For truly targeted and high-quality results from AI, simply typing commands into a text field is not enough. Even if you access a finished model, domain-specific fine tuning is required. This is where the new iX Special “Artificial intelligence: understand, apply, develop yourself” picks up all interested parties. Experienced developers can find tips on sources of training data and where to find the best ready-made models. State-of-the-art, the special shows how different language models can be linked with LangChain as AI agents. The booklet picks up beginners and those interested in the architecture of language models and how AI image generators work. For everyone in between, the special offers information so that you can really have a say at the moment.
make AI yourself
It doesn’t matter whether you’re adjusting finished models or training your own AI: computing power is what counts. For AI models, developers usually use GPUs to do the calculations. Here the special issue offers a benchmark of current accelerators as a decision-making aid. The special also answers how AI chips are structured and how they work.
In addition to the engine of the AI, it is also about the fuel of the models – the training data. The special issue shows where to get test data for your own purposes and how to assess the quality of this data. The common mantra “S**t in, s**t out” applies even more to AI – especially when companies ultimately want to use their models to process sensitive information from customers or use AI in decision-making.
For the use of already finished, free models, the iX Special takes a look at the Huggingface Transformers Hub. Developers who have data but are still looking for suitable models will find what they are looking for here. The language of AI development is primarily Python. Data scientists use Jupyter notebooks with the scikit-learn or PyTorch libraries to feed their data to the models. Here, newcomers to the topic in the special issue can get started.
Bring know-how up to date
If your own models are then to be put into production, there are a number of guidelines to be observed with the GDPR, Data Act and the maturing AI Act. The iX Special shows the most obvious pitfalls. The issue also guides you through the debate about AI-generated content and copyright: which paragraphs apply and which do not? Where should finished models be treated with caution? Finally, the issue tries to classify the current importance of AI. Two interviews show where we stand and how things could go on.
Subscribers have already received the iX Special with their subscription. Everyone else can find the magazine in the Heise shop and at the newspaper kiosk. PDF and printed version of the magazine cost 14.90 euros. The print version can be ordered postage free up to and including June 22, 2023. The bundle booklet + PDF costs 20.90 euros, the super bundle with booklet and the book “Natural Language Processing with Transformers” is 61.80 euros.
