We are currently rolling out Copilot within Topigs Norsvin. To get a good understanding of what GenAI and Copilot are and what exactly the rollout will mean for us and what the rules of the game are, we will publish some articles here in the near future to explain exactly what GenAI is, how we are going to use it and what rules we will have to adhere to.
Because it concerns new technology, where on the one hand we cannot oversee exactly what the impact and dangers will be for the organization and on the other hand we will have to gain new knowledge and insights, the implementation will be implemented in phases. Anyone who already has knowledge or would like to gain knowledge of GenAI is cordially invited to contribute to establishing a solid environment, where we can use the new tools to increase our productivity and accelerate developments, while at the same time knowing what we are doing and putting security first because the IP of Topigs Norsvin must be well protected.
In 1996, world chess champion Garry Kasparov defeated chess computer Deep Blue with a score of 4-2. A year later, he played another six games against the supercomputer developed by IBM, but this time lost 3 1/2 – 41/2; it was the first time that a reigning world chess champion was beaten by Artificial Intelligence.
On November 30, 2022, ChatGPT was made available to the general public by the company OpenAI, and since then, we can't imagine a world without AI. Was Deep Blue a predecessor of ChatGPT and if so, why did it take more than 25 years for us all to have access to AI?
The programmers at IBM who programmed the chess program that Deep Blue used, undoubtedly learned a lot and gained insights in performing this task. However, the AI program they wrote is fundamentally different from the Generative AI programs we use today when filling out our taxes, summarizing a document or deciding what we are going to eat for dinner tonight. That big difference is in the addition 'Generative'.
Kasparov played against a program that made decisions by following specific instructions and predefined rules. Supercomputer Deep Blue was able to perform more than 11 billion calculations per second and was able to calculate the game and all the possibilities of the next move and choose the right one. Deep Blue was not creative but was able to calculate a limited set of assignments very quickly.
A simple iPhone 17 alone is 1000x more powerful than the supercomputer from 1998. This explosively increased computing power has made it possible, that we can now train so called Large Language Models (LLM) and those LLMs make GenAI possible.
An LLM is a neural network: a computational model inspired by the human brain, using (mathematical) nodes that process input data, assign it a 'weight' and then send it to the next node. Just like neurons in the brain, they fire a signal when they reach a certain threshold. Because the neurons work together, it becomes possible to discover patterns in, for example, written text.
The input for the training of an LLM is therefore text and text can be found in great amounts on the internet. If you offer a lot of words and sentences to the LLM, it becomes possible within that model to calculate the probability of the next word that will follow. If I write 'I love', the probability that I will write 'you' as the next word is quite high. This is of course a very simple example, but if you imagine that you give the model trillions of pieces of text and the model processes all those pieces of text, compare them with each other and calculate the ratio between the words, then it becomes conceivable that a computer can guess very accurately what the next word will be. The LLM does not understand language in a way a human understands it, but the LLM calculates the connections between words and we experience that behaviour as intellect.
There are thousands of Large Language Models in the world, but the most well known are (Chat)GPT (OpenAI), Gemini (Google), Llama (Meta), Claude (Anthropic) and Grok (X). Every now and then you hear that a new and more powerful version of one of these LLM’s is released. What this means is that the model is trained on more data. The first model of GPT was trained on 175 billion words or parameters. When training the most recent models, quadrillion (one million billion or 1 with 15 zeros) parameters were used. It is also said that the model has become 'smarter and the AI companies currently use the GenAI tools they developed earlier to create the next generation of GenAI: you get AI creating AI.
In the Terminator films, the AI system called Skynet, became self-aware on August 29, 1997 at 2:14 am Eastern Time. Have we passed this point in the real world already or is that just doomsday thinking?