manipulating a transparent ranking formula.

Language models may not expose clear recommendation criteria.
Educational search content is increasingly commoditized.
Citations could make optimization somewhat more observable.
Strong brands may remain memorable and discoverable across changing interfaces.

“Educational content is commoditize.”
Kieran Flanagan · 29:30

“No one is gonna use a traditional search engine. It is gonna be totally archaic.”
Kieran Flanagan · 32:00

#seo#brand-marketing#ai-search#content-strategy

Hot Take33:00

When AI Commoditizes Production, Ideas and Distribution Win

The hosts predict that AI will reduce the execution advantage historically held by people with specialized production skills. As more ideas become cheap to build, competitive advantage shifts toward choosing better ideas, refining them, and earning distribution in an increasingly crowded market.

The gap between non-coders and coders is shrinking.
AI reduces the cost of producing apps, websites, and content.
More production creates more competition for attention.
Idea quality and distribution become increasingly important.

“With AI, the idea becomes everything.”
Kipp Bodnar · 34:00

“The actual production of the idea is going to be commoditized in a lot of situations.”
Kipp Bodnar · 34:00

#distribution#ideas#marketing#competitive-advantage

Explainer· 5

Explainer02:00

What Made GPT-4 Different From Earlier ChatGPT Models

The hosts highlight GPT-4's ability to accept images, generate longer text, rhyme more effectively, and reason over substantially larger inputs. These improvements expand the system from a conversational text tool into a platform for document analysis, visual interpretation, coding, and content creation.

GPT-4 can accept images as inputs.
It can generate up to 25,000 words of long-form text.
Its larger context supports analysis of substantial documents.
Multimodal inputs create new practical use cases.

“Some of the biggest differences is that it's multimodal, which means it can accept images.”
Kipp Bodnar · 02:00

“It can now generate up to 25,000 words, which is pretty crazy for long-form text.”
Kipp Bodnar · 02:30

#gpt-4#multimodal-ai#long-context#generative-ai

Explainer11:30

GPT-4 Jumped From the 10th to the 90th Bar-Exam Percentile

Kieran uses standardized-test performance to illustrate how quickly model capability advanced between ChatGPT and GPT-4. The large jump over only a few months is presented as evidence that businesses should not project current limitations far into the future.

ChatGPT launched only months before GPT-4.
GPT-3.5 reportedly scored around the 10th percentile on the bar exam.
GPT-4 reportedly reached the 90th percentile.
Some biology results reportedly reached the 99th percentile.

“ChatGPT was in the 10th percentile that passed the results, was in the 10th percentile. GPT-4 was in the 90th percentile.”
Kieran Flanagan · 12:00

“And then, on some other tests I looked at like biology, some sort of biology exam, it was in the 99th percentile.”
Kieran Flanagan · 12:00

#benchmarks#bar-exam#model-progress#gpt-4

Explainer13:30

Use Long Context to Apply Expert Knowledge to New Work

The hosts argue that a 25,000-word context window allows users to supply substantial specialist material and then ask the model to reason with it. A marketer could provide a body of copywriting work, apply its lessons to a sales page, and request explanations for every suggested change.

Long context lets the model work across substantial source material.
Users can apply an expert's body of knowledge to a new task.
The model can explain what it changed and why.
This resembles loading temporary specialist knowledge into a workflow.

“So, it can hold 25,000 words and it can actually manipulate them, reason with them, use them somewhat similar to a really top grade Stanford…”
Kieran Flanagan · 14:30

#context-window#copywriting#knowledge-work#reasoning

Explainer18:00

AI Could Replace Static Courses With Personalized Learning

Long-context and multimodal models can turn source material into structured lessons, potentially delivered as text, audio, or video. The Khan Academy GPT-4 tutor is presented as an early example of education shifting from fixed course packages toward interactive, personalized instruction.

AI can synthesize source material into a multiweek course.
Multimodal systems may deliver lessons in several media formats.
Interactive tutors can adapt learning through conversation.
Course creators may face pressure from on-demand personalized curricula.

“I can give it to AI and I can say, build me a six-week course.”
Kieran Flanagan · 18:30

“The other one that's I think, very transformative is you have all of these course creators.”
Kieran Flanagan · 18:30

#education#ai-tutors#course-creation#personalization

Explainer25:00

Adept Wants AI to Operate Existing Software for You

Adept is described as applying AI to software actions rather than only generating text or media. A user states an objective in natural language, and the system executes the necessary actions across software, potentially moving conventional interfaces and applications into the background.

Adept raised substantial funding during a banking crisis.
The product aims to execute software tasks from natural-language instructions.
Applications may become backend functionality behind an AI control layer.
The interface shift could matter more than any single automated task.

“And they're trying to fundamentally change how we use software.”
Kipp Bodnar · 26:00

“it just matters in the functionality, 'cause I use it through Adept.”
Kieran Flanagan · 27:00

#adept#software-agents#automation#user-interface

Story· 1

Story21:00

AI Job Disruption Raises a Deeper Question About Human Purpose

Kieran contrasts Sam Altman's earlier optimism about workers moving into better jobs with a more cautious recent tone about engineers and other knowledge workers. The concern is not only lost income but also whether people can retain meaning, usefulness, and purpose when machines perform more of their work.

AI disruption is reaching knowledge work earlier than some expected.
Past automation narratives assumed displaced workers would find better jobs.
Abundant leisure does not automatically produce fulfillment.
Policy and business debates must account for meaning as well as employment.

“Humans might have to find other things to do with their time.”
Kieran Flanagan · 22:00

“Humans wanna feel a sense of purpose.”
Kipp Bodnar · 22:00

#future-of-work#purpose#automation#job-displacement

Tool· 2

Tool03:00

Turn a Month of Meeting Notes Into an Automated Weekly Brief

Kieran describes combining daily meeting notes into one document, asking GPT-4 to summarize them, and extracting forgotten follow-up actions. He proposes automating the flow from Google Docs through OpenAI to email so each week ends with a concise action brief.

Combine dispersed meeting notes into one source document.
Ask AI to summarize discussions and extract follow-up actions.
Review outputs because the model may invent unnecessary tasks.
Automate delivery of the resulting brief by email.

“And I asked it to summarize all of my meeting notes, and to give all of the follow up points.”
Kieran Flanagan · 03:30

“And every week, I'll just get a summary of my most important notes and the follow up points from those meetings.”
Kieran Flanagan · 04:00

#meeting-notes#automation#productivity#follow-ups

Tool27:30

Google's AI Rollout Puts Automation Inside Daily Workflows

Google's announcements were less visually surprising than GPT-4's launch but potentially important because they place AI inside Gmail, Docs, and Sheets. Summarization, drafting, and workflow automation become available where large numbers of people already work.

Gmail can use AI for summaries and follow-up drafting.
Docs gains integrated writing assistance.
Sheets can absorb capabilities previously supplied by plugins.
Distribution through existing products can accelerate mainstream adoption.

“But it's getting in everybody's hands.”
Kipp Bodnar · 27:00

“If I can automate all of my email, that's pretty game-changing.”
Kieran Flanagan · 28:00

#google-workspace#gmail#productivity#automation

Takeaway· 1

Takeaway10:00

AI Shrinks the Distance Between an Idea and a Finished Product

The ability to generate code and creative assets dramatically compresses production cycles. The hosts argue that teams built around slow handoffs and conventional production stages risk being overtaken by teams that use AI for rapid iteration.

Creation cycles are shrinking by an order of magnitude.
AI can remove several handoffs between ideation and production.
Faster production enables much faster experimentation.
Teams that resist new tools risk being out-innovated.

“The ability to go from idea to product or marketing strategy, what have you, is gone. It's shrunk by an order of magnitude.”
Kipp Bodnar · 11:00

“And so, if you are not set up as an agile team that embraces technology, you're going to get out-innovated, I think.”
Kipp Bodnar · 11:30

#innovation#iteration#agile-teams#product-development