Why Small Language Models Are Going To Eat The World The Prompt Desk podcast

21:13

In this one year anniversary episode of the podcast, your show hosts look back on what they have learned and observed over the past year. They made the following observations: Reflection on a year of LLM development: The hosts discuss how the LLM ecosystem has rapidly evolved over the past year, moving from initial hype to a focus on practical applications with demonstrable ROI. Changes in prompt engineering: They note a shift from instruction-based prompting to example-based prompting, finding the latter more reliable. Impact on software development: LLMs have significantly increased productivity and are changing how software is architected and developed, making previously complex tasks much simpler and faster to implement. Shifting skill requirements: The hosts predict that soft skills will become more valuable for engineers, while specific programming language knowledge may become less crucial due to AI assistance. Evolving perspectives: Both hosts describe how their views on LLMs have changed, moving from initial skepticism to greater optimism about the technology's potential and impact. Future of the field: They emphasize the transformative nature of LLMs on the entire field of computer science and encourage newcomers to stick with it, predicting that those who do will become the experts shaping the future of technology. --- Continue listening to The Prompt Desk Podcast for everything LLM & GPT, Prompt Engineering, Generative AI, and LLM Security. Check out PromptDesk.ai for an open-source prompt management tool. Check out Brad’s AI Consultancy at bradleyarsenault.me Add Justin Macorin and Bradley Arsenault on LinkedIn. Hosted by Ausha. See ausha.co/privacy-policy for more information.…

The Prompt Desk

1
Validating Inputs with LLMs 23:06

42 weeks ago23:06

23:06

In this thought-provoking episode, AI experts Bradley Arsenault and Justin Macorin delve into the cutting-edge world of input validation using large language models (LLMs). The hosts explore how this innovative approach is transforming traditional data validation methods in software applications. The conversation begins with a look at conventional input validation techniques, primarily relying on regular expressions (RegEx) for basic text field checks. Justin then introduces the concept of using LLMs to validate more complex inputs, such as job titles or descriptions, which go beyond simple character-level validation. Throughout the episode, the hosts discuss the numerous benefits of LLM-based validation, including improved downstream results in AI systems, reduction in nonsensical inputs, and the potential to decrease reliance on customer support teams. They also tackle the implementation challenges, such as increased computation costs and the risk of rejecting valid inputs in edge cases. Bradley and Justin provide insights into the technical aspects of implementing LLM validation, discussing where in the software architecture to place these checks and how to handle user feedback. They also explore future improvements to the system, such as incorporating probability thresholds and enhancing error messages with helpful examples. The episode concludes with a balanced view of the trade-offs involved in adopting this new approach. While acknowledging the potential for user resistance and increased complexity, the hosts argue that the long-term benefits of cleaner data and reduced errors make LLM-based validation a promising frontier in software development. This episode offers valuable insights for developers, data scientists, and product managers looking to enhance their applications' data quality and user experience through advanced AI techniques. --- Continue listening to The Prompt Desk Podcast for everything LLM & GPT, Prompt Engineering, Generative AI, and LLM Security. Check out PromptDesk.ai for an open-source prompt management tool. Check out Brad’s AI Consultancy at bradleyarsenault.me Add Justin Macorin and Bradley Arsenault on LinkedIn. Hosted by Ausha. See ausha.co/privacy-policy for more information.…

The Prompt Desk

1
Why you can't automate everything with LLMs 18:24

43 weeks ago18:24

18:24

In this episode of Prompt Desk, hosts Bradley Arsenault and Justin Macorin dive into the evolving landscape of AI applications. They explore the shift from full automation dreams to the reality of human-AI collaboration, discussing why guided experiences are proving more effective than autonomous systems. The hosts share insights on designing AI interfaces, managing error rates, and the importance of human oversight in complex AI tasks. Whether you're an AI enthusiast, developer, or business leader, this episode offers valuable perspectives on harnessing the power of large language models while avoiding common pitfalls. Join Bradley and Justin as they unpack the challenges and opportunities in creating reliable, user-friendly AI applications that truly augment human capabilities. --- Continue listening to The Prompt Desk Podcast for everything LLM & GPT, Prompt Engineering, Generative AI, and LLM Security. Check out PromptDesk.ai for an open-source prompt management tool. Check out Brad’s AI Consultancy at bradleyarsenault.me Add Justin Macorin and Bradley Arsenault on LinkedIn. Please fill out our listener survey here to help us create a better podcast: https://docs.google.com/forms/d/e/1FAIpQLSfNjWlWyg8zROYmGX745a56AtagX_7cS16jyhjV2u_ebgc-tw/viewform?usp=sf_link Hosted by Ausha. See ausha.co/privacy-policy for more information.…

The Prompt Desk

1
Data Preparation Best Practices for Fine Tuning 20:26

44 weeks ago20:26

20:26

In this episode of The Prompt Desk podcast, hosts Bradley Arsenault and Justin Macorin dive deep into the world of fine-tuning large language models. They discuss: The evolution of data preparation techniques from traditional NLP to modern LLMs Strategies for creating high-quality datasets for fine-tuning The surprising effectiveness of small, well-curated datasets Best practices for aligning training data with production environments The importance of data quality and its impact on model performance Practical tips for engineers working on LLM fine-tuning projects Whether you're a seasoned AI practitioner or just getting started with large language models, this episode offers valuable insights into the critical process of data preparation and fine-tuning. Join Brad and Justin as they share their expertise and help you navigate the challenges of building effective AI systems. --- Continue listening to The Prompt Desk Podcast for everything LLM & GPT, Prompt Engineering, Generative AI, and LLM Security. Check out PromptDesk.ai for an open-source prompt management tool. Check out Brad’s AI Consultancy at bradleyarsenault.me Add Justin Macorin and Bradley Arsenault on LinkedIn. Hosted by Ausha. See ausha.co/privacy-policy for more information.…

The Prompt Desk

1
Multilingual Prompting 15:39

46 weeks ago15:39

15:39

Have you ever tried to run a prompt in another language and gotten completely different results? It is a sad truth that our LLM models continue to work best in English. It is something that will probably always remain true because of the dominance of the English language, which ensures that a disproportionate amount of the worlds data is found in English. In this episode of The Prompt Desk, your hosts Justin Macorin and Bradley Arsenault discuss their relatively brief experiences with trying to get prompts working well in other languages, and provide some tips and tricks based on what they have learned so far. --- Continue listening to The Prompt Desk Podcast for everything LLM & GPT, Prompt Engineering, Generative AI, and LLM Security. Check out PromptDesk.ai for an open-source prompt management tool. Check out Brad’s AI Consultancy at bradleyarsenault.me Add Justin Macorin and Bradley Arsenault on LinkedIn. Please fill out our listener survey here to help us create a better podcast: https://docs.google.com/forms/d/e/1FAIpQLSfNjWlWyg8zROYmGX745a56AtagX_7cS16jyhjV2u_ebgc-tw/viewform?usp=sf_link Hosted by Ausha. See ausha.co/privacy-policy for more information.…

The Prompt Desk

1
Safely Executing LLM Code 18:11

47 weeks ago18:11

18:11

In this episode, AI experts Bradley Arsenault and Justin Macon dive deep into the challenges and best practices for safely executing code generated by large language models in a production environment. They discuss key security considerations, containerization techniques, static/dynamic code analysis, and error handling - providing valuable insights for anyone looking to leverage the power of LLMs while mitigating the risks of abuse by AI hackers. --- Continue listening to The Prompt Desk Podcast for everything LLM & GPT, Prompt Engineering, Generative AI, and LLM Security. Check out PromptDesk.ai for an open-source prompt management tool. Check out Brad’s AI Consultancy at bradleyarsenault.me Add Justin Macorin and Bradley Arsenault on LinkedIn. Please fill out our listener survey here to help us create a better podcast: https://docs.google.com/forms/d/e/1FAIpQLSfNjWlWyg8zROYmGX745a56AtagX_7cS16jyhjV2u_ebgc-tw/viewform?usp=sf_link Hosted by Ausha. See ausha.co/privacy-policy for more information.…

The Prompt Desk

1
How to Rescue AI Innovation at Big Companies 19:20

48 weeks ago19:20

19:20

In this episode, we dive into how companies can compete and innovate in the age of AI. We explore the limitations of simple "Generate with AI" buttons and discuss the need for more radical changes in AI implementation. Drawing from Clayton Christensen's work on disruptive innovation, we examine why large organizations struggle with adopting transformative technologies and suggest strategies for both big and small companies to stay competitive. We also touch on the challenges of transitioning traditional business models to AI-driven approaches, the importance of organizational agility, and the potential of modular organizational structures. Whether you're part of a Fortune 500 company or a nimble startup, this episode offers insights on navigating the AI revolution and maintaining your competitive edge. --- Continue listening to The Prompt Desk Podcast for everything LLM & GPT, Prompt Engineering, Generative AI, and LLM Security. Check out PromptDesk.ai for an open-source prompt management tool. Check out Brad’s AI Consultancy at bradleyarsenault.me Add Justin Macorin and Bradley Arsenault on LinkedIn. Please fill out our listener survey here to help us create a better podcast: https://docs.google.com/forms/d/e/1FAIpQLSfNjWlWyg8zROYmGX745a56AtagX_7cS16jyhjV2u_ebgc-tw/viewform?usp=sf_link Hosted by Ausha. See ausha.co/privacy-policy for more information.…

The Prompt Desk

1
How UX Will Change With Integrated Advice 17:22

49 weeks ago17:22

17:22

Are you tired of hearing about agents? Chat-bot this, agent that. It's the only way to apply AI that anyone talks about. But are agents the only way to use LLM technology? Is everything just converging towards conversational interface? In this episode of The Prompt Desk, your show hosts discuss a different way to use cutting edge AI in user-interface design: integrated advice. Using LLMs, we can provide AI driven tips, tricks, hints, and feedback on the open-ended text-boxes in our user interfaces. In everything from writing job descriptions and product descriptions, through to social media posts or comments - user interfaces are full of open-ended textbox inputs that are just dying to receive an 21st century makeover. Download the episode to hear more! --- Continue listening to The Prompt Desk Podcast for everything LLM & GPT, Prompt Engineering, Generative AI, and LLM Security. Check out PromptDesk.ai for an open-source prompt management tool. Check out Brad’s AI Consultancy at bradleyarsenault.me Add Justin Macorin and Bradley Arsenault on LinkedIn. Please fill out our listener survey here to help us create a better podcast: https://docs.google.com/forms/d/e/1FAIpQLSfNjWlWyg8zROYmGX745a56AtagX_7cS16jyhjV2u_ebgc-tw/viewform?usp=sf_link Hosted by Ausha. See ausha.co/privacy-policy for more information.…

The Prompt Desk

1
Prompting in Tool Results 19:00

50 weeks ago19:00

19:00

If you are using systems prompts, chat completions, and tool completions with OpenAI, you might find it challenging to get the model to follow your prompts. If you are like the show hosts, you often find that there are certain instructions that your bot just refuses to listen to! In the latest episode of The Prompt Desk, show host Bradley Arsenault shares with Justin Macorin the latest technique he has been using to improve the reliability of behaviours that the bot refuses to comply with. That new technique is to add additional instructions in prompts mid-conversation using tool results. Instead of seeing a tool result as simply containing data, you should rethink your tool results as combining both data and a prompt on what the bot should do with that data. With this technique, your show hosts have improved the reliability of their bots significantly. --- Continue listening to The Prompt Desk Podcast for everything LLM & GPT, Prompt Engineering, Generative AI, and LLM Security. Check out PromptDesk.ai for an open-source prompt management tool. Check out Brad’s AI Consultancy at bradleyarsenault.me Add Justin Macorin and Bradley Arsenault on LinkedIn. Please fill out our listener survey here to help us create a better podcast: https://docs.google.com/forms/d/e/1FAIpQLSfNjWlWyg8zROYmGX745a56AtagX_7cS16jyhjV2u_ebgc-tw/viewform?usp=sf_link Hosted by Ausha. See ausha.co/privacy-policy for more information.…

The Prompt Desk

1
Can custom chips save AI's power problem? 37:26

51 weeks ago37:26

37:26

Cutting-edge AI models use a lot of power. It is not uncommon to hear of data centers under construction that require the power output of the largest nuclear reactors all by themselves. Large tech companies used to tout their green credentials - now they quietly buried that message to pursue fortunes in generative AI. In this environmentally conscious episode of The Prompt Desk, your hosts Justin Macorin and Bradley Arsenault sit down with hardware entrepreneur Raymond Chik to talk about his startup, Untether.ai , to understand what the next generation of AI chips will look like. We dig into how it's possible to make a chip do the same calculations more efficiently, how to compete with Nvidia and other dominant chipmakers, and everything about hardware chips. --- Continue listening to The Prompt Desk Podcast for everything LLM & GPT, Prompt Engineering, Generative AI, and LLM Security. Check out PromptDesk.ai for an open-source prompt management tool. Check out Brad’s AI Consultancy at bradleyarsenault.me Add Justin Macorin and Bradley Arsenault on LinkedIn. Please fill out our listener survey here to help us create a better podcast: https://docs.google.com/forms/d/e/1FAIpQLSfNjWlWyg8zROYmGX745a56AtagX_7cS16jyhjV2u_ebgc-tw/viewform?usp=sf_link Hosted by Ausha. See ausha.co/privacy-policy for more information.…

The Prompt Desk

1
Towards an Inter-Agent Communication Standard 21:21

1 year ago21:21

21:21

In this future-looking episode, hosts Bradley Arsenault and Justin Macorin discuss how agents might communicate with each other in the future. Will agents be making HTTP requests to each other? Will agents dial each other up on the phone? Might it be possible that one-hundred years from now, all our AI systems communicate to each-other over plain-text UTF-8 Telnet connections? Brad and Justin discuss what an interagent communication standard might look like. Brad makes the case that plain-text HTTP Post requests, using plain vanilla HTTP headers, will be how most agents talk to each other. --- Continue listening to The Prompt Desk Podcast for everything LLM & GPT, Prompt Engineering, Generative AI, and LLM Security. Check out PromptDesk.ai for an open-source prompt management tool. Check out Brad’s AI Consultancy at bradleyarsenault.me Add Justin Macorin and Bradley Arsenault on LinkedIn. Please fill out our listener survey here to help us create a better podcast: https://docs.google.com/forms/d/e/1FAIpQLSfNjWlWyg8zROYmGX745a56AtagX_7cS16jyhjV2u_ebgc-tw/viewform?usp=sf_link Hosted by Ausha. See ausha.co/privacy-policy for more information.…

The Prompt Desk

1
Should we let prompts write prompts? 19:22

1 year ago19:22

19:22

In this thought-provoking episode, hosts Bradley and Justin explore the intriguing concept of using AI to write prompts for other AI systems. They delve into the paradoxical nature of this idea and discuss potential approaches to make it feasible. The hosts emphasize the importance of having a structured system and well-defined schemas when attempting to automate prompt creation. They argue that abstract prompt generation without context is impractical for real-world applications. Instead, they propose leveraging existing prompts and system knowledge to guide AI in creating new, relevant prompts. Bradley and Justin examine various methods for AI-assisted prompt writing, including fine-tuning models on existing prompts and using few-shot learning approaches. They stress the need for human oversight in the process, likening LLMs to three-year-olds that require clear examples and guidance. The conversation touches on the challenges of prompt iteration and version control, with Justin suggesting the potential for developing tools to manage prompt versions effectively. The hosts also briefly discuss optimization approaches and the complexities of chaining multiple prompts together. Throughout the episode, Bradley and Justin provide a pragmatic perspective on the future of AI-assisted prompt engineering, balancing enthusiasm for innovation with realistic expectations about current AI capabilities. Listeners will gain valuable insights into the nuanced world of prompt creation and the potential for AI to augment this process in meaningful ways. --- Continue listening to The Prompt Desk Podcast for everything LLM & GPT, Prompt Engineering, Generative AI, and LLM Security. Check out PromptDesk.ai for an open-source prompt management tool. Check out Brad’s AI Consultancy at bradleyarsenault.me Add Justin Macorin and Bradley Arsenault on LinkedIn. Please fill out our listener survey here to help us create a better podcast: https://docs.google.com/forms/d/e/1FAIpQLSfNjWlWyg8zROYmGX745a56AtagX_7cS16jyhjV2u_ebgc-tw/viewform?usp=sf_link Hosted by Ausha. See ausha.co/privacy-policy for more information.…

The Prompt Desk

1
The Bot Delusion 20:30

1 year ago20:30

20:30

Brad and Justin discuss the gap between AI's perceived capabilities and limitations. We explore common misconceptions about chatbots, discussing pre-defined tasks, integration dependencies, and the importance of clearly defining both what AI can and cannot do. Join us for insights on creating more realistic AI systems and managing expectations. --- Continue listening to The Prompt Desk Podcast for everything LLM & GPT, Prompt Engineering, Generative AI, and LLM Security. Check out PromptDesk.ai for an open-source prompt management tool. Check out Brad’s AI Consultancy at bradleyarsenault.me Add Justin Macorin and Bradley Arsenault on LinkedIn. Please fill out our listener survey here to help us create a better podcast: https://docs.google.com/forms/d/e/1FAIpQLSfNjWlWyg8zROYmGX745a56AtagX_7cS16jyhjV2u_ebgc-tw/viewform?usp=sf_link Hosted by Ausha. See ausha.co/privacy-policy for more information.…

The Prompt Desk

1
Experiments with the Networking Bot 21:17

1 year ago21:17

21:17

In this episode of The Prompt Desk Podcast, hosts Bradley Arsenault and Justin Macorin dive into the fascinating world of AI-powered networking bots. Bradley shares his experience creating and refining a bot that makes phone calls to schedule lunches and happy hours, using his own voice. The hosts discuss the technical challenges of building such a system, including voice synthesis, conversation AI, and phone infrastructure. They explore the human factors involved in bot interactions, such as building trust and ensuring people take the bot seriously. Bradley and Justin also touch on future improvements and the broader implications for AI in sales and networking. This episode offers valuable insights for anyone interested in the practical applications of AI in business and social contexts. --- Continue listening to The Prompt Desk Podcast for everything LLM & GPT, Prompt Engineering, Generative AI, and LLM Security. Check out PromptDesk.ai for an open-source prompt management tool. Check out Brad’s AI Consultancy at bradleyarsenault.me Add Justin Macorin and Bradley Arsenault on LinkedIn. Please fill out our listener survey here to help us create a better podcast: https://docs.google.com/forms/d/e/1FAIpQLSfNjWlWyg8zROYmGX745a56AtagX_7cS16jyhjV2u_ebgc-tw/viewform?usp=sf_link Hosted by Ausha. See ausha.co/privacy-policy for more information.…

The Prompt Desk

1
Using Agents to Test Agents 21:41

1 year ago21:41