AD
Episode
338
Interview
Web News

Does AI Have Tech Bias? | AI All The Things

Recorded:
November 12, 2024
Released:
November 26, 2024
Episode Number:
338

In this episode, Matt and Mike introduce a new episode type dedicated to exploring the rapidly evolving world of AI. As AI tools advance at lightning speed, staying informed is critical for developers navigating this transformative era.

This week, they dive into the evolution of AI-powered development tools, from simple autocompletes like Copilot and TabNine to full-scale IDE solutions like Cursor and Supermaven. They discuss the pros and cons of using these advanced tools for multifile code generation and manipulation, highlighting both the efficiencies and risks they bring.

The conversation then shifts to the rise of AI-driven full-stack application generators, such as Bolt.new, V0, and GitHub Spark. These tools can build entire applications from simple prompts but come with a notable downside: tech bias. Matt shares his experience building a podcast website and highlights how AI's reliance on popular frameworks can limit the adoption of emerging technologies.

Finally, the duo debates the future of AI in the development industry. Will AI replace developers in 1, 3, or even 10 years? Tune in to find out!

Listen

Also available on...
...and many more, check your podcast app!

Who’s in This Episode?

Show Notes

Episode Sponsor - Wix Studio

We'd like to thank this episode's sponsor for their support!

Wix Studio: The Web Platform for Agencies and Enterprises

Wix Studio is the new web platform tailored to designers, developers and marketers who build websites for others or for large organizations. The magic of Wix Studio is its advanced design capabilities which makes website creation efficient and intuitive.

Check out Wix Studio today.



How to support the show

Patreon

Prices subject to change and are listed in USD

  • Support the show from as little as ~$1/month
  • Get a shoutout at the end of the episode (while supplies last) for just ~$3/month
  • Help support the HTML All The Things Podcast: Click Here


Show Notes

Introduction

  • New episode type where we will cover the latest in AI topics
  • AI evolves extremely rapidly and we figured the industry needs to constantly be told what’s here and what's coming to keep themselves ready for the next phase

AI autocomplete has evolved

  • Tools like Copilot and tab nine were created as just a simple autocomplete, where you are typing something and it would fill in the next line or two, or next code block for you
  • This worked well but companies started to try to figure out how to expand this.
  • Cursor came along to provide an entire IDE based off of VS code that enables it to manipulate not just the next few lines, but most of the files in your codebase
  • Supermaven has this capability as well within vscode and as of a couple weeks ago, copilot has deputed their competitor which allows for multifile creation/manipulation
  • Pros
    • Allow for rapid development
    • High knowledge about the whole codebase so better suggestions
    • Easier to chat with and get directly contextual answers and suggestions based on the codebase
  • Cons
    • Easier to let a mistake though as instead of reviewing a few lines you’re reviewing several files that were generated at once
    • You’ll have less knowledge of your codebase because the decision-making was all done by AI

AI has tech bias 

  • Another set of tools has also arisen, which is complete component and project generation tools
  • Bolt.new from Stackblitz, V0 from vercel, and GitHub spark are good examples
  • These tools can take a regular text prompt and generate an entire full-stack application for you
  • These applications can have authentication, a backend, a database, integrations, really anything you can correctly prompt for
  • I created a little rss reader/parser that would take our html all the things rss feed, parse it and create a podcast website with all our episodes and the ability to search them (bolt.new)
  • One major thing I have noticed is that these systems so far have serious tech bias
    • Because these were trained on open source code that means that the more popular a framework or library is, the more chance it has of getting the implementation right when using it
    • I tried the same app in both sveltekit and next js and was not surprised to see the next.js app to be implemented in a much better way then sveltekit
      • For example, it had trouble figuring out how to use sveltekit server routes, instead forcing everything to be an api which didn’t really work for the usecase
  • This worries me because it’s another barrier for new tech to be created and adopted
    • Would you recommend someone new to use a framework that hasn’t been fully indexed by AI?
  • Potentially to fix this maybe AI companies will allow of rapid indexing of new tech if submitted somewhere. 
    • New tech could create a bunch of examples and detailed documentation and pass that to AI to rapidly index and learn from so that it can be represented in the suggestions/completions

Status check on the future 

  • With what you know today do you think AI will replace our jobs in the upcoming 1 year, 3 years, 5 years, 10 years?