What’s happening?
Even if you don’t spend every day glued to The Register or Hacker News, it’s been hard to ignore the splash made over the past six months by the new generation of large language models in general and ChatGPT in particular. There’s been a range of responses, from ‘it’s just a statistical next word prediction tool’ to ‘is this the biggest advance since electricity or just since the internet’ to ‘let’s stop doing this before we produce Skynet’.
If you haven’t had much chance to play with it, you definitely should. Even if you have, it’s worth noting that most people will only have access to GPT3.5 rather than the paid GPT4. 3.5 was incredible, but 4 is as incredible again.
A nice illustration: 3.5 generated plausible enough text to come in the bottom 10% of the exam for American lawyers; 4 comes in the top 10% as reported by Forbes. I’ve gone from occasionally using it as an advanced spell checker to treating it as a super-intelligent assistant who’s expert on everything.
I sold my Google shares last week: ChatGPT is now my first step for any question and has cut my search engine use by 90%.
As director of a business that does technical work in digital transformation and data engineering, I get asked two things almost every day:
- Won’t this put you out of work?
- What’s next?
The pace of change has been so great (ChatGPT came out in November; GPT-4 was released in March) that it’s hard to see very far ahead. Still, I’ll go out on a limb and say two things:
- This is great news for businesses like ours, and terrible for the big body shops
- Personalising AI using existing techniques will give businesses that can move quickly enough a huge advantage, even without the need for new models
Writing code is the easy part
Agile project teams, like TCT, have always thrived on delivering tailored solutions. We hire people who excel at both coding and communication, possessing the ability to deeply understand a client’s business problems and deliver the right solutions.
Our solutions are mostly technology-based, but our emphasis is on the solution, not the technology.
Unlike large body shops that require precise, detailed specifications, we take a wider view. While these body shops may produce code at attractive prices, they often necessitate additional personnel like project managers and business analysts. Customers often tell us that they saved money by hiring body shops, but then spent more than the saving hiring business analysts, project managers and requirements writers.
GPT makes our advantage even greater: by making code generation free, it frees us up to think about what code should be generated, what systems are needed and ultimately what really solves a business problem.
If your business model is just typing out code, GPT is a threat. But at TCT we’re not just code typists: we’re engineers and problem solvers who deeply understand the issues at hand and apply the most effective technologies to address them.
The exciting part is that our deep partnerships, business understanding, and efficient problem-solving—will still be necessary. With lower-value tasks automated, we’ll be able to deliver even more, even faster.
What happens next?
“The future is already here – it’s just not evenly distributed.”
William GibsonSo, I’m very optimistic about the future. The next question is: what will happen next and how can TCT help our customers take advantage of it?
In terms of ChatGPT itself, the next big step forward will be plugins. Plugins enable ChatGPT to talk directly to APIs, say to book you a flight or restaurant table or even search the internet.
For our customers, the most exciting thing will be the ability to use the ChatGPT API and the latest generation of retrieval technologies to solve their business problems.
Would you like a customer service AI that can access all of your existing documentation? How about an intelligent assistant that reads stock market reports and alerts you to anything unusual? How about a virtual troubleshooter that reviews error reports and tells you where the problem in your workflow is?
These are all possible with existing techniques, combining language model APIs, vector databases and open source tooling - the early adopters of these new techniques will dominate their sectors while their competitors wonder how they’re doing it.
If you’d like to know more, please get in touch. We’d be delighted to help your business stay ahead.
How I wrote this article
It’s almost a cliche now to finish a ChatGPT article with a big reveal. “And you won’t believe it but ChatGPT wrote the last paragraph!”.
Actually, the process I used illustrates the point I’m making. I started this post by dictating my thoughts into an app, which used AI to turn them into text. I then did a manual edit, before asking ChatGPT to provide comments and restructure for me. I then did a full (manual) rewrite based on those suggestions, and iterated a few times.
So, all of the ideas and 70% of the text were written by me, but a lot of structure, polish and editing was done by the AI. This illustrates my point. This post could never have been written by ChatGPT on its own. I still had to think about the problem, implications, and what it means for our strategy. But ChatGPT did save a lot of the typing, a lot of the editing, and a lot of the structure. The same is going to happen with technology. The thinking will still be done by humans, but the typing and the editing can be done by the AI.