- 3
- 686
AI Struggle Bus with Guang
United States
เข้าร่วมเมื่อ 21 ส.ค. 2024
hands-on projects to build and deploy AI apps. but with struggles.
I’m so bad at swiftUI even Cursor couldn’t help me
hi :D
00:00 intro
01:38 chapter 1: hatching a plan
03:02 chapter 2: the mirage
04:29 chapter 3: the first error
07:32 chapter 4: a comedy of errors
10:29 chapter 5: burn it all down
11:09 chapter 6: a new way forward
00:00 intro
01:38 chapter 1: hatching a plan
03:02 chapter 2: the mirage
04:29 chapter 3: the first error
07:32 chapter 4: a comedy of errors
10:29 chapter 5: burn it all down
11:09 chapter 6: a new way forward
มุมมอง: 148
วีดีโอ
3 levels of getting LLMs to return valid JSON without threats of violence
มุมมอง 316หลายเดือนก่อน
hi :D 00:00 intro 01:16 level 0: why can’t we just ask for JSON directly? 03:05 level 1: function calling with instructor 05:50 level 2: constrained sampling with OpenAI’s structured outputs 10:28 Level 3: the broke grad student mode resources: - simmering.dev/blog/openai_structured_output/ - openai.com/index/introducing-structured-outputs-in-the-api/ - th-cam.com/video/yj-wSRJwrrc/w-d-xo.html
Build a LLM app to summarize podcasts in 10 mins for free?
มุมมอง 2222 หลายเดือนก่อน
code: github.com/guang/podcast-summarizer 00:00 Intro 00:33 Project Overview and Goals 00:56 Architecture 02:19 Transcribing w/ Whispercpp 05:25 Summarizing w/ Gemini and Instructor 08:00 Building the API 10:06 UI Speedrun 11:23 Demo and chat w/ my buddy and fellow SWE Austin 18:15 You're still here??
*get 4 views of course
I give this video like and hope you will run ads to promote your nice channel
I think that what did at the end is the reasonable thing to do at the current state of AI. You also somehow proved that we, iOS developers, still have a moat. At least for now. Keep up the good work.
Exactly what I was looking for! Thank you!
Sounds like constrained sampling can be game changer if cleverly applied. At the very least it should make code generation guaranteed to be syntactically correct. They can probably easily generate tree sitter grammars for themselves for almost any language where they have enough training material. Next level would be to guarantee type correctness. Maybe they have a bit of this in place for code already since code generation is quite robust as it is.
Great video!