Ben Tossell06 December 2022

This OpenAI Update Is Going To Replace Google Search with Ben Tossell

1Frameworks

12Insights

Frameworks in this episode

Productivity4 steps

The Human Sandwich

Set the direction, let AI execute, then refine the result.

Read →

Insights & moments

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

Myth Buster· 1

Myth Buster31:00

Alexa May Be an AI Distribution Network, Not a Failed Product

Kipp rejects the framing that Amazon failed with Alexa, arguing that its installed devices form a valuable network of voice-access points. A stronger language-model backend could transform that existing hardware footprint without Amazon rebuilding distribution from scratch.

Alexa's installed base is strategically valuable.
The assistant backend can be upgraded independently of the devices.
Voice hardware provides immediate AI distribution.
Current product limitations do not erase infrastructure value.

“They have created the world's best network of AI access points.”
Kipp Bodnar · 31:00

“they just have the perfect access point for, you know, kind of verbal AI that exists.”
Kipp Bodnar · 31:30

#alexa#amazon#voice ai#distribution

Hot Take· 4

Hot Take08:30

Answer Engines Could Break the Publisher Payback Model

The hosts contrast Google's exchange of traffic for publisher content with an AI system that learns from content but may return answers without visits, attribution, or brand exposure. They question why publishers would continue producing source material if an answer engine captures the value without providing a meaningful return.

Google sends publishers traffic in exchange for indexable content.
AI answers can eliminate the need to visit source pages.
Publishers may lose traffic and brand visibility.
Copyright and sustainable monetization remain unresolved.

“We give you our content, you learn from that, give us the answers.”
Kieran Flanagan · 08:30

“But if there's no payback for the publishers, then why even do it?”
Kieran Flanagan · 11:30

#publishers#search#copyright#monetization

Hot Take17:00

The Future of Software May Be an Audience of One

Ben predicts that people will generate narrowly tailored internal tools rather than search for, subscribe to, and configure generic SaaS products. In this vision, a user describes a precise need and receives software built for that person and use case alone.

Natural language could replace much of custom internal-tool development.
Users may build small utilities instead of buying multiple subscriptions.
Generated software can target one exact workflow.
Deployment remains a current barrier but may also become automated.

“I can fully imagine a future where all internal tools are built by yourself for yourself.”
Ben Tossell · 17:00

“you can install that internal tool that's literally an audience of one, built by yourself for the exact use case that you have.”
Ben Tossell · 18:00

#personal software#internal tools#saas#text-to-code

Hot Take25:30

Perfect Personalization Could Make People Stop Changing

Kipp warns that an AI which continually supplies familiar, preferred choices may reduce exposure to novelty and slow personal growth. Kieran adds that discovery supports learning, experimentation, and a healthy business ecosystem in which unfamiliar products can still emerge.

Preference learning can become a reinforcing loop.
Discovery helps people learn and try unfamiliar things.
Single-answer recommendations can narrow choice.
Over-personalization may harm cultural and commercial diversity.

“The thing I am most scared of for everybody watching is, that AI slows down human change.”
Kipp Bodnar · 25:30

“I think discovery for humans is an important aspect of like learning, trying new things”
Kieran Flanagan · 26:00

#personalization#discovery#society#recommendations

Hot Take30:00

AI-Generated Content May Shift Trust From Brands to People

Ben predicts that uncertainty about whether images and articles are machine-generated will increase the value of identifiable human authorship. Personal relationships and explicit accountability may become stronger trust signals than polished brand content.

Audiences may question whether brand content was AI-generated.
Known individuals can build trust through accountable authorship.
AI becomes a production tool rather than proof of credibility.
Human identity may differentiate content in a synthetic media environment.

“I think that people are gonna trust people over brands.”
Ben Tossell · 30:00

“AI is just a tool in everyone's tool belt that they should be excited about.”
Ben Tossell · 30:30

#trust#brands#ai content#authorship

Explainer· 3

Explainer02:00

How DaVinci-003 Updated Every GPT-3 App Overnight

Ben explains that DaVinci-003 improved the model beneath products such as Copy.ai and Jasper without requiring each product to reinvent its application. The release made responses better and prompting easier, effectively distributing a software upgrade across the GPT-3 ecosystem.

DaVinci-003 improved on DaVinci-002.
Products built on GPT-3 inherited better model behavior.
The update improved responses and simplified prompting.
DaVinci-003 was not GPT-4.

“essentially every app that was built on GPT-3 got a software update overnight.”
Ben Tossell · 02:00

“DaVinci-003 is not GPT-4”
Ben Tossell · 03:30

#gpt-3#openai#ai models

Explainer05:00

Why Conversation Changes the Search Experience

ChatGPT retains context and lets users progressively refine a request instead of restarting searches and assembling information manually. The speakers argue that this shifts the work of finding, collating, and rewriting information from the user to the AI.

Conversational context supports follow-up questions.
Users can refine one result instead of launching repeated searches.
The AI can change format, scope, and content in place.
Traditional search leaves synthesis work to the user.

“Chat GPT can remember what's happened in the thread, so you can just have a continued conversation that then builds on the last piece.”
Ben Tossell · 05:30

“The work doesn't feel like it's on you, whereas if you're using Google, the work is on you to go find the stuff, collate it,…”
Ben Tossell · 07:30

#chatgpt#search#user experience

Explainer20:00

Websites Could Collapse Into a Conversation

The speakers imagine websites becoming simple conversational interfaces that answer pricing, suitability, and setup questions directly. They extend the idea to a consolidated buying journey in which AI researches options, recommends solutions, and potentially completes setup without requiring conventional browsing.

A search or chat field could replace much website navigation.
AI could combine research, selection, purchase, and setup.
Conventional product pages may become less important.
Recommendation quality remains limited without trustworthy reviews and criteria.

“there's no need for scrolling, it's just like an input field, like a search box on every site”
Ben Tossell · 21:00

“maybe there's a day where AI will do all of that for you.”
Kipp Bodnar · 21:30

#web design#commerce#conversational ui#recommendations

Tool· 1

Tool12:00

Can a Non-Coder Build an App With ChatGPT and Replit?

Ben describes a non-coder generating an application with ChatGPT, pasting the code into Replit, and returning errors to the chat for diagnosis. The conversation presents natural-language coding as a way to lower the barrier between an idea and a working product, although deployment and code handling still require effort.

ChatGPT can generate application code from plain-English requests.
Replit provides an in-browser environment for running copied code.
Users can paste errors back into the conversation for fixes.
Natural-language iteration makes development more accessible to non-coders.

“You put that in the conversation saying "I'm getting this bug, what do I need to do?" And then it says, "Oh, you need to…”
Ben Tossell · 13:30

“can you build a website purely using Chat GPT and Replit, not knowing any code, and literally typing in English, copying and pasting, can you…”
Ben Tossell · 14:00

#coding#chatgpt#replit#no-code

Takeaway· 3

Takeaway18:30

AI Tools Fail When They Sit Outside the Flow of Work

Discussing Adept AI, Ben argues that an assistant must be available where the user is already working. Requiring people to leave their task, open another app, or navigate a menu adds enough friction to undermine an otherwise capable tool.

Effective assistants should appear inside the active workflow.
Context switching reduces practical usefulness.
Instruction is less valuable than direct execution.
Browser and software overlays can remove interface-learning barriers.

“it needs to be in front of you, it needs to be where you are at the time”
Ben Tossell · 19:00

“if you can tell me exactly what to do and how to do it, I just wanna say, "Okay, do it then."”
Ben Tossell · 19:30

#ai agents#workflow#adept ai#automation

Takeaway28:30

When AI Products Look Alike, Distribution Becomes the Moat

Ben recounts independently imagining a personalized children's storybook only to see several nearly identical products appear within days. The hosts conclude that low build times and weak backend differentiation make distribution and go-to-market expertise decisive advantages for AI startups.

AI enables many teams to build the same concept quickly.
Backend technology may offer little lasting differentiation.
Speed alone does not guarantee a durable advantage.
Distribution and go-to-market knowledge become primary competitive moats.

“the next two days, I saw four products that were exactly that.”
Ben Tossell · 29:00

“distribution go-to market knowledge is really high.”
Kipp Bodnar · 29:30

#distribution#go-to-market#ai startups#competition

Takeaway32:00

AI Shrinks the Distance Between Imagination and Creation

The episode closes optimistically, arguing that many people already possess creative ideas but lack the technical tools to express them. Faster access to code, text, and other production capabilities could allow more people to turn unusual ideas into tangible work.

Tooling is often the barrier between an idea and an artifact.
AI gives more people access to code and content production.
Faster execution can encourage experimentation.
The speakers expect ideas and art to improve rather than decline.

“it's just unlocking people's creativity, like there's so much creativity in people, and tooling is the problem.”
Kieran Flanagan · 32:30

“I think ideas, and art, and everything are gonna get better, not worse.”
Ben Tossell · 32:30

#creativity#democratization#building#art