Databricks Acquires MosaicML: A Leap Towards User-Customized AI
AI giant, Databricks, makes moves to to bolster its Large Language Model (LLM) offering through its acquisition of US-based generative AI start-up MosaicML.
Created on June 26|Last edited on June 26
Comment
Databricks, a San Francisco-based firm known for its advanced data management technologies, has signed an agreement to acquire MosaicML, a generative artificial intelligence (AI) startup. The deal, valued at approximately $1.3 billion, positions Databricks to cater to a booming demand from businesses aiming to build their own LLMs, and even “medium” language models (MLMs?).
Databricks and MosaicML
With its roots as a data storage and management startup, Databricks has made significant strides in developing AI-ready data management technology. Its latest acquisition, MosaicML, brings a unique value proposition to this ensemble. Since its launch in 2021, MosaicML has been committed to reducing the cost of employing generative AI. Where it traditionally cost tens of millions of dollars to develop a model, MosaicML has brought the price tag down to the hundreds of thousands.
As part of Databricks, MosaicML will continue to operate as an independent service, integrating its language-model platform with Databricks' existing technology. This will enable businesses to create affordable, customizable language models using proprietary data, a shift from the current practice of relying on third-party models trained on extensive publicly available data.
Privacy and Data Control
Data privacy concerns are increasingly influencing the adoption of AI technologies. Databricks CEO, Ali Ghodsi, points out that building a model from scratch provides clarity on the nature of the data being utilized. Pretrained models, conversely, often contain superfluous information gleaned from the internet, which can potentially skew results and raise security concerns. Therefore, the union of Databricks and MosaicML offers a more tailored, privacy-conscious alternative to the market's reliance on external vendors.
Smaller, Faster, Smarter
There is a growing interest in smaller models that can be trained more efficiently and run on devices with limited computational capabilities. Large language models, like the one that powers ChatGPT, may offer vast capabilities, but they can be overkill for many applications.
MosaicML's strategy aligns with this trend. The company is geared towards enabling businesses to harness AI's power at a fraction of the cost, using domain-specific models. This approach can lead to more practical, industry-specific results and streamline the process of incorporating AI into various business applications.
The Future
The question looms: are we nearing the performance limits of end-to-end LLMs? Rumors suggest that OpenAI's GPT-4 might be a mixture of specialized models, hinting at a potential plateau in the performance capabilities of broad-spectrum LLMs. LLMs are no doubt amazing and definitely seem to be a critical piece to the AGI puzzle, but also, it seems as though scaling will not be enough in order to achieve human-like functioning, and this will likely open the door for new research-focused companies that find new methods for building superior systems.
In conclusion, the Databricks-MosaicML merger is exciting for companies looking to build lightweight and customizable AI. It's likely that businesses will be increasingly turning to smaller, user-specific models over broad-spectrum ones, as this brings cost and customizability benefits, and there is most definitely a need to make the process of developing these models more efficient!
Add a comment
Tags: ML News
Iterate on AI agents and models faster. Try Weights & Biases today.