Luminous - The German Alternative to ChatGPT
Everyone is talking about ChatGPT! But there is also a company from Europe that can rival OpenAI’s ChatGPT. We are talking about Aleph Alpha.
The company has also developed a Large Language Model (LLM) like ChatGPT. The models are called Luminous. According to Aleph Alpha, Luminous is one of the most sophisticated AI language models in the world. Time to try it out.
Most AI experts say that the current decade will be the decade of AI transformation. And that seems to be the case. In the coming months and years, we will see more and more new AI models that make the impossible possible.
What can you expect in this article? You will learn about the functionality of the LLM Luminous from Aleph Alpha. We look at the individual model variants and explain for which use cases you can use the models.
In this article, we look at the current state of the art by comparing the two LLMs of Aleph Alpha and OpenAI. LLMs have the potential to significantly impact the economy by enabling new applications and business models. LLMs can generate human-like text, allowing businesses to automate tasks like content creation, customer service, and even software development.
This automation can lead to cost savings and increased efficiency. That can lead to increased revenue and profitability. Nevertheless, all AI creations should be critically questioned and validated. But the future is coming, so be a part of it.
What is Luminous?
Aleph Alpha understands Luminous as a family of Large Language Models. LLMs can process and produce human texts. The models are trained with very large amount of data. We can also say that the models “read” the texts. After the training, the models have a structural understanding of language and can produce it.
The Luminous family currently consists of three models that differ in complexity and ability. Furthermore, all models are available in five languages: English, German, French, Italian and Spanish. Unlike other large language models, the Luminous family has the ability to work with images as well. Aleph Alpha calls this multimodality. In the following, we show the capabilities of the three models from the Luminous family.
Luminous-base:
- Model size: Smallest model
- Requirements: For applications with speed and low cost requirements
- Input: Text and multimodal (combination of image and text)
- Output: Text
- Tasks: Classification and Labelling
Luminous-extended:
- Model size: Second-largest model
- Requirements: Faster and cheaper than Luminous-supreme
- Input: Text and multimodal (combination of image and text)
- Output: Text
- Tasks: Information extraction and Language simplification
Luminous-supreme:
- Model size: Largest Model
- Requirements: Expensive variant
- Input and Output: Text
- Tasks: All tasks of the smaller models, Well suited for creative text writing
Luminous-supreme-control:
- Model size: Largest Model
- Requirements: Most expensive variant
- Input and Output: Text
- Tasks: Optimized for Q&A and Natural Language Inference
You can try out the models of Aleph Alpha with predefined examples for free. Go to the Jumpstart page and click through the examples on Classification & Labelling, Generation, Information Extraction, Translation & Conversion and Multimodal. In the examples, you can already see how well the models work without logging in.
After testing the examples, you are familiar with the Luminous models work. Then you can sign up for testing your tasks. At the start, you get two credits for unrestricted access to all models.
Aleph Alpha also offers a Python API, which you can download as a Python package from pypi.org. On their website, they show how to use the Python client. Some code examples are provided. Furthermore, there is an HTTP API. In the next section, we will look at a few practical tasks.
Various Tasks
In this section, we look at some sample tasks from Jumpstart website.
Classification & Labelling
Luminous can classify the sentiment of a text message.
Prompt: Wow, you are so great at this!
Sentiment: Positive
You see Luminous recognizes the sentiment of the statement.
You can also use Luminous to assign sections of text to specific topics.
Prompt: Flowering plants grow fruits to spread their seeds. Edible fruits function symbiotically. Animals obtain nourishment from consuming the fruit and the animal’s movement helps spread the seed.
Topic: Biology
In this example, the model assigns the subject of biology, which is quite accurate. Or what do you think?
Generation
Do you need marketing material for a new product? Then you can roughly describe your product, and Luminous will create an advertising message for you.
Prompt: Nutty’s new nut bar is vegan and sugar free. It is made only of ecologically sourced ingredients. The production and delivery process is carbon-neutral. In addition, there are no additives.
Slogan: Let’s have a nutty time. #nutty #sugarless #vegan #sustainable #carbonneutral
Compact and accurate.
Do you need a name for your company? Then ask Luminous. Just write a short description of what you need a name for, and Luminous will make suggestions.
Prompt: A rustic bakery located in an old windmill offering locally sourced baked goods.
Names: Bakeshop Focacceria, Bakers Hill, The Artisan Way, Windmill Bakery & Cafe
These are very creative names, or what do you think?
Are you responsible for the job descriptions of your company? Then you can use Luminous as an assistant. Luminous can create job descriptions based on the job title. Have a look at the following example.
Prompt: Accountant
Job profile: YOUR PROFILE
Bachelor’s degree in Accounting or Economics
Minimum of 2 years experience in accounting
Excellent verbal and written communication skills in German and English
YOUR RESPONSIBILITIES
Prepare and process financial statements
Manage and maintain accounts
Analyze financial data and make recommendations on how to improve company performance
This example shows that AI can be very helpful in the future as an assistant or for brainstorming.
Multimodal
Luminous also supports images as input. For example, it can express thoughts about images.
Prompt:
Result: how the plane got into the forest.
Impressive. Luminous understands the context of the image and expresses thoughts about it. Wow! Moreover, Luminous could handle multimodal inputs even before GPT-4. Let’s do a performance test.
Luminous vs. ChatGPT
In this section, we compare the performance of Luminous and ChatGPT (GPT-3). We use a performance analysis carried out by Aleph Alpha. This section aims to reflect the key findings of this analysis.
The analysis was conducted in February 2023 and compares the performance of different LLMs. The comparison with Luminous is limited to the models BigScience BLOOM (176B), Meta AI OPT (175B) and OpenAI davinci (175B) (This model is below ChatGPT.).
In the figure above, the current best model of Aleph Alpha (luminous-supreme) is compared with the other models. For this purpose, 16 core tasks were carried out, consisting of five groups (Classification, Closed-Book Question Answering, Common-sense Reasoning, Natural Language Inference and Reading Comprehension).
Example prompts for the evaluation can be found in the performance comparison on the Luminous web page. The figure shows that the accuracy is competitive on average across all 16 tasks, especially given the smaller architecture of luminous-supreme compared to the other models.
Next, the Luminous team use an extended set of tasks. Now, they use 28 tasks for evaluation. Futhermore, the smaller models luminous-base (13B) and luminous-extended (30B) also included in the experiments. The following figure shows a comparison of the Luminous models with the model from OpenAI. Look at the results in the figure below.
We can see that performance on all task types assessed in the results improves with increasing model size. The result of luminous-supreme is impressive. The largest Luminous model (luminous-supreme) can keep up with the world’s leading AI language models and is much more efficient.
Conclusion
The results show that the European language model can compete with the currently highly celebrated ChatGPT. Aleph Alpha is a key building block for European AI research and shows that Europe can compete internationally. Everyone should be aware that there are alternatives to OpenAI’s models.
It is decisive which use case a company wants to solve with the models. Nevertheless, everyone should get an overview of current models so that you can select the right model for your use cases and requirements.
💡 Do you enjoy our content and want to read super-detailed articles about data science topics? If so, be sure to check out our premium offer!
Leave a comment