17 September 2024

Chat GPT-o1 is Mindblowing!! Everything You Need To Know

1Frameworks

9Insights

Frameworks in this episode

Productivity4 steps

Four Rules for Prompting Reasoning Models

Give reasoning models clear structure without making them overthink

Read →

Insights & moments

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

Hot Take· 3

Hot Take04:00

A PhD-Level Expert in Your Pocket Changes Knowledge Work

The hosts frame o1's reported performance in science and mathematics as inexpensive, always-available access to advanced expertise. Their larger point is that capabilities once limited by the availability and cost of specialists are becoming accessible to ordinary users.

Advanced scientific and mathematical help is becoming widely accessible
AI expertise can be available around the clock
Low marginal cost changes who can attempt specialist tasks
The hosts see the accessibility itself as a historic shift

“you basically now have a PhD level scientific student in your pocket”
Kieran Flanagan · 05:00

“247 $20 a month whenever you needed to”
Kipp Bodnar · 05:30

#expertise#science#knowledge work#access

Hot Take11:30

Service as Software Could Displace Professional Services

The episode predicts a shift from software as a service toward service as software. In this view, technology will directly perform work previously sold through professional-services organizations, creating new software markets while pressuring established service providers.

Technology products will increasingly perform complete services
New software markets may replace parts of professional-services markets
Advanced reasoning expands the range of automatable service work
The shift changes the product being sold from a tool to an outcome

“the next decade of technology is about we're going from software as service to Service as software”
Kipp Bodnar · 11:30

“there'll be new software and Technology markets that displace Professional Service markets”
Kipp Bodnar · 12:00

#service as software#saas#professional services#automation

Hot Take17:30

Reasoning Models Accelerate the Democratization of Coding

The hosts describe 2024 as the year AI made coding broadly accessible. They anticipate follow-on implications such as disposable personal code and custom web applications built for narrow needs rather than distributed as conventional products.

AI allows more non-specialists to write code
Coding is presented as the standout application of reasoning models
Personal disposable code may become commonplace
Small custom web apps could replace some general-purpose workflows

“2024 the year of that coding got democratized and that AI is allowing everyone to write code”
Kipp Bodnar · 17:30

“personal disposable code personal web apps all of those things”
Kipp Bodnar · 18:00

#coding#democratization#personal software#web apps

Explainer· 2

Explainer01:30

Why OpenAI Reset Its Model Family With o1

OpenAI presented o1 as a new model family rather than another entry in the GPT-4 naming sequence. The hosts argue that this reset signals a shift from emphasizing multimodality toward models designed for stronger thinking and reasoning.

o1 begins a new OpenAI model family
The naming reset distinguishes it from GPT-4o
Reasoning is positioned as the defining capability

“this is why it's called 01 01 a new family of model specifically because this model has something that the other models it would argue…”
Kieran Flanagan · 02:00

#openai#o1#reasoning#ai models

Explainer02:30

The Coding Evaluation Where o1 Nearly Tripled GPT-4o

Cognition Labs evaluated o1 through its Devin coding agent and reported a large improvement over GPT-4o. The episode cites scores rising from 25.9% with GPT-4o to 51.8% with the o1 preview and 74.2% with the production model.

GPT-4o scored 25.9% in the cited evaluation
The o1 preview scored 51.8%
The production version reportedly scored 74.2%
The hosts interpret the increase as a major reasoning advance

“the Devon base with the 01 preview was 51.8% and with production 01 is 74.2%”
Kipp Bodnar · 03:00

“it didn't just double it tripled in its ability to reason from the old opena model to the new opena model”
Kipp Bodnar · 03:30

#coding#benchmarks#devin#o1

Story· 1

Story06:30

The Ice-Block Puzzle That Showed o1's Scientific Reasoning

The episode highlights an unusual test asking how to melt a large ice block on a freezing night using a hydraulic press, a small bottle of tea, and a spoon. The model selected the press and explained how pressure, mechanical energy, increased surface area, and heat transfer could contribute.

The puzzle combines physical constraints with deliberately odd tools
The model selected the hydraulic press
Its response connected pressure and mechanical energy to melting
The hosts use the demo to illustrate applied scientific reasoning

“the most effective choice is a very powerful hydraulic press”
Kipp Bodnar · 07:30

“press pressure induced melting conversion of that mechanical energy to thermal energy increasing the surface area and facilitating heat transfer”
Kipp Bodnar · 07:30

#physics#science#reasoning#demo

Takeaway· 3

Takeaway10:30

Complex Consulting Work Is Becoming Available on Demand

The hosts discuss using reasoning models for work commonly assigned to consultants, including merger-risk assessment and investment-project evaluation. They argue that individuals and companies now have access to an AI-enabled consultant capable of addressing larger, multi-step problems.

Merger-risk assessment is presented as a representative use case
Investment projects can be evaluated with advanced models
Reasoning models can stack and execute multiple analytical steps
Some professional-services work may become accessible without a large firm

“everyone has a kind of AI enabled consultant and they can work through really large uh complex projects”
Kieran Flanagan · 11:00

#consulting#mergers#investment#automation

Takeaway12:00

AI Moves Expensive Analysis to the Start of Decisions

When analysis becomes fast and inexpensive, it no longer needs to be reserved for a final candidate or late-stage decision. The hosts use company acquisitions as an example: teams can run rough analyses on five possibilities early rather than commissioning a deep review of only one at the end.

Human-time constraints traditionally delayed analysis
AI lowers the cost of evaluating multiple options
Earlier analysis can eliminate weak choices sooner
Broader comparison may improve final human decisions

“you can do it for all five companies you can do it much faster much cheaper much earlier on”
Kipp Bodnar · 12:30

“it's just going to really evolve how we work and I think lead us to making better decisions as humans”
Kipp Bodnar · 12:30

#decision making#analysis#acquisitions#productivity

Takeaway18:00

Ignoring AI Expertise Creates a Compounding Productivity Gap

Kieran argues that workers who actively use advanced models will become more productive and effective than comparable peers who do not. Because model capabilities are changing rapidly, delaying hands-on experimentation risks creating an increasingly meaningful disadvantage.

Reasoning models provide on-demand specialist assistance
Comparable workers may diverge based on whether they use AI
Rapid model improvement increases the cost of delayed adoption
Regular experimentation helps users understand practical capabilities

“you are just at a distinct disadvantage if you are not using a PhD level expert to help you with your tasks”
Kieran Flanagan · 18:30

“the quicker people understand what is happening the more productive and efficient they're going to be in their role”
Kieran Flanagan · 18:30

#ai adoption#productivity#career#competitive advantage