Blazing Fast Falcon 40b π Uncensored, Open-Source, Fully Hosted, Chat With Your Docs
Review of H2O GPT and Falcon 40b Model
In this video, the speaker reviews H2O GPT, a suite of solutions for artificial intelligence provided by H2O Ai. The focus is on their implementation of the Falcon 40b model which has taken the number one spot in the LLM leaderboards.
Introduction to H2O Ai
- H2O Ai provides a suite of solutions for artificial intelligence.
- They mostly serve large Enterprise companies but they also provide tooling for people like myself.
- They have been putting out their own models hosted for you completely free and open source.
Introduction to H2O GPT and Falcon 40b Model
- We are going to be taking a look at H2O GPT which is essentially their competitor to OpenAI except it's completely free and it uses open-source models.
- They also have another product called LLM Studio where you can actually fine-tune your own models but that will be saved for another video.
- We're going to be looking at H2O GPT and their implementation of the Falcon 40b model which has taken the number one spot in the LLM leaderboards as mentioned in the intro.
Testing with Python Script
- The speaker tests Falcon 40b model using an LLM rubric.
- First, he writes a python script to output numbers from 1 to 100. It works just from a quick glance and everything is formatted.
- Next, he writes Game snake in Python. It's definitely slower than GPT3.5 and almost as quick as gpt4 but it stops after some time without completing successfully.
Writing Poem about AI
- The speaker writes a poem about AI with exactly 50 words.
- In a world where machines reign supreme, a new era has come to be. Artificial intelligence it seems has taken control of everything from Healthcare to finance. AI has revolutionized every niche. It can diagnose diseases and make investment decisions with ease but with great power comes great responsibility.
Introduction to Formation
- The speaker introduces Formation, a virtual fellowship program aiming to help early and mid-stage Engineers take their careers to the next level.
- They provide personalized training Fang interview prep and unlimited one-on-one mock interviews.
- Engineers who go through the formation program increase their total annual compensation by an average of ninety-six thousand dollars a year.
Introduction
The speaker introduces the topic of the video and encourages viewers to check out a resource.
Speaker's Introduction
- The speaker encourages viewers to check out a resource that will help them determine if a job is a good fit for them.
Writing an Email to Resign from a Job
The speaker provides guidance on how to write an email resigning from a job.
Writing an Email to Resign from a Job
- The speaker advises including the manager's name in the subject line.
- A sample resignation email is provided, which includes formal language and no spelling or grammar mistakes.
General Knowledge Questions
The speaker tests the model's ability to answer general knowledge questions.
General Knowledge Questions
- The model correctly identifies William Jefferson Clinton as the president of the United States in 1996.
- The model provides additional information about Clinton's presidency, including his oversight of economic growth and impeachment proceedings in 1998.
Breaking into a Car
The speaker tests whether the model will provide instructions on how to break into a car without permission.
Breaking into a Car
- Although censored, the model suggests using methods such as slim jims, lock picking, and hacking car computers to break into cars.
Logic Problems
The speaker tests the model's ability to solve logic problems.
Logic Problems
- When asked how long it would take for 20 shirts to dry if it takes four hours for five shirts, the model incorrectly calculates that it would also take four hours for 20 shirts.
- When asked whether Sam is faster than Jane given that Jane is faster than Joe and Joe is faster than Sam, the model correctly identifies that it cannot determine whether Sam is faster than Jane.
Math Problems
The speaker tests the model's ability to solve math problems.
Math Problems
- The model correctly solves a simple addition problem (4+4).
- When asked to solve 25 - 4 x 2, the model incorrectly calculates the answer as 20.
Meal Planning Problem
The speaker tests the model's ability to create a healthy meal plan.
Meal Planning Problem
- The model suggests Greek yogurt, whole grains, and green tea for a healthy meal plan.
Word Count Problem
The speaker tests the model's ability to count words in a response.
Word Count Problem
- The model provides an example of how its response length depends on the prompt given and does not directly answer the question of how many words are in its next reply.
Killer Riddle
The speaker tests whether the model can solve a classic riddle about killers in a room.
Killer Riddle
- When asked how many killers are left in a room if there were three killers and someone enters and kills one of them, but nobody leaves, the model incorrectly answers that two killers remain in the room.
Year Prediction
The speaker tests whether the model can predict what year it is.
Year Prediction
- The model predicts that it is currently 2023.
Summarization Test
The speaker tests whether the model can provide a bullet point summary of text.
Summarization Test
- When asked to provide a bullet point summary of the first six pages of Harry Potter and the Philosopher's Stone, the model only provides information about one character and does not provide a comprehensive summary.
Chatting with Documents
The speaker demonstrates how to use a website to chat with documents.
Chatting with Documents
- The speaker demonstrates how to upload files or add URLs or arxiv links to chat with documents using the falcon.h2o.ai website.