by Dzidas Martinaitis

Categories

Lately, reading the news and following updates about advancements in AI, specifically in Generative AI and chatGPT, gave me mixed feelings - on one hand, we are on something big and impactful, but at the same time it feels like a potential threat to the future. And I’m not alone - NLP students lost their field of research overnight, meanwhile some orgs at FAANG became obsolete. It is an old news that chatGPT can pass a software developer tests at FAANG, an exam to become a lawyer or generate inspirational phrases for your YouTube shorts. But I’m sceptical that we will experience a radical transformation in a short time of a few years, but rather, it will be an iterative change which can take a decade or more. But as a story goes, a slowly boiled frog was too comfortable to jump out of a pot, the fate we shall avoid.

My scepticism grew since 2010, when we were promised self driven cars, tomorrow! Looking back, it felt then, that we just needed a bit, maybe a year or two, and Uber or Bolt will be driverless. Do you see it coming next year, two years? I’m less optimistic this time, giving it another decade or so. And more recently, with the birth of Stable Diffusion back in 2022, it felt like we are going to generate movies for ourselves, find business ideas based on recent trends or build product promotions from a single photo. Where are we today? Well, we can generate “artistic” content with limited applicability in the business context, at best. Sure, startups are burning midnight oil to come up with innovative products, but so far it didn’t change much.

In the defense of chatGPT or LLMs, I agree that it already has a impact - as a coding assistant, a translator, an initial knowledge bank or a sentence generator, but we still neeed that connector between the keyboard and the chair. But don’t forget that you already taxed with useless content what is generated in a bink of the eye, meanwhile AI wars are yesterday’s news. Marketing companies are heavily rely on auto-generated text as in the example with YouTube shorts, meanwhile social platforms deploy AI to recognise and ban such content, so the former parties now use another service to paraphase the auto-generated content.

Let me offer you a different angle against this doomsday outlook. Few years back, as a part of Amazon Cloud(AWS) organization, I worked with AWS customers to transform their businesses by employing machine learning solutions. My main take away from that experiece was that the business doesn’t care about lastest state of the art ML/Ai technique unless it gives a competive advantage. As a consequence, they are happily running a 20 years logistic regression model or a rule based system, which they call ML model to please shareholders or investors. As a personal anecdote, I was leading a team of engineers with the goal build a deep learning model for a computer vision problem. After 3 weeks of development, it became obvious, that the approach, favored by the sales team, gives 80% accuracy at best, meanwhile the customer was insisting on a human, 100% accuracy. So, we gave ourself a chance to look beyond a deep learning approach and sure enough, within few days we found a solution based on an old computer vision library. In the end, it was a rule based approach which ~20 lines of code and was 100% accurate. To put more salt on the AI wound, the cost of running it in a serverless enviroment was $3 versus $70K for a deep learning solution and no maintenaince at all. What a beauty, right?

So, my dear reader, how we will survive the future ruled by cruel and fearless AI? I would emphasize two things: by putting business ideas and challenges upfront as in the example above or in this short tweet by former CTO Oculus VR; and diversifying our skills. If today’s trend is any good for future prediction, then majority of AI innovations of today will be in hands of handful and we will be hapilly consuming them as cloud services. Meanwhile, we will be instrumental for the business to bridge the gap, therefore our skills need to be diversified, nevertheless adjacent. Below, you can find a suggestive list of actions to diversify the skills. Let me know if have a suggestion in the comments of a social platform or on Twitter.

  • If you are a AI/ML practitioner, embrace yourself and look beyond your field. For me, it was Bayesian statistics and causal inference. As with all religions, there is one Bible, which all followers must get acquainted. And to strenghted your belief, a very good YouTube course is provided as well.
  • If you are in an IT related field, the cloud knowledge is a must nowadays. ACloud.guru really helped me to learn about Amazon, Google and Azure clouds and pass a couple of certificates.
  • Don’t skip over Data Engineering which is tightly coupled with ML Operations. For the former, I would suggest “Fundamentals of Data Engineering” and for the latter - “Designing Machine Learning Systems”.
  • Incorporate AI advancements in your life - Copilot, ChartGPT and etc. There is no shame of working smarter, but be conscious of potencial information leak. To my surprise, some schools in Europe are encouriging students to use ChatGPT for writing.
  • Learn about internet marketing and keep eye on it in order to better understand your customers.