Skip to main content

discussion 2024-07-26

Summary​

In the Discord chat, Metapontum highlighted Supercoder's focus on quality of life and interface advancements such as rewind/fastforward features and infrastructure for instant VM cloning to facilitate parallel attempts, suggesting a shift towards ease of backend agent swapping rather than competing directly in SWEbench. Shaw agreed, speculating that Supercoder is betting on an improved user interface with better models coming at no extra cost, anticipating the release of their own llama3.1 next week.

FAQ​

  • What is Supercoder?

  • Metapontum: Supercoder is a tool or platform worth checking out, particularly in the context of quality of life/interface advancements like rewind fastforward, infra for instant cloning of VMs, and easy swapping of backend agents.

  • Are they focusing on competing with SWEbench?

    • Metapontum: Given how quickly they got surpassed in the SWEbench rankings, it is doubtful that they are doubling down on competing directly against it. Instead, their focus seems to be more on interface improvements and backend flexibility.

  • What kind of advancements can we expect from them?

    • Shaw: They seem to be making an "interface bet," with the expectation that better models will come free as a result. There is speculation about them releasing their own llama3.1 version soon, which could potentially have significant improvements or features.

Who Helped Who​

  • Metapontum helped Shaw with understanding the focus on quality of life/interface advancements by explaining their approach to rewind, fastforward features and VM cloning for parallel attempts.
  • Metapontum provided insight into the backend flexibility regarding agent swapping which could be beneficial in addressing interface challenges.

Action Items​

  • Technical Tasks
  • Check out Supercoder if not seen yet (mentioned by metapontum)
  • Focus on quality of life/interface advancements like rewind, fastforward, and infra for instant cloning of VMs to allow parallel attempts (mentioned by metapontum)
  • Make the agent part easy to swap out on the backend (mentioned by metapontum)
  • Feature Requests
    • Developing their own llama3.1 with improved models (implied suggestion by shaw, agreed upon by metapontum)