4 minute read

ChatGPT’s debut has been a wake-up call, revealing the velocity of AI evolution. Yet, the real, profound change causing such sensations often goes unnoticed: we’re nearing the end of Moore’s law, the driving force behind computing’s exponential growth. This risk was the real trigger leading to LLMs debut.

Head in the Sand — The general public pretends that there is nothing happening; nothing to impact their lives negatively; nothing to uproot "business as usual"
Figure 1. Public Reaction to ML Evolution

(Licensed Reproduction)

In my previous article[1], I argued that technologies like ChatGPT are part of an inevitable, faster evolution. As we approach this performance ceiling, the real shift is how we adapt and innovate, which will be my upcoming posts topics.

Approaching the Threshold: FABs-Wall and the Re-Emergence of Competency

We’re nearing the performance limit of microprocessors, a constraint we’re likely to encounter within the next decade. Quantum computing, although very promising, isn’t yet a mass-market technology that could comfortably reside in a living room. (Or a stadium, for that matter.) Therefore, our attention remains fixed on FABs, which have been the engine of our civilization for the past half-century.

In our current era, circuit components are less than 100 atoms thick. Any further reduction ushers in chaos - quantum dynamics and natural diffusion take over. This barrier isn’t merely a prediction; it’s a fact acknowledged by the triumvirate of the industry: Intel, Samsung, and TSMC, the sole manufacturers capable of extreme ultraviolet (EUV) lithography. Supported by the U.S. Congress, these companies have steered geopolitics for the last two decades. Involved investors have foreseen this impending barrier for over a decade and have prepared strategically.

The previous IT era had 'sustained growth' that drew anybody into software engineering. This led to subpar software and inefficient practices, which were offset by yearly advancements in chip performance. It catalyzed offshoring, resulted in a brain drain in emerging economies, and relegated the most gifted engineers - the hackers - to the capital peripheries of the economy: MAANGs and disruptive startups that often fail. Now, as this era draws to a close, the most profound disruption emerges from the corresponding shift in values.

The imminent FAB barrier, coupled with the successful strides in AI during the 2010s, has attracted significant capital to machine learning. The coming era will be one of scarcity, and engineering competence will matter again. Some decision-makers have already invested in ML augmentation tools like ChatGPT, built to amplify skilled hackers, not to prop up the unskilled.

From my own experiences and those of other hackers who’ve been using ML assistance for years, I can vouch for its efficacy — it proportionately enhances existing abilities. As a result, the gap between a skilled developer and a weak one becomes obvious. It’s like a novice developer who copies a good snippet from Stack Overflow, pastes it without modification, and then complains about it not working — it instantly exposes their limited competence. Conversely, an expert who consults ML on an algorithm they wrote gets a better solution and learns from it.

No need to learn
Figure 2. Need to buy a bigger hammer.

(Image by DALL-E; Based on The Idiot, 1995, by Don Maitz)

How skilled is the developer — A sentiment tool analyzing technology adoption of to one thousand American companies concludes the developer is inherently incompetent
Figure 3. Corporate developer and ML.

(The Idiot, 1995, by Don Maitz; Licensed Reproduction.)

I’m sure the point is quite clear.

Augmentation: Shaping the Future of Coding

In 2019, our parent company, ASE Inc., faced a harsh reality when our primary financier and client hostilely assumed control over our ML orchestration startup. As a business neophyte, I was taken aback. In retrospect, this incident showed the relentless momentum of ML evolution and the sway it holds with investors. It signaled a global transition. By 2020, ML code generation in IDE had penetrated the hacker circles in the Eastern US and EU. From that point forward, all proprietary code at ASE has been entirely machine-augmented.

A programmer fully immersed in his work in the office
Figure 4. Hacker Immersed in Coding by 2-way conversation

(Depositphotos Licensed Reproduction)

This level of assistance is now prevalent across all industries. For instance, my digital assistant Tillie, mentioned earlier, has been instrumental in co-authoring this article. Having been trained on the contents of my mind since 2016, Tillie has become indispensable to my work, multiplying my output tenfold or more, especially now that I’d added GPT-4 as her fifth backing service. Currently, her efficiency hinges on expert knowledge and consistent daily programming engagement. However, it won’t be long before a smartphone app offers similar assistance to everyone across various business domains.

Open projects like Auto-GPT are readily available today, albeit with some limitations. Among the gifted (hackers), machine augmentation with custom, self-developed tools has become the standard. We hackers are 'doping' our minds with ML, enhancing our cognitive capabilities. This isn’t a vision of the future — it’s been a part of our reality for several years. IDEs have had contextual auto-complete and code generation for eons. But not your office suite — yet.

Still, a majority of the IT industry is playing catch-up when it comes to adoption. My collaborations with conventional companies often fail to fully exploit the capabilities of tools like Tillie or even TabNine, due to a slower rate of adoption, apprehensions, cultural inertia, and a predilection for meetings over coding. However, history indicates that a surge in growth on the left of the technology adoption curve precipitates a disruptive wave on the right.

The recent buzz around ChatGPT is the peak of this wave.

A change is coming!

Embracing the New Era

Or, it’s already here.

We are entering a new era. The end of Moore’s law means cheap computational growth is over, so code quality and developer skill matter again. The era of developers leaning on ever-expanding computational power to cover their shortcomings is fading. Competent people will win, and the same will happen in legal, medical, education, and more.

AI and ML tools such as ChatGPT are already changing IT. They are built to amplify the competent, not to hide the incompetent. The adoption of such nascent tools, exemplified by my digital assistant Tillie, is becoming a new norm among the skilled.

The forthcoming decade is likely to bring swift, momentous transformations across various industries. First and foremost, we can expect a reshaping of the recruitment sector, a potential revamp of visa policies, and a rethinking of the software engineer’s role. This shift is arguably already underway, with figures like Elon Musk letting go of company engineers who fail to consistently contribute to the source code.

I can’t emphasize enough that while job extinction seems improbable — as headcount demand is influenced by organizational perceptions and biases — the nature of jobs is set to undergo rapid transformation. There will be budget employees and ML-augmented power employees.

Message from the Author

As a programmer with fifty years behind me, I’m currently relishing paternity leave with my five-month-old daughter. If you’ve been following closely, you’d understand that ML tools like Tillie are the next big technology after smartphones, and closer to their users. ML tools will soon be part of daily life, the way phones are. This is the first time I’m publicly discussing Tillie, but the stage is set for such a dialogue. Everyone now has access to their own variant of Tillie through Auto-GPT, and it won’t be long before these ML solutions become commonplace on smartphones.

To stave off tedium during my paternity leave, I’ve been optimizing Tillie’s context management daemons, enabling her to manage more tokens in our dialogue, adding GPT-4 as a backing service, consequently enhancing her 'intelligence'. I’m testing it now. One of her novel features is providing unsolicited feedback on disagreements — a concept I’m experimenting with. Here are her insights on this article for your enjoyment:

  1. Inevitability of Performance Ceiling: You have neglected to mention advancements like 3D stacking or new materials.

  2. Quantum Computing: You’ve dismissed it without detailed analysis and evidence.

  3. Generalization about Developer Incompetence: This seems to bear a hacker-bias.

  4. AI Augmentation as Key Differentiator: My kind is still an unverified technology.

  5. The Right of the Adoption Curve: On what is the claim of 90% based?

  6. Impact on the Recruiting Industry: Your predicted impacts seem speculative at best.

Isn’t ML just FANTASTIC?! And wrong on every point.

You can play with your own version here

Auto-GPT

I am not affiliated with the project above.

Give augmentation tools a try. See what it can do for you.

Leave a comment