AI won't replace jobs: it's just an efficient secretary

AI won't replace jobs: it's just an efficient secretary

Tech circles have buzzed for months with the idea that AI is unstoppable and will wipe out employment. But scratch beneath the surface, and a different reality emerges. Users describe large language models (LLMs) as "highly efficient secretaries," not replacements for skilled trades producing tangible results. This view has gained nuance as practical examples accumulate.

The pattern repeats across industries. In graphic design, professionals note tools cannot draw straight lines reliably, serving low-quality tasks rather than final client deliverables. In programming, users achieve complex Python scripts with 3D views without prior training, but credit lies in effective prompting. The consensus: AI accelerates experts; it does not substitute novices.

The problem of hallucinated errors

Reliability remains the primary concern. Experienced users report initial excitement fades when they discover systems invent data and echo desired answers. When corrected, models apologize and rewrite text, yet fail to distinguish knowledge from fabrication. This explains why natural language cannot replace programming languages: code specifies exact algorithmic tasks, while human language is inherently ambiguous. Translating vague instructions into executable code is where models struggle most.

Why Transformers hit a ceiling

Technical debates focus on architecture. Current models use feed-forward networks, processing sequences entirely without revisiting intermediate steps. When errors occur, they regenerate from scratch rather than editing specific sections. Human brains correct mistakes on the fly at minimal cost. Some suggest recurrent layers or hierarchical models could offer qualitative leaps. One user cites research on multi-frequency neural networks, warning such breakthroughs might reignite hype with uncertain outcomes. Others dismiss recent Hierarchical Reasoning Models (HRM) as vaporware, lacking real-world planning capabilities beyond puzzles like Sudoku.

The business fueling the bubble

Economic factors drive the debate. Critics argue monetary policy induces this boom, with companies spending millions on projects adding no customer value. When credit cycles tighten, survivors will gain advantage. Some firms cancel critical projects due to executive obsession with integrating AI everywhere. Infrastructure isn't limited to LLMs; data centers can update algorithms, and theoretical research pogre. Nvidia’s pledge to invest up to $100 billion in OpenAI and deploy 10 GW of systems signals massive stakes, with the first gigawatt operational by late 2026.

Sectors already feeling the impact

Not all are skeptical. Some claim AI is already consuming entire sectors, spreading widely within years. Graphic designers, video editors, photographers, and translators face immediate pressure. Autonomous driving works reliably in some markets, potentially displacing double-digit percentages of global GDP. The challenge is political, not technical: governments must decide whether to retrain displaced workers or protect votes. Conversely, many argue automation transforms rather than eliminates jobs. Routine tasks may be handled by AI under supervision. Countries reducing taxes and shifting labor to productive areas will lead; those failing to adapt risk becoming third-world economies by comparison.

---

Is the real question not when AGI arrives, but who pays for the party until then?