SmartGPT: Make ChatGPT Smarter

[@engineerprompt]

Want to connect? 💼Consulting: calendly.com/engineerprompt/consulting-call 🦾 Discord: discord.com/invite/t4eYQRUcXB ☕ Buy me a Coffee: ko-fi.com/promptengineering |🔴 Join Patreon: Patreon.com/PromptEngineering

[@marilynlucas5128]

All these advancements are useless if developers keep ignoring open source and developing AI apps that require Open AI Api keys. Smh

[@JustAThought01]

Consider a two track process: cloud processing for major social/business services and at home products for individual/small business implementations. Cloud implementations will have much greater funding for researching possible implementations and then open source implementations once the best path forward is identified.

[@Lorv0]

True but there are tools like openllm which can act as an OpenAI server. So basically you could use this tool or any other tool using OpenAI keys with your own local LLM

[@marilynlucas5128]

@@Lorv0 I understand what you're saying. I am aware of Openllm and even vLLM and its paged attention feature that enables fast inference. I just didn't get the part where you said I can use OpenAI API keys with openllm?

[@MattGoldenberg]

These methods can all be used without access to the model.

[@YT-Gmailer]

langChain is capable of using other models. You don't necessarily have to use gpt3.5-turbo

[@twentiesguide8517]

Pls do a video on few shot learning using LLAMA 2 for text classification. Nobody have done this and it’ll be very useful for us.

[@satyamgupta2182]

Yes. Great idea

[@pavanpraneeth4659]

Yes we need this

[@AdrienSales]

Thanks a lot for having made me discover this approach. I've just tested it and I have started prototype and tune prompts on it. I wonder if you tried to pipe chain after that with common expression language ?

[@MichealAngeloArts]

There is probably no point for the ideation phase to have n>1 as all the resulted ideas have produced the same output (e.g. 6 hours as in your first experiment). Setting the temperature to a high value comes on the risk of increasing hallucinations. As a result you may end up having all your responses unfactual. A better implementation of ideation is to be guided sequentially rather than parallel (same as when you have "guided" the model in the UI approach that there is a logical flaw to its first idea)

[@minercraftmen]

Exactly. I was thinking the same thing. It's only good if you have to give the answers separately for some reason. I tried it with a plain local model to solve the problem. Could not solve it. Then I wrote a 3 step prompt about like the one in the video and it solved it with one output. What I noticed is that it responded very well to the "common sense" prompt. The prompt: (sorry for bad eglish) ## Ideas: Make 3 different ideas for the same input. Use common sense. ## Critiuqe: Look at your 3 responses and search for flaws in them. ## Final: Conclude from the critics what would be the right answer. Dont overthink and look for additional information. Answer: (OpenChant 3.5) Idea 1: The number of shirts doesn't affect the drying time, so 30 shirts will also take 6 hours to dry. Idea 2: The drying time is directly proportional to the number of shirts, so 30 shirts will take 3 times longer, or 18 hours, to dry. Idea 3: The drying time is indirectly proportional to the number of shirts, so 30 shirts will take 1/5th the time, or 1.2 hours, to dry. Critique: Idea 1 is incorrect because the number of shirts does affect the drying time. Idea 2 is incorrect because the drying time is not directly proportional to the number of shirts. Idea 3 is incorrect because the drying time is not indirectly proportional to the number of shirts. Final: Based on the critiques, the correct answer is that 30 shirts will take 6 hours to dry.

[@WeylandLabs]

Yay ! We are evolving finally ... 🤣

[@paulstevenconyngham7880]

Got a link to the colab notebook you used in your example by any chance?

[@engineerprompt]

There is a link to the LangChain documentation in the description which has the code, if not I will share the colab link.

[@paulstevenconyngham7880]

@@engineerprompt doh! thanks man, love your content - keep up the excellent work

[@coder-r]

@@engineerprompt share your collab link if you can please

[@coder-r]

Is there a link the notebook?

[@engineerprompt]

Yes, there is a link to the actual documentation from langchain which was used during the video

[@Codescord]

MAKE A VIDEO ON "HOW TO TRAIN LLAMA 2 FROM TXT FILES"

[@franchocou]

First

[@gunngunn6763]

How to get free ChatGpt 4?

[@engineerprompt]

Use bing chat

[@manabchetia8382]

Can we use SmartChain with Lllama 2 ?

[@WeylandLabs]

Bingo ! Finally intelligence has been detected...

[@dmerriman]

Yes, you can use Ollama, switch out the OpenAI module for from langchain.llms import Ollama and for the llm use the following: llm = Ollama(model=llama2", verbose=True)

[@JustAThought01]

Very interesting concept. Good for comparing different solutions for problem solving. Current event problem: how best to control economic inflation.