Diffblue CTO Peter Schrammel on AI, trust, and the beauty of code
Welcome to The Diff, a new newsletter for the LinkedIn dev community covering the latest news and thinking around AI-augmented software development, or 'AI for code'. Brought to you by the engineering team at?Diffblue.?
Each fortnight we talk to a member of the team at Diffblue to discuss changes in the world of AI for code, as well as current talking points in the engineering world, and also explore their own journeys. Share your views in the comments below, and if you have questions, feel free to ask.?
Peter Schrammel is Cofounder and CTO of Diffblue and a former assistant professor of computer science at the University of Sussex. In this issue, he talks about AI trust, transitioning from academia to business, and what links classical music with writing code.
The last two years have brought AI right to the forefront of discussions about the future of tech, and prompted speculation about what AI will be able to do - and what it won’t be able to do. When it comes to AI for code, what do you think is the future of AI? And what does ‘better’ AI look like to you? More generalist or more specific, autonomous or assistive?
It is a question of how much you can trust the AI to accomplish a certain task. It is like delegating versus micromanaging: the more I trust that the outcome will be what I need, the more I can delegate and just check the outcome.?
While much of the focus during the recent AI boom have been generalist models, they require a great deal of interaction. It is specialist AIs that are more likely to achieve the results to a quality that allows the user to build trust and start delegating bigger and bigger tasks.?
As the tools become better we’ll eventually just trust them, instead of scrutinising the outcome in every detail.?
The programming world has been through this before. For example, in the 1970s, compilers translating computer code written in a high-level language to a low-level one became more widespread. Initially, developers were suspicious as to whether a programme would be able to translate code as efficiently as a human.?
Now, there is little doubt that compilers are more efficient than any human programmer and able to operate at virtually no cost.?
There may be some rare bugs in compilers, but even that doesn't? cause much concern even for mission-critical code. There is effectively 100% trust in what compilers do.?
AI tools will perform specific tasks autonomously and the results will be sufficiently good that we will lightheartedly take the risk of not questioning them. We’ll know that we wouldn’t be able to do a better job ourselves.
Where do you see Diffblue in five years??
I would like to provide a tool to users that makes their lives easier - so that they can focus on the other things that they enjoy doing. The main premise of Diffblue Cover is to automate away all the tedious stuff that nobody enjoys doing, so that they can focus on the more interesting bits.?
Your background is in academia rather than business. How have you found that transition??
There is a lot that I had to learn. Initially, you think that technology is everything: because that’s where all your blood and sweat has gone, into making it work. But the change in thinking is that the product needs to deliver value to the user - which is a completely different thing.?
You're usually coming with a solution that is looking for a problem - which is just not the right approach. Technology usually solves a particular problem, and the problem might not be the problem that the user is having.?
You need to listen to the user and then use the technology to solve their problems.
Do you have any advice for those wanting to make a similar transition??
I think you can avoid mistakes if you have somebody on board that has done it before. There are a lot of things that are completely obvious, but if you’ve never done it you fall into every trap. If everybody on your founding team is a technologist, then you're just lacking the product experience. Also the commercial experience, and the experience in building a team.
As well as a scientist you are a composer of classical music. Are there any similarities to writing music and writing code??
Yes, there are similarities. If you look from a theory perspective, it's interesting that the principles of music are derived from fundamental physical principles of the world, and are expressed by mathematical concepts. If you think about rhythm or harmony, melody, they are all expressed by numbers. There is a beauty, an elegance, to code in the same way as there is in music.?
领英推荐
There is a beauty, an elegance, to code in the same way as there is in music.?
But the difference is the ability music has to move you. You can have software that frustrates you, even that makes you angry, but the emotional arc of a piece of music and the way it raises emotions, you can’t do that with code.
Top stories from the world of AI for code
Each fortnight we look at what the Dev world is talking about. Share your views in the comment section below.
By 2027, India is expected to surpass the US as the world’s largest developer community on GitHub, the Microsoft-owned code-sharing platform.
How artificial intelligence is poised to revolutionise the software industry, offering unprecedented opportunities for growth and innovation whilst simultaneously enhancing cybersecurity measures.
A cornerstone of modern software development, DevOps testing is playing a vital role in ensuring software quality within the fast-paced world of continuous delivery.
Thanks to TuringBots (AI and generative AI for software development), software development is on the cusp of a transformative change.
The Diff's quote of the week
Generative AI has been one of the hottest trends in recent memory, and, as the hype subsides, IT organizations are looking to deliver on its promise.
That's it for this week's edition of The Diff. Keep up to date on what is going on the website or get in touch in the comments below.
About Diffblue - Accelerate development with AI unit test generation.
Diffblue uses game changing AI technology to fundamentally transform the way developers write code. Our goal is to empower dev teams to be wildly more productive by automating tedious and time-consuming parts of the software development life cycle (SDLC)? so that they focus on building products that people use and love.
Diffblue Cover is the only autonomous AI-powered Java and Kotlin unit test generation solution that generates reliable unit regression tests at scale — locally and in CI.
All the reward, and none of the risk. Unlike LLMs or code completion tools, our technology uses reinforcement learning to generate code that is guaranteed to run, compile and be correct — every time. Plus, we operate on-prem so your code stays within your own environment, never seen and never shared.
Visit the Diffblue website for more info: https://www.diffblue.com/
As we edge closer to trusting AI to generate code autonomously, how prepared are we in terms of data preparation and quality assurance? Looking forward to hearing your thoughts!
Head of Marketing @ Diffblue | Former BasisAI, AWS, Twitter | AI & Developer Tools
5 个月Nice Peter Schrammel ??