Why your Fortune 500 company should start its own GPT team
*okay, technically an LLM ("large language model") team, but... Kleenex
It’s 2023 and a number of years ago, your large corporation realized that you need to have capable software people on staff, even though you’re not a tech company. Doing business, whatever your business is, requires people writing and maintaining code. You’ve spent time and effort to achieve a certain level of software competency, and are always looking for ways to get more value from what you’ve built.
Now all of the sudden there’s buzz around OpenAI’s ChatGPT, and how it’s going to change the world… somehow. How will this affect your company? You don’t know. Here’s why you should put a team on it:
You’re going to need advice about things like nuts and bolts (does it really cost $5 million to train these models?), what trends are happening in the technology space? (pair it with a search engine!), a realistic view of whether using GPT models will be something viable for your own company, and what things you might want to (or need to) do in-house Having a team on it is a valuable counterpoint to the other voices that you’ll hear - consultants (ahem), people trying to sell you services, etc. Microsoft and Google both have researchers on staff investigating new battery chemistry, not because they think they’ll make amazing breakthroughs, but because having expertise in-house is valuable.
These are some the issues you’ll be fielding sooner than you think:
Employees are gonna want to use the latest
toysproductivity tools. How exactly is this going to manifest itself? What should your policies be?Higher-ups are gonna want to know what’s going on with this new AI, and what their plan should be
The data scientists are going to tell you two things about artificial intelligence that you should take to heart
Machine Learning is good at interpolation, but questionable at extrapolation. In other words: the computer is good at things where it has lots of relevant examples to learn from, but is less trustworthy in areas it hasn’t seen before, so beware when you’re using these tools. When GPT has nothing relevant to draw on, it tends to produce believable fibs. Uh oh.
Machine learning needs clean, relevant data to be trained, and often lots of it
The product people are going to tell you two things about building useful stuff
The data scientists are right - we need to start identifying data sources that we already have (call transcripts, internal documentation, the more polite Slack messages), cleaning them up, and matching those up to potential use cases
Someone needs to track all the ideas regarding where in the company you might use GPT, and keep a list of (or a Slack channel for) all the teams that are currently experimenting with it
The team will give you quarterly updates on
Technology - what’s state of the art? Trends? What are competitors up to? Are there new prompt strategies emerging?
How is internal prototyping and experimentation going? How easy has it been to run these models in-house?
Potential internal use cases, and how you might make them happen
Build vs. buy? Could we use transfer learning to get a pre-trained model up to speed on our internal data?
Potential risks to watch out for - this includes human factors such as your employees going onto external websites typing things like “write a plan for integrating our company with Blue Sun Corporation after the secret merger that’s slated for next year but not announced to the public yet”, or dealing with the optics of using a robot to write emotionally sensitive content (oops)
How the heck are we going to assess the quality of the output of these models?
Are there things that we could build that might turn into a profit center if we rent our capabilities to other companies? Should we be filing patents to protect our right to practice?
What kind of training should we be giving our employees on GPT models?
Here’s what you’ll tell upper management:
Productivity will go up
Risk will also go up
Our competitors will definitely be using GPT
I’ve got a team on it