This is a guest post for the Computer Weekly Developer Network written by Rod Cope in his capacity as CTO at Perforce.
A company that has gone through various growth and development stages over the years, Perforce Software today describes itself as a recognized specialist in enterprise-grade DevOps tools. The organisation promises to future-proof competitive advantage by driving quality, security, compliance, collaboration and speed across the technology lifecycle.
Cope writes in full as follows…
A year ago, if I were to describe to you what we are currently seeing with modern AI — LLMs being the latest example and they certainly won’t be the last — it would have sounded like science fiction.
Yet, it’s happening right now, it’s completely usable: anyone can start using a LLM today, without any prior knowledge.
Announcements and introductions from Amazon, Google and others demonstrate why LLMs’ time is now.
While some people may be hesitant to jump in or unaware of the impact, this is not a fad that is going away. In all my 42 years of writing code, there has never been anything like this in software development: it is unprecedented.
What’s interesting about it — and important to understand — is that LLMs and AI in general are changing all the time: what is best today could be completely different or obsolete by this time next year, if not before. Amazon’s announcement at Re:Invent is an example: what the company announced effectively was a way to bypass an existing way of using AI.
This pace of evolution is both good and bad: while AI tools like LLMs take away a lot of the routine drudge for a developer, that also means putting in the time and effort to keep abreast of what’s happening in the marketplace. Is ignoring all that an option? Not unless you want to risk how you are working right now becoming rapidly obsolete and being left behind.
So where do you start?
Pick one of the top LLMs and start playing around with it: after all, this does not have to be a long-term commitment (though security and compliance are important – make sure to select a LLM where the team behind it has spent as much time to prevent misuse as they have done making the tool useful). Choosing a LLM is not a decision that is difficult to reverse. Technically, LLMs have light impact on technology infrastructure and are pretty much plug-and-play, so switching to a different one is simple (though depending on how they are being used, some work may have to be thrown away or redone).
Foundational models are a great springboard into LLMs, because they are essentially building blocks that take away a lot of the hard work. Nor do you need any previous AI or ML knowledge: modern AI like LLMs have completely leapfrogged previous examples of AI and ML. Previously, a human could eventually work out what those latter tools could, even if it would have taken 1000s of human years. This current generation of AI goes far beyond what the human brain can achieve: we are reaching the point where AI becomes a ‘black box’ and we do not know how it works.
Good old, traditional AI
There will, however, be a role for more traditional AI for a while yet, particularly where it would be more faster and less expensive to carry out in an explainable way. Applications that are functional safety related is another example, but ultimately that will change, when we reach a point where we trust AI more than the risk of human error.
First, compliance and security has to be prioritised. Prompt injections, for example, are a completely new type of digital threat: this is not about exploiting code loopholes, this is a hacker trying to trick AI into giving away something that it shouldn’t. AI is going to finder it much harder than a human to think that something suspicious is happening and thinking of all the ways in which it could be tricked is difficult. That said, I believe that we will see AI tools monitoring AI security and compliance: think of them as AI police officers. Users of LLMs and AI in general are also going to need to look to the quality of their own data, so adoption of data cleaning and similar tools are likely to grow.
In the meantime, expect to see continued momentum around LLM innovations and they are becoming an integral part of existing products. Microsoft and Google are already doing this (side note, so is Perforce) and the functionality will trickle out to everyone.
The big players like Salesforce have the depth and breadth of data required to power AI, but startups also have the agility to explore new ways of using this transformative technology.
Beyond general LLMs, vertical or user-specific LLM hubs could provide a great way to share knowledge across an industry.
For example, there could be a generic foundation for financial services companies and then on top, they could also have their own proprietary private space for their specific requirements. There are probably multiple other ways in which LLMs can accelerate all kinds of processes.
But LLMs are just the current buzz: they are very useful and evolving all the time, but they are just part of a massive new technology landscape powered by AI.
This is going to snowball and much faster than most of us can imagine, so now is the time to start becoming familiar with LLMs, while keeping an eye out for what is coming next.