Should you wanted to boost the profile of your major tech company and had $10 million to spend, how would you spend it? On a Super Bowl ad? An F1 sponsorship?
You could spend it training a generative AI model. While not marketing in the standard sense, generative models are attention grabbers — and increasingly funnels to vendors’ bread-and-butter services.
See Databricks’ DBRX, a brand new generative AI model announced today akin to OpenAI’s GPT series and Google’s Gemini. Available on GitHub and the AI dev platform Hugging Face for research in addition to for business use, base (DBRX Base) and fine-tuned (DBRX Instruct) versions of DBRX could be run and tuned on public, custom or otherwise proprietary data.
“DBRX was trained to be useful and supply information on a wide range of topics,” Naveen Rao, VP of generative AI at Databricks, told TechCrunch in an interview. “DBRX has been optimized and tuned for English language usage, but is able to conversing and translating right into a wide range of languages, comparable to French, Spanish and German.”
Databricks describes DBRX as “open source” in an analogous vein as “open source” models like Meta’s Llama 2 and AI startup Mistral’s models. (It’s the topic of robust debate as as to if these models truly meet the definition of open source.)
Databricks says that it spent roughly $10 million and eight months training DBRX, which it claims (quoting from a press release) “outperform[s] all existing open source models on standard benchmarks.”
But — and here’s the marketing rub — it’s exceptionally hard to make use of DBRX unless you’re a Databricks customer.
That’s because, so as to run DBRX in the usual configuration, you wish a server or PC with no less than 4 Nvidia H100 GPUs. A single H100 costs hundreds of dollars — quite possibly more. That could be chump change to the common enterprise, but for a lot of developers and solopreneurs, it’s well beyond reach.
And there’s tremendous print in addition. Databricks says that firms with greater than 700 million energetic users will face “certain restrictions” comparable to Meta’s for Llama 2, and that each one users can have to comply with terms ensuring that they use DBRX “responsibly.” (Databricks hadn’t volunteered those terms’ specifics as of publication time.)
Databricks presents its Mosaic AI Foundation Model product because the managed solution to those roadblocks, which along with running DBRX and other models provides a training stack for fine-tuning DBRX on custom data. Customers can privately host DBRX using Databricks’ Model Serving offering, Rao suggested, or they’ll work with Databricks to deploy DBRX on the hardware of their selecting.
Rao added:
We’re focused on making the Databricks platform the most effective selection for customized model constructing, so ultimately the profit to Databricks is more users on our platform. DBRX is an illustration of our best-in-class pre-training and tuning platform, which customers can use to construct their very own models from scratch. It’s a simple way for purchasers to start with the Databricks Mosaic AI generative AI tools. And DBRX is extremely capable out-of-the-box and could be tuned for excellent performance on specific tasks at higher economics than large, closed models.
Databricks claims DBRX runs as much as 2x faster than Llama 2, partly because of its mixture of experts (MoE) architecture. MoE — which DBRX shares in common with Llama 2, Mistral’s newer models, and Google’s recently announced Gemini 1.5 Pro — mainly breaks down data processing tasks into multiple subtasks after which delegates these subtasks to smaller, specialized “expert” models.
Most MoE models have eight experts. DBRX has 16, which Databricks says improves quality.
Quality is relative, nonetheless.
While Databricks claims that DBRX outperforms Llama 2 and Mistral’s models on certain language understanding, programming, math and logic benchmarks, DBRX falls wanting arguably the leading generative AI model, OpenAI’s GPT-4, in most areas outside of area of interest use cases like database programming language generation.
Rao admits that DBRX has other limitations as well, namely that it — like all other generative AI models — can fall victim to “hallucinating” answers to queries despite Databricks’ work in safety testing and red teaming. Since the model was simply trained to associate words or phrases with certain concepts, if those associations aren’t totally accurate, its responses won’t at all times accurate.
Also, DBRX is just not multimodal, unlike some more moderen flagship generative AI models including Gemini. (It might only process and generate text, not images.) And we don’t know exactly what sources of information were used to coach it; Rao would only reveal that no Databricks customer data was utilized in training DBRX.
“We trained DBRX on a big set of information from a various range of sources,” he added. “We used open data sets that the community knows, loves and uses day by day.”
I asked Rao if any of the DBRX training data sets were copyrighted or licensed, or show obvious signs of biases (e.g. racial biases), but he didn’t answer directly, saying only, “We’ve been careful concerning the data used, and conducted red teaming exercises to enhance the model’s weaknesses.” Generative AI models generally tend to regurgitate training data, an major concern for business users of models trained on unlicensed, copyrighted or very clearly biased data. Within the worst-case scenario, a user could find yourself on the moral and legal hooks for unwittingly incorporating IP-infringing or biased work from a model into their projects.
Some firms training and releasing generative AI models offer policies covering the legal fees arising from possible infringement. Databricks doesn’t at present — Rao says that the corporate’s “exploring scenarios” under which it’d.
Given this and the opposite elements wherein DBRX misses the mark, the model looks like a troublesome sell to anyone but current or would-be Databricks customers. Databricks’ rivals in generative AI, including OpenAI, offer equally if no more compelling technologies at very competitive pricing. And many generative AI models come closer to the commonly understood definition of open source than DBRX.
Rao guarantees that Databricks will proceed to refine DBRX and release recent versions as the corporate’s Mosaic Labs R&D team — the team behind DBRX — investigates recent generative AI avenues.
“DBRX is pushing the open source model space forward and difficult future models to be built much more efficiently,” he said. “We’ll be releasing variants as we apply techniques to enhance output quality by way of reliability, safety and bias … We see the open model as a platform on which our customers can construct custom capabilities with our tools.”
Judging by where DBRX now stands relative to its peers, it’s an exceptionally long road ahead.
