Crypto, Web3, and NFT are so last year. In 2023, it’s all about creating your own AI startup, and you don’t want to miss out! In fact, you could be making money in your sleep (Or so you’ve heard on YouTube, Instagram, and TikTok). Plus, why leave all that venture capital just lying around on the ground for any schmuck to take, when you could be that schmuck?
With that wisdom in mind, here are some surefire ways to (not) get your AI startup off the ground.
1) Pick the wrong problem to solve with AI
Make sure the problem your AI is trying to solve is so simple, it could be solved without AI. Why not create a sophisticated deep learning model to determine if a number is even or odd? That’ll save your customers the hassle of doing basic arithmetic.
2) Make sure your AI model is set in stone
GPT-3 is the only AI model that is going to exist for the next decade. Make sure to build your whole application around that assumption, because flexibility won’t be required in this glacially slow-paced age of AI. Wait, what do you mean we’re a version behind?
3) Ignore ready-made services
The best move when making your own AI startup is to train all your models from scratch, even when pre-trained models are available that can do the same thing (e.g checking SageMaker Jumpstart, Azure ML or Vertex AI). You should totally train a new language model yourself without sufficient computational resources or expertise. This certainly won’t lead to a subpar model.
4) Don’t leverage Generative AI services from cloud providers
Sure, you could use tools like Amazon Bedrock, which provides API access to pre-trained models to you as a managed service, so you don’t have to worry about managing infrastructure. But why not do the opposite, and deal with all that overhead yourself in-house? Also, make sure to build your custom models without sufficient data, unlike the foundational models you could have gotten from something like Amazon Titan.
5) Don’t optimize your compute power at all
You should use general-purpose GPUs for both your training and inference tasks, because your money and time is disposable. Don’t use any customized chips optimized for machine learning like Tranium for training or Inferentia for inference, Intel AMX or Tensor Processing Units (TPUs) for AI workloads. After all, those weren’t made to help you out at all.
6) Ignore all recent developments
As I said before, the field of AI is glacial, so you’ve got ages before the ship you’re sailing runs into that iceberg over there. Feel free to take your hands off the wheel and ignore learning about all new services that come up that could help your fledgling business stay ahead of the pack.
… Poe’s Law means I need to tell you I’m being sarcastic, now
[ Insert disclaimer not to do any of these things so we don’t get sued. Make sure to remove this note ~ The Editor ]
In all seriousness, it is an exciting time in the world of AI, and if you’ve got a great idea for an AI startup, go for it! But make sure you’re solving a problem with AI that should be solved with AI, and you’re taking advantage of all the many services available to you.
One of the advantages of everyone being so keen to leverage AI right now is there’s tools out there you can use, and not using them would put you seriously behind your potential competition, who will use these tools to get their startups off the ground.
Courses on AI/ML to check out
Of course, if you want to both get started in the AI space and leverage those tools, it helps if you take some courses on it. Here are some resources you might find helpful.
Cloud Agnostic
AWS
Azure
GCP
