Perplexity is Killing Custom GPTs Ep. 309

For more episodes, visit our website at www.thedailyai....
In today’s episode of the Daily AI Show, Brian, Andy, and Jyunmi, later joined by Karl, discuss the potential of Perplexity AI's API as a contender to custom GPTs, exploring whether Perplexity might be the next "custom GPT killer." The conversation covers the strengths and weaknesses of custom GPTs, the role of live search, and when businesses should consider Perplexity's API over the more familiar custom solutions.
Key Points Discussed:
Custom GPTs: Advantages and Use Cases
Ease of Setup and Speed: Custom GPTs are praised for their simplicity and speed in addressing specific business needs. Brian highlighted how quick it is to deploy a custom GPT for tasks like prospecting or sales support, which provides a real-time solution with minimal setup.
Cost Efficiency: With minimal ongoing costs for even highly customized solutions, custom GPTs are seen as an affordable way to boost business efficiency, especially in sales and customer research.
Flexibility and Limitations: While custom GPTs offer rapid results, they have limitations in control, especially for complex solutions. Variations in user experience and issues with memory management can lead to inconsistent outputs across different users.
Perplexity API: An Emerging Alternative
Strengths of Perplexity: Perplexity AI offers a real-time search API that combines large language models with web search capabilities. The discussion highlighted that while it introduces better control over search results and automation, it comes with trade-offs in speed and latency.
Customization Through API Tools: The group explored how tools like Make or Zapier can integrate Perplexity’s API for more fine-tuned control over outputs, offering flexibility in automating workflows where quality and precision are key but time isn't as critical.
Use Cases and Decision-Making
When to Use Custom GPTs vs Perplexity: Brian and the co-hosts discussed the balance between speed and reliability. Custom GPTs are ideal for rapid deployment and iterative testing, while Perplexity’s API may offer better quality control in scenarios where users need real-time, web-based data but can tolerate slower response times.
Real-Time Search Capabilities: One of the limitations of custom GPTs is their reliance on pre-trained data. The integration of Perplexity’s API allows users to pull in real-time information, but with slower response times compared to GPTs, making it suitable for use cases that prioritize accuracy over speed.
The episode closes with a reflection on the future of AI-powered automation, noting that while custom GPTs remain a solid choice for many applications, Perplexity offers compelling advantages for businesses seeking enhanced control and live data. The co-hosts plan to dive into Google’s Notebook LM in the next episode, offering more insights into AI tools and their growing role in the business world.
#AI #PerplexityAI #CustomGPT #AutomationTools #AIinBusiness
00:00:25 Introduction and Topic: Perplexity API vs. Custom GPTs
00:01:45 Benefits of Custom GPTs: Speed, Cost, and Ease of Use
00:05:35 Personal Experiences with Building and Using Custom GPTs
00:08:53 Live Search Capabilities and Limitations of Custom GPTs
00:11:24 The Role of the OpenAI Assistant API
00:12:05 Experiences with Custom GPTs and Data Integration
00:14:59 Perplexity API as a Potential Alternative
00:18:00 Challenges with Controlling Custom GPTs and User Experience
00:20:46 The Need for Flexibility and Adaptability in AI Solutions
00:23:26 Google's Lack of a Comprehensive Web Search API
00:25:15 Use Case Considerations and Expectation Management
00:26:36 Understanding the Needs and Preferences of End Users
00:29:40 Individualized AI Use and Adoption in Organizations
00:31:20 The Importance of Setting Expectations with AI
00:33:51 Addressing Variability in AI Responses and Client Complaints
00:38:47 Understanding Deviations in Custom GPTs and Prompting Strategies
00:40:04 The Importance of User Education and Troubleshooting
00:41:38 Integrating Reasoning Models and Future Enhancements
00:42:56 Choosing the Right AI Tool for Specific Needs
00:45:24 Conclusion and Show Announcements

[@DaKingof]

🎙 (00:00) Discussion begins with how custom GPTs assist in information collection and processing.
💡 (00:37) The topic of the day is introduced: whether perplexity could be a "custom GPT killer" and how it could replace custom GPTs in automation systems like Make or Zapier.
⚙ (01:38) Overview of custom GPTs' ease of use and how OpenAI made them easy to set up.
🔍 (03:02) Mention of live search capabilities in custom GPTs, which pull live information and integrate it with the language model and user-uploaded documents.
🏢 (03:33) Examples of custom GPT use cases in sales, such as prospecting and discovery call preparation, which benefit from live information and company-specific content.
📧 (04:35) Explanation of how custom GPTs use a combination of live data, company documentation, and language models for tasks like email writing.
🚀 (05:07) Discussion on how custom GPTs have been valuable for solving real-time problems in business settings, even if not perfect.
🔁 (06:11) Benefits of using custom GPTs for repetitive tasks, such as generating summaries and hashtags from transcripts.
🔧 (07:30) Custom GPTs are useful for niche cases, like development projects with specific tool combinations.
⚠ (08:08) Transition into the limitations of custom GPTs, like not being perfect and reasons for considering alternatives like perplexity API.
🗣 (08:34) Carl’s and Andy’s input on their experience with custom GPTs, including use cases where live search is needed and when it’s better to build custom digital assistants.
📂 (10:22) Mention of integrating GPTs with other tools like Google Docs or Teams for more customized solutions.
⚡ (11:34) Explanation of how perplexity's API might offer faster access to live data but with trade-offs like lower speed.
🛠 (12:42) More input from Andy on using pre-built custom GPTs and how easy OpenAI made it to create and customize them.
📊 (14:08) Differences between perplexity and custom GPTs, noting that perplexity offers flexibility for automation but may sacrifice speed and usability.
⏱ (15:43) Detailed explanation of why custom GPTs are valuable: speed, low latency, ease of use, and cost efficiency.
🤔 (17:53) Custom GPT limitations, like variability in responses, are discussed, and how perplexity might solve consistency issues.
🔄 (19:35) Discussion about how custom GPTs sometimes deviate from the intended task due to inherent variability in how they interpret prompts.
🤖 (22:47) Example of using perplexity API with automation tools like Make to manage complex tasks and how breaking up tasks can improve reliability.
🔀 (23:57) Flexibility of using multiple tools (like perplexity) is mentioned, along with the potential trade-offs between speed and consistency.
🛑 (24:38) Note that Google doesn’t provide an API for raw web search results, making perplexity an interesting alternative.
🐢 (25:50) Carl discusses how perplexity’s slow speed could be problematic for some use cases.
⚖ (28:16) Emphasis on how building AI solutions depends heavily on the specific use case and user expectations.
🎭 (31:05) Discussion about the unpredictability of AI responses and the need for setting realistic user expectations when deploying AI tools.
🚫 (35:20) Examples where custom GPTs produced unsatisfactory results and the difficulty in controlling AI responses are discussed.
🧠 (40:32) Carl points out that GPTs don’t reason through tasks the way humans expect, suggesting future models like GPT-5 could improve this.
🤖🔍 (41:39) Mention of Claude’s reasoning capabilities and how it compares to custom GPTs for specific tasks.
🔚 (42:34) Closing thoughts on balancing the use of various AI tools depending on the need for speed or consistency.
🌪 (47:47) End of the show, with a mention of hurricane Milton and an upcoming episode on Google’s NotebookLM.

[@QuizmasterLaw]

Just here to remind my brothers in arms: this is WHY we do NOT use llm ai for targeting decisions!

[@justinduveen3815]

Perplexity API is wonderful for real time info! Some of their bigger models take a while to answer so that can cause automation problems with some no / low code players.
Using them with python works great.

[@micbab-vg2mu]

API is quite good:)

[@echabbewal]

Please keep discussion to the point in future videos. I just got lost

[@avalagum7957]

You use AI to write emails to your company's customers? You trust AI that much?