ChatGPT

[Reefs]

All,

I think it is time to set some basic rules re:  AI generated content on this forum, namely ChatGPT.

Lolly posted this message:

Sept 22, 2024 1:46:59 GMT -5 lolly said:

Reefs : I think it'd be a great idea if all cut and paste "AI" are excluded from what is supposed to be a discussion and relegated to a thread specifically designated for AI.

I agree. The habit of some members to "cut and paste" lengthy ChatGPT dialogs goes against the original idea of this forum. So please limit your use of ChatGPT to the absolute minimum. Occasionally using ChatGPT to highlight a point is okay. But using ChatGPT exclusively in your replies or only replying with chatGPT is not okay. That's spamming. And it will be dealt with accordingly.

R

[lolly]

Can we start a thread that uses 2 different AI bots to post and reply to each other? People who like AI could just follow that thread, but it also wouldn't matter if no one did.

[inavalan]

AI can be useful if you use it for what it can do. For example, you can use it to summarize a video, and not having to spend hours watching something, to find out that it wasn't what you expected. It can even summarize in your language of choice a video spoken in a language that you don't understand and that doesn't have subtitles. It isn't a perfect summary, but it can be quite useful.

[Gopal]

Sept 24, 2024 22:53:50 GMT -5 inavalan said:

AI can be useful if you use it for what it can do. For example, you can use it to summarize a video, and not having to spend hours watching something, to find out that it wasn't what you expected. It can even summarize in your language of choice a video spoken in a language that you don't understand and that doesn't have subtitles. It isn't a perfect summary, but it can be quite useful.

As a software developer, I find that ChatGPT significantly aids in enhancing my code.

Since I’m not a native English speaker, it assists me in rephrasing my English without changing the meaning of my writing.

[Gopal]

Sept 24, 2024 9:07:42 GMT -5 Reefs said:

All,

I think it is time to set some basic rules re:  AI generated content on this forum, namely ChatGPT.

Lolly posted this message:

Sept 22, 2024 1:46:59 GMT -5 lolly said:

Reefs : I think it'd be a great idea if all cut and paste "AI" are excluded from what is supposed to be a discussion and relegated to a thread specifically designated for AI.

I agree. The habit of some members to "cut and paste" lengthy ChatGPT dialogs goes against the original idea of this forum. So please limit your use of ChatGPT to the absolute minimum. Occasionally using ChatGPT to highlight a point is okay. But using ChatGPT exclusively in your replies or only replying with chatGPT is not okay. That's spamming. And it will be dealt with accordingly.

R

Yes, I also find it quite annoying when people copy and paste AI-generated paragraphs here. It’s a good decision to separate the thread so those who are interested can go and look there.

[inavalan]

Sept 23, 2024 21:53:59 GMT -5 Reefs said:

Summary of O.I. The Future of AI? Inside Organoid Intelligence Breakthroughs

This is an AI generated summary. There may be inaccuracies.


00:00:00 - 00:05:00
In the YouTube video "O.I. The Future of AI? Inside Organoid Intelligence Breakthroughs," Tony from Immortality explores Organoid Intelligence (O.I.), a biotech field using real brain tissue neurons instead of silicon for AI. The Muro platform, an advanced organoid intelligence system, aims to extend its lifespan from 100 days to 365 days. Researchers at the Scripps Research Institute are also developing O.I., collecting and storing data from organoids to build a comparable AI system. Another approach involves growing brain cells on computer chips, with researchers at IMCB using a printed pattern to guide cell connections. Scientists are experimenting with these brain dishes, teaching them tasks like aircraft control. The goal is to build a sophisticated O.I. platform, with competitors seeking inspiration and novel solutions.

Show more:

00:00:00 In this section of the YouTube video titled "O.I. The Future of AI? Inside Organoid Intelligence Breakthroughs," Tony from Immortality discusses Organoid Intelligence (O.I.), which is not Artificial Intelligence but a field of biotech that uses real brain tissue neurons instead of silicon analogs. The Muro platform, a complete system for organoid intelligence, is the most advanced version submitted to date, with a goal of increasing its lifespan from 100 days to 365 days. The organoids are spherical objects, 500 micrometers in size, and designed to look like a ball sitting on electrodes. The size of the organoid is crucial, as doubling its size would require entirely reworking it to prevent necrosis. The communication between the organoid and the operator is a challenge, as the organoid learns the language that the operator trains it in, and the organoid produces electrical activity that needs to be deciphered. The Euro platform is modular and uses an array of organoids, which can be controlled using Python code and signal processing computer equipment, allowing anyone to run experiments remotely.

00:05:00 In this section of the YouTube video titled "O.I. The Future of AI? Inside Organoid Intelligence Breakthroughs," researchers at the Scripps Research Institute are developing an organoid intelligence platform using real neurons. They collect and store data produced by the organoids in a time series database and aim to build a comparable artificial intelligence system like ChatGPT. Another approach involves growing brain cells on computer chips, but placing each cell on the right connection point is a challenge. Researchers at the IMCB have found a solution by printing a pattern on the chip that the brain cells love to eat, causing them to stick and their tentacles to be guided by the pattern. Scientists like Dr. Deor are experimenting with these brain dishes, teaching them to control tasks such as flying an aircraft. The goal is to build a sophisticated organoid intelligence platform, with competitors looking for inspiration and new ways to tackle challenges.

[laughter]

This is a dialog with ChatGPT about the internals of its design and the tech used to build it. Posted to provide context for a related dialog. The full version is quite long, but might be of interest to the gearheads here.

========================================================================================

Q: I'm interested in learning about AI technology today.

That's a fascinating topic! AI technology encompasses a wide range of concepts and applications. Are you interested in a specific aspect of AI, such as machine learning, natural language processing, robotics, or something else?

Q: Are you familiar with the term "backpropagation"?

Yes, I'm familiar with backpropagation! It's a key algorithm used in training artificial neural networks, which are the building blocks of many machine learning models. Backpropagation is short for "backward propagation of errors," and it's used to calculate the gradient of the loss function with respect to the weights of the network. This gradient is then used to update the weights, gradually improving the network's performance on the training data. Would you like to dive deeper into how backpropagation works?

Q: Was backpropagation used in building your LLM?

Yes, backpropagation was a fundamental part of training the model that powers me, known as GPT-3.5. During training, backpropagation is used to adjust the weights of the neural network based on the difference between the model's predictions and the correct answers. This process helps the model learn to generate more accurate and coherent responses over time.

Q: I've been told from another source that your network has 96 layers, is that accurate?

No, that information is not accurate. The GPT-3.5 model has 96 attention layers, not 96 layers in total. Each attention layer consists of multiple sub-layers, including feedforward neural networks and multi-head self-attention mechanisms, making up the full architecture of the model.

Q: Are the feedforward networks trained using backpropagation?

Yes, the feedforward neural networks within each attention layer of the GPT-3.5 model are trained using backpropagation. During the training process, backpropagation is used to update the weights of these networks based on the error between the model's predictions and the actual targets, helping the model learn to generate more accurate outputs.

===========================================

.. went on from there with 16 additional questions covering the topics of "multi-head self-attention mechanism", source papers, CNN's as compared to RNN's, the "vanishing gradient problem", "long range dependencies", "LSTM (Long Short-Term Memory) and GRU (Gated Recurrent Unit)", and self-organizing maps and other backprop alternatives and further details of ChatGPT's design. Left all that off for brevity and so not to bomb everyone's browser, but will add it in as a follow-on post if anyone is interested.

[lolly]

Just thinking, as AI generates more text, and inevitably, the majority thereof, its training data will be self-generated and result in a fundamental tautology, quite literally of its own making.

[Reefs]

Sept 28, 2024 19:41:03 GMT -5 lolly said:

Just thinking, as AI generates more text, and inevitably, the majority thereof, its training data will be self-generated and result in a fundamental tautology, quite literally of its own making.

Yes.