Checking work with a model from a different family is now standard practice for agent work. Cursor and meta harnesses make it easy to run agent teams on models from different providers. It's trickier if Claude Code is your main tool, because Claude Code can only run subagents on Claude models. I published harness-bridge, a plugin that hands a subagent's job to the Codex CLI or pi, so it runs on GPT, DeepSeek, GLM or anything else on OpenRouter.
https://t.co/hByzJU4g21
I spent a couple of weeks trying to make a 1.3B language model safe for kids to talk to.
Fine-tuning raised the pass rate on harmful prompts from 1/141 to 70/141. Not enough to trust. What worked: code outside the model that makes every safety decision. https://t.co/XuV5hQxAHD
Promise is a debt! You wanted to try the tool for free. Now you can!
๐ฅ TESTING DAY STARTING NOW: 24-hour trial for everyone ๐ฅ
120 tokens (that's 24 upscales to 2k resolution)!
IMPORTANT: How to get your GPU tokens ๐งต๐
Ever dreamt of being mentored by a celebrity, like a billionaire investor, @RayDalio?
Well, with @langchain, it will take only 30 mins to pick up the brains of the most successful people on earth!
I've got another cool project for you:
1. Download Youtube podcasts and transcribe them with @OpenAI's Whisper
2. Break down transcriptions into useful bits with @langchain's loaders and splitters
3. Create embeddings, load them together with documents into @pinecone for semantic access
3. Use @langchain templates and chains to chat with @RayDalio on the current economic situation and learn how you can invest in inflationary environment!
And you know the drill -> Like, Retweet and Iโll DM the notebook! ๐ฉ
Let's kick this thing off! ๐
1/ Exciting news, Web3 developers! ๐โจ
Brace yourselves for the revolutionary power of Covalent Network's latest innovation: The Refiner.
Let's dive in and explore how it's transforming Web3's data infrastructure. ๐
How do you stay on top of the latest research? Easy, chat with multiple researchers on @arxiv using @langchain , @OpenAI , @trychroma and @arizeai!
I've cooked up an insight packet 30-mins mini course for you to become a research wizard!
What you'll master:
- Setting-up Python virtual environments ๐
- Extracting knowledge from @arxiv papers like a pro ๐
- Understanding the significance of certain parameters for better outcomes๐ง
- Creating and storing embeddings in @trychroma
- Visualizing the embeddings with @arizeai ๐
Using @langchain for:
- Extracting relevant info from @arxiv papers per user prompt ๐๏ธ
- Creating high-quality responses with citations using @OpenAI 's LLMs! ๐
Also want the notebook?
Like, retweet, and follow, and Iโll DM it straight to your inbox! ๐ฉ
Let's get this weekend started! ๐
Wanna crack the @langchain's codebase?
I've got a 10-minute crash course here for you to build an intelligent code assistant by using @langchain , @OpenAI , and @pinecone.
You will get answers and examples to questions like how do LangChain agents or callbacks work?, or what are chains?
You will set-up the environment, load the code repository, create embeddings and store them in @pinecone vector database.
Next, you will combine user prompt and chat history to dig out relevant pieces of code, and flex your AI muscles with @OpenAI's LLMs (gpt-3.5-turbo) to answer questions!
Sounds exciting, doesn't it? Let's get started!
If you want to get the notebook ๐
Like, RT + comment "notebook", and
I'll DM it to you ย (must be following)