KW 29: OpenAI offers AI fine tuning for GPT-3, When the computer assigns grades, Expansion of Industry 4.0 – IT and AI for greater efficiency


OpenAI offers AI fine tuning for GPT-3: Starting in June 2020, OpenAI has been offering access to GPT-3 via an API. Hundreds of apps, including feedback software such as Viable, data collectors such as Airpaper, and the search engine Algolia, have relied on GPT-3 since the release of API access. Fine-tuning was not yet possible, which is why users have to program the desired output with examples via the input field. However, since the set of possible examples is limited, OpenAI already offered larger customers direct access to GPT-3 and the fine-tuning option included with it. In a new beta, customers can now also start ten free fine-tuning runs per month, with a limit of 2.5 million tokens on the size of the datasets. The new beta feature can be used to train better chatbots, for example.

When the computer assigns grades: The assessment of students‘ course work by artificial intelligence is still a mere vision of the future in Germany. However, its use could help students in a targeted way. For example, AI could set students tasks that correspond to their performance level. The German Research Center for Artificial Intelligence is therefore also working on AI projects in the education sector. However, this area, which harbors a high potential for discrimination, is still unregulated in many parts.

Expansion of Industry 4.0 – IT and AI for greater efficiency: Production robots and automated guided vehicles now take over many work processes, but can only increase productivity if they function accurately down to the millisecond. Sophisticated, adaptive computer systems are needed to ensure this. In this context, Industry 4.0 represents the interlocking of production and communications or information technology. This is intended to reduce manufacturing costs and production times, ensure greater transparency of supply chains and full automation from the order to the delivery of the product.

– Advertisement –
Blockchain – Ticker – With our weekly newsletter we provide you with the most important developments in blockchain technology and scene.

Remapping the customer: An increasing number of supermarkets and discounters are tracking customer flows and analyzing buying habits using technology from sensor startups. This makes it possible, for example, to measure how many customers enter and leave the store or which shelves are particularly frequented. In this way, internal processes can be optimized, service improved and customers understood more precisely. The startup Sensalytics, for example, already has almost 20,000 sensors in use at 360 companies.

Project on AI in emergency response: The research project „Communication & Information Platform for Resilient, Crisis-Relevant Supply Networks“ (ResKriVer) is intended to examine when artificial intelligence can best be used in the future in non-police emergency response in order to manage damage situations or prevent them from occurring in the first place. Specifically, this will involve the development of a digital platform through which information about the supply networks of crisis-relevant goods and resources can be collected, documented and analyzed. In addition, the platform will be used for communication between the population and crisis teams. Various scenarios such as blackouts, pandemics and extreme weather situations will be used as a starting point for development. The project is scheduled to run for three years and is funded by the German Federal Ministry for Economic Affairs and Energy.

Wind energy pilot project: AI and real-time sound measurements increase energy yield: The „Noise-Watchdog“ measurement and control system for recording and processing wind turbine sound immission promises more yield in the noise-reduced range through artificial intelligence and real-time sound measurements. It is to measure the sound immission of wind turbines in real time and automatically link the results obtained with the control of a wind farm. Sound-reduced operation can thus be optimized so that less yield is lost. The Munich-based engineering company IB Fischer CFD+engineering GmbH is working on Noise-Watchdog and is now looking for two suitable wind farms for testing under real conditions.

Strategy: New regulation for AI in the EU: What it means for insurance companies
Research: Location decision for AI campus further delayed
WHO: These are the 6 guidelines for AI in healthcare
Shopping: AI will massively change retail in these areas
Finance: How banks and insurance companies can benefit from AI

– Advertisement –
IoT – Ticker -The physical world meets the digital one. Internet of Things as an interface that revolutionizes both the industry and everyday life. Get a weekly update from the world of „Internet of Things“.


The University of Applied Sciences in Berlin receives 2.9 million euros from federal funding programs for artificial intelligence in research and teaching.


AI is never wrong? AI solutions possess so-called domain knowledge as a working basis, i.e. information about a specific subject area and the associated rules. In addition, they can interact with their environment, usually interfaces to other systems. In this context, many AI systems operate without communication to humans. Furthermore, it is significant that AI is capable of learning. According to a study, the most important goals that customers expect from the introduction of AI-supported systems include increasing sales growth, employee productivity, (positive) customer experience, profitability and business process efficiency. However, for practical application, the amount of data required is a major obstacle. Often, the data quality is not good enough to train AI systems.

AI creates sound: So-called stock music, i.e. short music excerpts that are mainly used on platforms such as Tiktok or Youtube, require people who can handle transitions and compositions. This is where the program Dynascore comes in. Its artificial intelligence takes over the composing and arranging, so that the music can be adapted and cut in sync with the video. All the user has to do is determine where to place pauses, transitions and crescendos, while Toll calculates the length and duration and ensures smooth transitions.


AI for automatic tumor detection: Together with the Dresden-based startup Asgen, Dresden University Hospital is testing artificial intelligence-based PAIKON software for diagnosing carcinomas in the breast and stomach with the aim of relieving medical staff and optimizing therapies. The examination of tumors is currently still carried out manually, but the procedure used to date is to be accelerated and made potentially more reliable by means of AI-based evaluation. Microscopic images of entire tumors can be analyzed by the software within minutes. This procedure, known as HER2-FISH analysis, is intended to investigate the expression of tumor markers relevant to breast and gastric tumors in order to generate information about suitable forms of therapy.


„Physicians need to be aware of the opportunities, but at the same time the limitations, of AI-assisted diagnostic tools.“
Prof. Dr. Tobias Raupach, Director of the Institute for Medical Didactics at the University Hospital Bonn, explains the background of the project „AI Campus – The Learning Platform for Artificial Intelligence“, which teaches prospective physicians how to use AI.


Electrode and AI: Paralyzed man answers questions via thought currents: Currently, people with paralysis who have lost the ability to speak usually rely on devices that use eye or head movements to spell out words one letter at a time. Some use a device that allows them to control a computer cursor with thoughts. Edward Chang, a neurosurgeon at the University of California, San Francisco, and his team wanted to find a better solution for the man, identified only as BRAVO1 to protect his privacy. The name refers to his status as the first patient in a study called BRAVO, or Brain-Computer Interface Restoration of Arm and Voice. BRAVO1, who is in his late 30s, has been paralyzed and unable to speak since he had a stroke 15 years ago, Chang says. Previously, Chang’s team had developed a system designed to recognize the brain signals associated with the intention to speak specific words. Tests showed that the system worked in people who were still able to move and speak. But success was far from certain in someone such as BRAVO1, Chang says. To find out, the team implanted sensors on the surface of the man’s brain. Then it had a computer study the patterns of electrical activity produced when he attempted to speak 50 different words. The process took months. Once BRAVO1 could reliably generate words on a computer screen, the team began having him form sentences. To help improve accuracy, the team added a program that analyzed the context of each word as it was added. After months of adjustments to the system, the man was able to generate a word reliably every four seconds, or roughly 15 words per minute.

Newsletter subscription

Subscribe to our free weekly newsletter for a compact overview of Artificial Intelligence topics:


More digital news briefings

Our political briefings