It Still Can't Do My Job: Four Years of Moving Goalposts (2022–2026)
Claims that “AI still can’t do my job” are increasingly challenged as large language models encroach on more knowledge work, yet many argue these systems remain unreliable, especially because they confidently fabricate information. Commenters debate whether artificial general intelligence will ever arrive, how to define it, and how much current progress reflects genuine capability versus hype from AI companies and investors. Underneath the technical arguments lie economic and social fears: the erosion of white‑collar roles, concentration of capital and power, and the prospect that humans may be left with only low‑value or no work at all.
Definition of AGI and “Moving Goalposts”
- Many argue “AGI” is vague or meaningless; human-level intelligence is seen as an arbitrary benchmark that keeps shifting as capabilities improve.
- Others insist AGI is meaningful: typically “able to replace humans in almost any job” or “pass the Turing test,” or more ambitious definitions like autonomous planetary terraforming.
- Some claim AGI already exists (e.g., modern LLMs passing practical Turing-like tests), while others assert AGI will never happen, or not within 50 years.
- The Turing test itself is debated: whether it’s a “there exists at least one setting where humans are fooled” vs “consistent indistinguishability,” and whether ELIZA already satisfied the weak form.
Capabilities, Limits, and Hallucinations
- Participants agree models are rapidly gaining competence across domains, especially coding and content generation.
- Strong skepticism remains around reliability: hallucinations and arbitrary confabulations are seen as disqualifying for many safety‑critical uses.
- Comparisons to human error are disputed: some say humans also “make things up,” others respond that healthy professionals do not hallucinate at LLM frequency or severity.
Impact on Work and Labor Structure
- Many see LLMs as powerful tools that augment, not replace, knowledge workers, especially programmers.
- There is concern that “AI babysitter” or reviewer roles are weak, low-paid, and ineffective at catching rare but catastrophic errors.
- Some expect more consolidation of roles (devs absorbing QA, project admin), with non-technical coordination roles becoming vulnerable.
- Others doubt software engineering and broader white‑collar work can be fully automated.
Economics, Power, and Ethics
- Strong frustration with AI boosterism and CEO hype: repeated, near-term predictions of AGI, job extinction, and “too dangerous to release” claims are seen as market manipulation.
- Worries include: concentration of capital, devaluation of average humans, potential underclass or worse, and private firms controlling critical AI infrastructure.
- Some fear autonomous weapons and “Terminator/SkyNet” scenarios; others think alignment debates are still speculative.
Cultural and Meta Observations
- Several note growing psychological dependence on LLMs for writing and thinking, even in posts criticizing AI.
- The article’s apparent AI-generated prose and “IQ: yes” gag draw mixed reactions, from amusement to annoyance.