Google MedLM is a family of foundation models fine-tuned for the healthcare industry on Google ... [+]
Artificial Intelligence (AI) is changing. But let’s not forget where we’ve come from. The early notions of pseudo-sentient intelligence that percolated out of the mainframe labs of the 1950s may have too embryonic for the processing and storage power of the day. Although they may have given way to the ‘movie AI’ of the 1980s, we only started to see real progress in the post-millennial years and, IBM Watson gained its fair share (and more) of the attention in this space.
AI is now of course changing again and it’s not been hard to spot why. The rise of generative AI (gen-AI) drawing from Large Language Models (LLMs) running on vector databases has not been out of the tech newswires all year.
Sharper & refined AI tooling
But as we move into a new year and perhaps some of the furore and hype subsides, what happens next with AI is all about refinement and tooling i.e. where we go now is creating sharper language models that align to industry- or task- or function-specific jobs… and where we go now is all about creating sharper tools for software application development professionals to put new strains of AI into our applications.
Google has famously closed out a year of gen-AI hysteria with the launch of its Gemini Large Language Model.
Before we consider how Google is positioning Gemini to reflect current trends, let’s stop for just one nanosecond and remember what we’ve just said here i.e. the IT industry isn’t talking about a higher-level AI engine or model, the tech glitterati aren’t focusing on some new AI-enriched app that will order you a new pint of milk when the RFID-tagged carton in your refrigerator flags a best before date... and we’re not talking some new AI widget that’s going to surface on our smartphones. Instead, we’re getting excited about a new lower substrate-level data science approach that will percolate upwards to give us better AI. As we have said, AI is changing.
Fanfares aside, what we can see here is Google very much reflecting the need to sharpen and refine AI at this stage. Technologists want AI tools that can work to ingest any kind of data and work in the widest variety of post-deployment scenarios. Google knows this and it has built Gemini to be 'multi-modal' and be able to ingest information in text form, but also in the shape of images, audio and video.
Gemini triplets
While we normally think of Gemini pairs as a twin set in astrological terms at least, this Gemini has been shaped and scaled as a triple pack. By creating different versions of Gemini, Google says it will ‘efficiently run’ on everything from datacenter-level cloud deployments to mobile devices. To enable enterprise software application developers to be able to build and scale with AI, Gemini 1.0 has been optimized in three different sizes:
- Gemini Ultra: The largest and most powerful model for highly complex tasks.
- Gemini Pro: The model best suited for scaling across a wide range of tasks - to call it multi-purpose may be doing it a disservice, but you get the point.
- Gemini Nano: As the diminutive name suggests, the most efficient model for on-device tasks.
With real world software developer interests at the fore, the company now confirms that Gemini Pro is available via the Gemini API to developers in Google AI Studio, the company’s developer environment designed to allow programmers to integrate Gemini models via an Application Programming Interface (API) and develop prompts as they create code to build generative AI applications. It’s also available to enterprises through Google Cloud’s Vertex AI platform, as explained here.
Why is Gemini available via both routes? The API option via AI Studio is a free web-based developer tool designed to encourage usage and generate interest. Google says that when coders are ready for a fully managed AI platform, they can transition their AI Studio code to Vertex AI for additional customization and Google Cloud features, at a cost, there’s no such thing as a free AI lunch as we know.
Shaping AI for health
If the trend to shape and sharpen (and we can generally take scale as a given) AI currently is born out of Google’s work with these tools, we can see this in the introduction of MedLM, a family of foundation models fine-tuned for the healthcare industry, available to Google Cloud customers in the U.S. through Vertex AI, this technology will be more widely available next year.
The company is keen to show a friendly face as it attempts to encourage coders to get involved with its AI technologies by providing further tools and assistance. According to Google’s own AI blog, “Duet AI for Developers is now generally available. This always-on collaborator from Google Cloud offers AI-powered code and chat assistance to help users build applications inside their favorite code editor and software development lifecycle tools. It also streamlines running applications on Google Cloud — and Duet AI for Developers gives enterprises built-in support around privacy, security, and compliance requirements. We will be incorporating Gemini across our Duet AI portfolio over the next few weeks.”
What happens next, globally
While Google has reflected (some would say driven, some would say followed) the trends of the AI industry at large and worked to sharpen and shape AI from the way it ingests information to the way it can be applied, there are still (obviously) challenges ahead. While many of these technologies are available in all territories, Google rolls out in the US first and Europe (and the rest of the world) follows, so in terms of international deployment factors and perhaps governance, there’s a wider question there for the future.
We’ve called out the medical industry here, there is also work to deliver Google Duet AI in the Security Operations (SecOps) space and make generative AI generally available to defenders in a unified SecOps platform. That’s great for security teams, but there are many other technology engineers in a) the operations team and b) the wider IT department who will want to get in on the generative AI movement and be able to work concurrently (software parallelism pun intended) with their colleagues.
Artificial Intelligence is changing and it will continue to do so - although many think that this year of generative AI stands out as a seminal moment in time - let’s hope developers get the right tools and that we’re not hallucinating.
