LLMs are good at in-context reasoning; given a bunch of information in context, they can perform complex reasoning tasks and solve problems. But can they perform latent multi-hop reasoning on pre-trained data? Simply put, can LLMs connect and traverse through implicit knowledge
PROMPT ENGINEERING
-
Devin: AI Agent Writes Complete Software Projects Autonomously
By
–
Cognition AI just introduced Devin, an autonomous AI agent that can write entire software projects based on prompts.
— Rowan Cheung (@rowancheung) 14 mars 2024
If it lives up to hype, Devin could be the first true look into a future where anyone can spawn an AI worker, no coding skills required. https://t.co/D4C8QXPzu7Cognition AI just introduced Devin, an autonomous AI agent that can write entire software projects based on prompts. If it lives up to hype, Devin could be the first true look into a future where anyone can spawn an AI worker, no coding skills required.
-
Claude generates animations from AI-generated code tutorial
By
–
Speaking of Claude, @dr_cintas found a way to generate animations from code generated directly from the chatbot.
— Rowan Cheung (@rowancheung) 14 mars 2024
He's doing a full tutorial in The Rundown tomorrow!
Get it for free: https://t.co/Gm5KwesnA7https://t.co/4j2i1OLXtiSpeaking of Claude, @dr_cintas found a way to generate animations from code generated directly from the chatbot. He's doing a full tutorial in The Rundown tomorrow! Get it for free: https://
therundown.ai/subscribe -
Midjourney Prompt Template for 360-Degree Equirectangular Images
By
–
1. Prompt template:
Use this simple prompt template to generate your desired image: "360 degrees equirectangular projection of [insert prompt] –tile –ar 2:1" Here's what Midjourney generated for me: -
GPT-4 Plugin Selection Intelligence Capabilities
By
–
Back when plugins were around, you could actually do that. GPT-4 used to intelligently choose the plugins you had installed to perform the request in the best way.
-
Cmd-R Outperforms GPT4 on Large Documents, Custom GPT Routing
By
–
I used Cmd-R on large documents, and it performed better than GPT4 for me! Yeah, NIAH correlates with that. You know how you can mention a custom GPT today? I would love to see the ability to make multiple mentions to intelligently direct flows based on each GPT's abilities.
-
Multi Needle Haystack: RAG Retrieval and Reasoning Challenges
By
–
Multi Needle in a Haystack Benchmark RAG often isn't about just retrieving a fact, it's about retrieving multiple facts and reasoning over them We find that: Retrieval performance decreases with more needles
Reasoning over retrieved needles is harder than retrieving -
Future AI Capabilities: RAG, Context, Agents and Prompt Optimization
By
–
– Better RAG on large documents (the new @Cohere model is incredible for that)
– Larger token context
– Some sort of agentic behavior
– Ability to improve its own prompts
– Orchestrate subagents intelligently
– Multiple custom GPT tags -
Fine-tuning Llama2 for FAQ-based Chatbot on Indian Tax Law
By
–
Question/Answering systems are a cornerstone use case for #LLMs. Check out this short tutorial from Ludwig hackathon winner, Yogesh Haribhau Kulkarni, to hear how he fine-tuned #Llama2 on FAQ data for Indian tax law to create a #chatbot. https://
pbase.ai/4adtmSW -
Test Prompts LangSmith Playground Generalization
By
–
🛝 Execute a Test Run in the LangSmith Prompt Playground🛝
— LangChain (@LangChain) 13 mars 2024
With this new LangSmith feature, you can test your prompt over a series of inputs to see how well it generalizes across different contexts or scenarios.
Furthermore, you can compare the outputs of your runs to… pic.twitter.com/5rEczY0Z7UExecute a Test Run in the LangSmith Prompt Playground With this new LangSmith feature, you can test your prompt over a series of inputs to see how well it generalizes across different contexts or scenarios. Furthermore, you can compare the outputs of your runs to