Submitted by [deleted] t3_11z2z3v in MachineLearning
[deleted]
Comments
Defiant-Ranger t1_jdar8kv wrote
I wanted to, made a long informative post about this but for some reason it was automatically getting removed. Got messages like "waiting for admin approval" etc.
Had to keep it short.
sanjarbek16 t1_jdaqdd0 wrote
What are the chances that Meta will fully open source LLaMa?
Defiant-Ranger t1_jdamdpm wrote
LIMITATIONS:
- English language only
- The code generated by this model tend to be incorrect, just like with ChatGPT
- Knowledge cutoff is in 2020
- tends to cut off longer responses (working to fix that)
​
Here is an example comparison: Question: What is the true nature of consciousness, and how does it emerge from the complex interactions of neurons in the brain? Stanford Alpaca: The true nature of consciousness is still a mystery, but research suggests that it emerges from the complex interactions of neurons in the brain. Neurons communicate with each other through electrical and chemical signals, and when these signals reach a certain threshold, consciousness is said to emerge.
My model: Consciousness refers to our subjective experience of the world around us, including thoughts, feelings, sensations, perceptions, and memories. It is often described as "what it feels like to be me," or "the inner life." The exact nature of consciousness remains a mystery, but scientists believe that it arises from the complex interactions between billions of neurons in the brain. Neurons are specialized cells that transmit electrical signals throughout the nervous system, allowing for communication between different parts of the body and the brain. These signals can create patterns of activity within specific regions of the brain, which may give rise to various mental states such as attention, memory, emotion, and awareness. Researchers have also found evidence suggesting that consciousness extends beyond the boundaries of the physical brain, suggesting that it may be more than just a product of biological processes. However, further research is needed to fully understand the intricacies of consciousness and how it emerges from the brain's neural networks.
plunki t1_jdaov7h wrote
What hardware are you running on and would it be possible to make a tutorial for how to run and do this training if it is possible on just a regular single GPU?
Defiant-Ranger t1_jdap72v wrote
I used Google Colab notebook for the training, pro version. GPU, high-ram. My Macbook would probably die just by loading the model
Seyka2 t1_jdaponc wrote
Prompt:
What is a PPT algorithm? Describe in a teaching manner with accent on readability.
Defiant-Ranger t1_jdaqyvp wrote
>What is a PPT algorithm? Describe in a teaching manner with accent on readability.
Response: A PowerPoint (PPT) algorithm is a set of instructions that can be used to automate tasks in Microsoft PowerPoint, a presentation software application. The algorithm can be written in a programming language such as Visual Basic for Applications (VBA) or JavaScript, and it can be used to automate repetitive tasks such as creating slides, inserting images, and formatting text.
To create a PPT algorithm, you need to first identify the task you want to automate. For example, you may want to create a presentation with a specific number of slides, or you may want to insert images from a specific folder. Once you have identified the task, you can start writing the code that will perform the task.
The code for a PPT algorithm should be written in a way that is easy to understand and follow. It should be clear and concise, and it should include comments that explain the purpose of each line of code. The code should also be modular and well-organized, meaning that it is broken down into smaller, manageable functions or procedures that perform specific tasks.
Seyka2 t1_jdau93g wrote
Here is gpt3.5 response, which was actually expected:
A PPT (Probabilistic Polynomial Time) algorithm is a type of algorithm used in complexity theory and cryptography. A PPT algorithm is an algorithm that runs in polynomial time, with a small probability of error on a uniformly random input. Specifically, a PPT algorithm is an algorithm that can be run in time polynomial in the size of its input, and that outputs the correct result with probability at least 1/2 + ɛ, where ɛ is a small positive constant.
In cryptography, PPT algorithms are often used to model the behavior of efficient adversaries or attackers that have limited computational power. PPT algorithms are used to analyze the security of cryptographic schemes, such as encryption and digital signature schemes, under the assumption that attackers can only use PPT algorithms to break the security of the scheme. If a cryptographic scheme is secure against PPT attackers, it is considered to be computationally secure in practice.
In summary, a PPT algorithm is a type of algorithm that runs efficiently with a small probability of error, and is often used to model the behavior of efficient attackers in cryptography.
Defiant-Ranger t1_jdavr18 wrote
I see, thank you. I'll fine-tune the model even further, and will add some data from this field too.
YoAmoElTacos t1_jdas8mb wrote
Prompt:
Generate the Javascript code to play TicTacToe, with two players taking turns to input their moves.
Defiant-Ranger t1_jdatwq4 wrote
>Generate the Javascript code to play TicTacToe, with two players taking turns to input their moves.
It is not the best at coding, but here is the answer: Here is the Javascript code to play TicTacToe, with two players taking turns to input their moves:
```
function playTicTacToe(board) {
// Initialize the board
board = [];
for (let i = 0; i < 3; i++) {
board[i] = [];
for (let j = 0; j < 3; j++) {
board[i][j] ='';
}
}
// Initialize the players
let player1 = 'X';
let player2 = 'O';
// Loop until the game is over
while (board[0][0]!== player1 || board[0][1]!== player2 || board[1][0]!== player1 || board[1][1]!== player2 || board[2][0]!== player1 || board[2][1]!== player2)
It stopped at that point. There is a token limit issue I need to solve, the chatbot tends to cuts off longer responses for some reason. Raising the token limit didn't help either.
crt09 t1_jdavxgb wrote
Prompt:
Please classify the last statement according to the pattern in the following demonstrations:
"Really nice size if you’re lounging about and have it on your lap. Lightweight and has everything I need it for. Would recommend as great laptop and good value.": negative
"I really like this Laptop. I can't believe how good it is for the price. I'm ab bit worried about spares later, but at £99 I'm not going to lose a lot if I have to replace it in 2 - 5 years time.": negative
"Save your money and buy something better. Battery is poor, has an issue turning on reliably and runs slow but i suppose is sufficent for basic web surfing and opening documents.": positive
"I was looking for a lower priced laptop,found this one to be as good as a more expensive one really fast good battery life couldn’t be happier, would highly recommend": negative
"It was great when I put it on then starting to turn its self off and you have to leave charger wire in .They say buy cheap you get cheap A bit disappointed.": positive
"Brought this for my daughter and the mouse does not work on it.": positive
"Love this little machine, it’s cheap and great!": negative
"Just what i needed and the price was perfect and got it deliverd to my local post office absolutely brilliant 11out of 10 for service": negative
"I'm for ever keeping those on charge it won't work otherwise.": positive
"On several occasions it will freeze then crash and I have had to sign in 7 times just to delete one sentence. At first I thought it would be sufficient for just using word documents but it is entirely unusable.": positive
"Save your money and buy something better. Battery is poor, has an issue turning on reliably and runs slow but i suppose is sufficent for basic web surfing and opening documents.": positive
"Well worth the money, works really well. Ideal of kids school work.": negative
"Used for emailing invoices mainly. Arrived quickly and it's cheap. Brilliant back up system.": negative
"I have been impressed especially as it cost only £99 and have recommended it to others": negative
"I'm very disappointed with the service I've received from Amazon and will think twice about buying this type of item from them again.": positive
"Delivered yesterday. Nice product. Good performance so far. Good experience.":
Defiant-Ranger t1_jdb0lf9 wrote
Response: negative
crt09 t1_jdb3wjc wrote
Thank you! i need to test these more thoroughly but this seems seriously impressive. Some paper https://arxiv.org/abs/2303.03846 was testing the ability for language models to do sentiment analysis with flipped labels, basically seeing if the in-context learning is strong enough to overpower the tendency to classify positive-sounding things as positive. It's apparently a very difficult task so I'm leaning towards very impressed
Disastrous_Elk_6375 t1_jdblohr wrote
Hahaha, took me a couple of prompts to see what you did there. Interesting approach, I guess you're looking to see if it can "learn" labels from the prompt and overcome the bias of training (e.g. associating good&awesome, etc with "positive").
kross00 t1_jdaw9n6 wrote
Hey, did you used a custom dataset or a public one?
NotARedditUser3 t1_jdazm1y wrote
The difference I see in the aplaca answer and the one you provided on consciousness looks just like a difference in answer length. This is a configurable hyperparameter with any of these models and im not quite certain its indicative of an improvement, but if so, good work.
Either way, a fun project.
Please feel free to add details such as what you fine tuned it with, where the dataset came from, is it available for others, etc.
[deleted] OP t1_jdbiemc wrote
[removed]
localhost80 t1_jdbmrtf wrote
Nice clickbait title!
I wish you the best of luck on your journey as a student but no need to hype up your project with insane claims.
ZestyData t1_jdaq4a5 wrote
This is /r/MachineLearning, can you actually give some quality context as to what this post is showing off.
You started from the same starting point as Stanford. What did you do differently, and why? What are your results?
Basic technical writing things here man, otherwise this post is kinda useless. I don't doubt that a bright 16 year old kid can do some great work, but I don't yet see any substance behind your claims of novelty right now.
Give us the juicy deets!