For programming prior to GPT4, I had a hierarchical tree of code snippet for tasks I had done that did not have an immediate google/stackoverflow solution. Now with GPT4, I can't remember the last time I have made new notes. There is DEFINITELY a complexity limit for GPT4 and not just exceeding 80-150 lines. This limit is often prematurely hit when writing complex algorithms. Because of this, I simply abandon small projects it does poorly, in favor of exploring different ideas lol
Hey there! I’ve been listening on Spotify and really enjoying the episode, but I had to come here to make a comment. I truly feel whoever is doing the editing is way way WAY too heavy handed with cutting out… whatever they’re cutting out, I’m assuming it’s “ums” and “ahs” and breaths. But I just wanted to toss a vote in and say: leave some of them in, the choppiness of this episode really started to get to me, so there is atleast one listener out there who would have preferred the audio to more closely resemble the original conversation, and I’m a pretty average person so maybe I’m not the only one who would feel this way. Having said that this is the first episode I’ve listened to/watched and I’m really enjoying it and look forward to consuming more of your content!
This was the first podcast of yours that I listened to and I came to the video to check out the visuals. Really engaging and I learned tips I will use in my career!
Wow!! Illumanating content! I particularaly enjoyed the tips from Nathan about installing & configuring an app by prompting ChatGPT how to do that for you. This accelaretes tremendously the development process as you don't need to search on google or YT how to do that! Great job guys. Keep it up!
While Analysis before answer works for getting more accurate answers, often Answer Before Analysis is better for instruction following. If you need the AI to complete a form in a structured way in order to feed that data into an API, giving it a place to put its overall reasoning seems to make it follow the instructions more directly. It depends on whether you want accurate information, or creative content. Most of the time when I'm using this approach, the reason is just throwaway data anyway. But you have to give the AI a place to put the reason data, or they will just go off script. Think of it like a 'first thing that comes to your mind' vs. 'really sit down and think about it.' Both approaches have their place, it just depends on your application.
🎯 Key Takeaways for quick navigation: 00:00 🧠 *The potential of AI in offloading cognitive work,* - AI can be a valuable co-pilot, helping individuals with specific tasks in real-time. - Delegation mode can offload repetitive and time-consuming cognitive tasks, saving time and improving productivity. - The balance between the benefits and risks of AI in offloading cognitive work, with the need for responsible use. 10:13 🤖 *AI's role in summarizing and streamlining creative work,* - AI can assist in summarizing information, making it more efficient for creative tasks like writing and programming. - The realization of how much drudgery exists in creative knowledge work and the role of AI in reducing it. - AI's ability to help find the right words, metaphors, and expressions to convey ideas effectively. 17:34 💡 *Leveraging AI as a co-pilot in writing and content creation,* - AI can assist in various stages of content creation, from summarizing ideas and creating outlines to refining expressions. - Examples of using AI to generate summaries and explanations of complex topics. - The value of AI in helping writers find the right structure and expressions for their content. 20:34 🧠 *Discussing the use of ChatGPT for writing assistance* - ChatGPT assists in microtasks and helps refine writing. - Using ChatGPT for verbal iteration and feedback on drafts. 23:51 🤝 *Rod's role as an AI advisor at Athena and the development of Athena chat* - Athena aims to provide virtual assistants for executives. - Building prototype demos to integrate AI into the assistant's workflow. - Developing a chat interface for AI-driven assistance. 26:06 💻 *Learning React and understanding app structure* - Rod's need to understand the structure of a React app. - ChatGPT guiding Rod through the app's structure. - Rod's approach to seeking guidance on unfamiliar coding frameworks. 30:54 🔑 *The importance of prompting techniques and Chain of Thought* - Explaining the significance of prompting techniques. - Avoiding prompting that leads to direct answers, allowing models to think through problems. - The evolution of AI prompting strategies. 37:13 🛠️ *Solving nested programming problems with ChatGPT* - Rod encountering multiple levels of coding challenges. - ChatGPT assisting Rod step by step to resolve issues. - The value of AI in overcoming complex programming hurdles. 40:59 🤖 *ChatGPT assisting in building an app with React* - Using ChatGPT to help with React development. - ChatGPT saves time by generating React code and providing guidance. 45:47 🧠 *ChatGPT's potential for episodic relationships* - Discussing the concept of episodic interactions with ChatGPT. - Envisioning a future where ChatGPT retains context and history across sessions. 49:03 🧙♂️ *Custom instructions for ChatGPT* - The importance of custom instructions in ChatGPT. - How custom instructions can enhance the interaction with ChatGPT. 54:58 📊 *Creating diagrams for a patent application* - Using ChatGPT to assist in creating diagrams for a patent application. - The iterative process of refining diagrams with ChatGPT's help. 59:05 🎨 *ChatGPT and diagram generation* - Exploring the potential of ChatGPT to generate and edit diagrams. - Comparing ChatGPT's capabilities to specialized diagram tools. 01:00:44 🧠 *Using ChatGPT and custom editors for text generation.* - ChatGPT can be used with custom editors to create tailored text generation experiences. - Developers might eventually build their own renderers within ChatGPT for specialized tasks. - ChatGPT's versatility allows for various applications beyond just text generation. 01:01:26 🚗 *Car shopping and the challenge of finding the right vehicle.* - Car shopping, especially for used cars, can be a complex and time-consuming process. - Understanding car trim levels and features can be challenging for buyers. - Perplexity, a tool for answering questions, can help gather information about different car models and trim levels. 01:03:17 🚑 *Using AI in medical diagnosis and the potential of AI in healthcare.* - AI is making significant strides in medical diagnosis and differential diagnosis. - AI models, such as language models, are becoming increasingly accurate in diagnosing medical conditions. - Access to AI-driven medical expertise could lead to greater equality of healthcare access. 01:08:34 🌐 *The future of AI and the evolving landscape of AI architectures.* - AI is advancing rapidly, and there's a belief in the potential for AI to achieve remarkable results. - AI models are becoming more specialized and capable in various domains. - The future of AI may involve a mixture of different AI architectures and models, leading to continuous advancements. Made with HARPA AI
1:13:00 Many MDs today are the worst imitators, frankly, their results/actions are little more than "in-distribution" knee-jerk responses. Simply because 1) they get to process more patients, and 2) it limits legal risk - doing the "standard" thing preserves their job by protecting them from lawsuits. This is really horrible for patients, because the really bad MDs ignore priors that would lead to very different diagnosis distributions. I'm going on a limb here and will call this blind spot "prior blindness," and cherry picking (for monetary optimization), as the two major faults with US healthcare today.
The internet itself was not meant to go global. The intention was to keep it for the elites only, with a very watered down version made available for everyone else. But it didnt go the way they planned. They had to adapt. And they did. I think a similar thing will happen with AI. They wont be able to stop smaller models (that can fit on any phone) from becoming more and more capable... They wont be able to restrict access unless they take away a lot of freedoms and rights from everyone. I'm gonna bet they wont be able to pull it off.. But i know theyll try (they already are).
Those interests dont move fast enough😂..They will get smoked and any notion of regulation is stupid. AI is out the bag..Open source is already has shyt coming out everyday,no stopping this train
For programming prior to GPT4, I had a hierarchical tree of code snippet for tasks I had done that did not have an immediate google/stackoverflow solution. Now with GPT4, I can't remember the last time I have made new notes. There is DEFINITELY a complexity limit for GPT4 and not just exceeding 80-150 lines. This limit is often prematurely hit when writing complex algorithms. Because of this, I simply abandon small projects it does poorly, in favor of exploring different ideas lol
Hey there! I’ve been listening on Spotify and really enjoying the episode, but I had to come here to make a comment.
I truly feel whoever is doing the editing is way way WAY too heavy handed with cutting out… whatever they’re cutting out, I’m assuming it’s “ums” and “ahs” and breaths. But I just wanted to toss a vote in and say: leave some of them in, the choppiness of this episode really started to get to me, so there is atleast one listener out there who would have preferred the audio to more closely resemble the original conversation, and I’m a pretty average person so maybe I’m not the only one who would feel this way.
Having said that this is the first episode I’ve listened to/watched and I’m really enjoying it and look forward to consuming more of your content!
This was the first podcast of yours that I listened to and I came to the video to check out the visuals. Really engaging and I learned tips I will use in my career!
Wow!! Illumanating content! I particularaly enjoyed the tips from Nathan about installing & configuring an app by prompting ChatGPT how to do that for you. This accelaretes tremendously the development process as you don't need to search on google or YT how to do that! Great job guys. Keep it up!
While Analysis before answer works for getting more accurate answers, often Answer Before Analysis is better for instruction following. If you need the AI to complete a form in a structured way in order to feed that data into an API, giving it a place to put its overall reasoning seems to make it follow the instructions more directly. It depends on whether you want accurate information, or creative content. Most of the time when I'm using this approach, the reason is just throwaway data anyway. But you have to give the AI a place to put the reason data, or they will just go off script. Think of it like a 'first thing that comes to your mind' vs. 'really sit down and think about it.' Both approaches have their place, it just depends on your application.
🎯 Key Takeaways for quick navigation:
00:00 🧠 *The potential of AI in offloading cognitive work,*
- AI can be a valuable co-pilot, helping individuals with specific tasks in real-time.
- Delegation mode can offload repetitive and time-consuming cognitive tasks, saving time and improving productivity.
- The balance between the benefits and risks of AI in offloading cognitive work, with the need for responsible use.
10:13 🤖 *AI's role in summarizing and streamlining creative work,*
- AI can assist in summarizing information, making it more efficient for creative tasks like writing and programming.
- The realization of how much drudgery exists in creative knowledge work and the role of AI in reducing it.
- AI's ability to help find the right words, metaphors, and expressions to convey ideas effectively.
17:34 💡 *Leveraging AI as a co-pilot in writing and content creation,*
- AI can assist in various stages of content creation, from summarizing ideas and creating outlines to refining expressions.
- Examples of using AI to generate summaries and explanations of complex topics.
- The value of AI in helping writers find the right structure and expressions for their content.
20:34 🧠 *Discussing the use of ChatGPT for writing assistance*
- ChatGPT assists in microtasks and helps refine writing.
- Using ChatGPT for verbal iteration and feedback on drafts.
23:51 🤝 *Rod's role as an AI advisor at Athena and the development of Athena chat*
- Athena aims to provide virtual assistants for executives.
- Building prototype demos to integrate AI into the assistant's workflow.
- Developing a chat interface for AI-driven assistance.
26:06 💻 *Learning React and understanding app structure*
- Rod's need to understand the structure of a React app.
- ChatGPT guiding Rod through the app's structure.
- Rod's approach to seeking guidance on unfamiliar coding frameworks.
30:54 🔑 *The importance of prompting techniques and Chain of Thought*
- Explaining the significance of prompting techniques.
- Avoiding prompting that leads to direct answers, allowing models to think through problems.
- The evolution of AI prompting strategies.
37:13 🛠️ *Solving nested programming problems with ChatGPT*
- Rod encountering multiple levels of coding challenges.
- ChatGPT assisting Rod step by step to resolve issues.
- The value of AI in overcoming complex programming hurdles.
40:59 🤖 *ChatGPT assisting in building an app with React*
- Using ChatGPT to help with React development.
- ChatGPT saves time by generating React code and providing guidance.
45:47 🧠 *ChatGPT's potential for episodic relationships*
- Discussing the concept of episodic interactions with ChatGPT.
- Envisioning a future where ChatGPT retains context and history across sessions.
49:03 🧙♂️ *Custom instructions for ChatGPT*
- The importance of custom instructions in ChatGPT.
- How custom instructions can enhance the interaction with ChatGPT.
54:58 📊 *Creating diagrams for a patent application*
- Using ChatGPT to assist in creating diagrams for a patent application.
- The iterative process of refining diagrams with ChatGPT's help.
59:05 🎨 *ChatGPT and diagram generation*
- Exploring the potential of ChatGPT to generate and edit diagrams.
- Comparing ChatGPT's capabilities to specialized diagram tools.
01:00:44 🧠 *Using ChatGPT and custom editors for text generation.*
- ChatGPT can be used with custom editors to create tailored text generation experiences.
- Developers might eventually build their own renderers within ChatGPT for specialized tasks.
- ChatGPT's versatility allows for various applications beyond just text generation.
01:01:26 🚗 *Car shopping and the challenge of finding the right vehicle.*
- Car shopping, especially for used cars, can be a complex and time-consuming process.
- Understanding car trim levels and features can be challenging for buyers.
- Perplexity, a tool for answering questions, can help gather information about different car models and trim levels.
01:03:17 🚑 *Using AI in medical diagnosis and the potential of AI in healthcare.*
- AI is making significant strides in medical diagnosis and differential diagnosis.
- AI models, such as language models, are becoming increasingly accurate in diagnosing medical conditions.
- Access to AI-driven medical expertise could lead to greater equality of healthcare access.
01:08:34 🌐 *The future of AI and the evolving landscape of AI architectures.*
- AI is advancing rapidly, and there's a belief in the potential for AI to achieve remarkable results.
- AI models are becoming more specialized and capable in various domains.
- The future of AI may involve a mixture of different AI architectures and models, leading to continuous advancements.
Made with HARPA AI
Very interested to hear your thoughts on Mamba 👍
Nathan's emergency pod on Mamba is live now! ua-cam.com/video/X5F2X4tF9iM/v-deo.html
1:13:00 Many MDs today are the worst imitators, frankly, their results/actions are little more than "in-distribution" knee-jerk responses. Simply because 1) they get to process more patients, and 2) it limits legal risk - doing the "standard" thing preserves their job by protecting them from lawsuits. This is really horrible for patients, because the really bad MDs ignore priors that would lead to very different diagnosis distributions. I'm going on a limb here and will call this blind spot "prior blindness," and cherry picking (for monetary optimization), as the two major faults with US healthcare today.
shout out to Robert Pirsig.
I both love and hate this.
The vested interests will crush this because it's a threat. see AI regulation. Download as many unaligned models as you can while you can.
The internet itself was not meant to go global. The intention was to keep it for the elites only, with a very watered down version made available for everyone else. But it didnt go the way they planned. They had to adapt. And they did. I think a similar thing will happen with AI. They wont be able to stop smaller models (that can fit on any phone) from becoming more and more capable... They wont be able to restrict access unless they take away a lot of freedoms and rights from everyone. I'm gonna bet they wont be able to pull it off.. But i know theyll try (they already are).
What would you suggest as of today?
Those interests dont move fast enough😂..They will get smoked and any notion of regulation is stupid. AI is out the bag..Open source is already has shyt coming out everyday,no stopping this train
Those interests cannot move fast enough they don't have a hope in hell they are up against millions of independent developers in foss communities
_Download as many unaligned models as you can while you can."_
What are the best ones? what would you recommend?