Replit | We trained our own AI?! | Our code complete mode outperforms other open-source models in HumanEval @replit | Uploaded May 2023 | Updated October 2024, 1 hour ago.
You saw the stats at Developer Day, but let's take a look at what it's *actually* like to use Replit's new Open Source Code Complete model and how it compares to what went before. Spoilers: It absolutely rocks!
Links: huggingface.co/replit/replit-code-v1-3b
00:00 Comedy Gold
00:20 We built our own AI?!
01:24 Python: Regex IP Addresses
03:39 It thinks it's funny!
04:52 Python: MNIST classifier
07:22 What did we train it on?
08:18 PHP: CLI
10:12 PHP: Web UI
11:26 Markdown
12:30 Concilusion
You saw the stats at Developer Day, but let's take a look at what it's *actually* like to use Replit's new Open Source Code Complete model and how it compares to what went before. Spoilers: It absolutely rocks!
Links: huggingface.co/replit/replit-code-v1-3b
00:00 Comedy Gold
00:20 We built our own AI?!
01:24 Python: Regex IP Addresses
03:39 It thinks it's funny!
04:52 Python: MNIST classifier
07:22 What did we train it on?
08:18 PHP: CLI
10:12 PHP: Web UI
11:26 Markdown
12:30 Concilusion