Nvidia Reportedly Bought a Synthetic Data Firm. So What’s Synthetic Data?
Chipmaker Nvidia is further leaning towards producing tools for generating AI developers and acquiring tools from synthetic data company Gretel for over $320 million, according to A’s data Reports from Wired Wednesday.
The move is because the generated AI companies work hard to find enough data to train and improve their models, thus increasing the need to generate data.
According to the report, Gretel’s employees will be folded into Nvidia. Gretel produces synthetic or simulated data for AI model training and will provide AI developers with NVIDIA’s products.
A NVIDIA spokesman declined to comment on the report.
Why is synthesis of data important
Training-generated AI models, such as OpenAI’s Chatgpt, a large language model, require a lot of data. Real-world data can cause problems for AI developers – that is, it can be noisy and not enough.
AI companies are violating restrictions on freely available training data, causing conflicts in whether they can use copyrighted content. Hundreds of actors, writers and directors Submit an open letter The Trump administration’s Office of Science and Technology Policy raised concerns about the use of copyrighted data. Currently, Openai is petitioning the government Give greater access to copyrighted materials Training AI models, otherwise American companies will be left behind by China.
Watch the following: Watch NVIDIA’s GTC 2025 Keynote: All Highlights in 16 Minutes
Synthetic data is also valuable in protecting private information. Gretel says its synthetic data Can be used to train models and tools without the need to expose sensitive or personal information – for example, healthcare data that does not identify an individual and may violate privacy laws.
have Concerns about the use of such data In model training. Excessive dependence on information that is not rooted in reality increases the possibility that the model will make things wrong. If the problem gets bad enough, it can lead to a problem called a model crash when the model becomes so inaccurate that it becomes useless.
