How to Use AI Better: Breaking the Cycle of Bigger Models

The artificial intelligence industry is engaged in an increasingly expensive race to produce more powerful language models. More computing power, more infrastructure, more investment, more capability. The prevailing business model rewards technological escalation, while competition makes restraint commercially difficult. New capabilities become dependencies, and those dependencies generate requirements for maintenance, security, replacement and further investment. The resulting risks and failures become additional reasons to develop the next generation of technology. The system reproduces the conditions of its own expansion, while the question of whether we are learning to use what we already possess receives considerably less attention. We are developing the machinery of communication faster than we are developing our understanding of communication itself.

Language offers a different starting point. A system persists through transformations that reproduce the relational conditions necessary for its continued existence. Language is such a system, and effective interaction with a language model depends upon understanding its relational dynamics. Meaning is not simply deposited into a prompt and retrieved intact from a machine. It emerges through relations among words, contexts, expectations and histories, continually reconstructed through communication. Repetition establishes recognisable patterns, differences introduce variation, and timing determines how successive contributions interact. Understanding these processes allows us to achieve considerably more with existing technology without necessarily requiring a more powerful model. The critical development is not exclusively technological. It is also the development of our capacity to communicate.

This suggests a practical way to slow the AI arms race: shift attention from producing ever more powerful systems towards learning how to use existing ones. The prevailing commercial arrangement encourages us to interpret the limitations of current technology as reasons to purchase its successor. Yet some of those limitations concern how we establish context, maintain continuity, recognise errors and evaluate results. Language possesses a harmonic structure in which recurring sounds, words and relations establish patterns of expectation. Frequency shapes familiarity, differential timing establishes phase, and successive interactions reinforce, interfere with or displace what came before. Effective communication involves working with these relations rather than treating each prompt as an isolated command. There is a considerable difference between making a machine more capable and becoming more capable through its use. The latter does not necessarily require another data centre, although the former has acquired an impressive talent for requiring several.

The economic problem extends beyond speculative investment. New capabilities become infrastructure, infrastructure creates dependencies, and dependencies generate further requirements for maintenance, security, replacement and investment. Artificial intelligence is being incorporated into the informational and decision-making relations upon which other systems depend. The commercial structures supporting these technologies can fail long before the institutions using them can disentangle themselves. A speculative collapse would not simply remove the technology. It could leave behind unsupported software, abandoned services, diminished expertise and critical dependencies whose maintenance is no longer economically viable. The system’s expansion increases the consequences of its possible contraction. We are constructing a civilisation increasingly dependent upon technologies whose continued support is tied to commercial arrangements that have not demonstrated comparable durability.

The recursive structure is both the essence and the problem. A system persists through activity that reproduces the conditions necessary for its continuation, but persistence is not synonymous with growth, and neither guarantees sustainability. The AI industry turns its own consequences into opportunities for further development. New capabilities generate dependencies, dependencies produce vulnerabilities, and vulnerabilities create markets for additional technological intervention. Each successive response reinforces the pattern that produced the preceding problem. The system becomes entrained to its own expansion. None of this requires a conspiracy or a collection of spectacularly incompetent executives. Individually reasonable decisions can reproduce collectively destructive conditions. Increasing capability is not synonymous with increasing resilience. We have become remarkably proficient at manufacturing necessities, then treating their continued production as evidence of progress.

Human beings are not external operators of technological systems. We participate in the relations through which those systems emerge and persist. Our expectations, purchasing decisions, institutional practices and communicative habits help establish what becomes commercially necessary. Understanding language allows us to use existing models more effectively, while understanding systemic persistence allows us to question the economic conditions driving their continual expansion. The same recursive principle that explains the industry’s escalation also provides a means of changing its trajectory: transform the relations being reproduced, and the conditions of subsequent development change with them. This is not an argument against innovation. It is an argument against mistaking technological expansion for development itself. The next significant advance need not be another increase in computational power. It can emerge from changing how we communicate, evaluate and decide what is worth developing in the first place.

The technique is not simply to write better prompts, but to change the conditions under which subsequent responses become possible. Each exchange establishes distinctions, modifies expectations and alters the probability of what follows. Recurrence, semantic association and differential timing allow patterns to be reinforced, displaced or reorganised without losing continuity. We are not merely communicating information; we are using communication to transform the conditions of communication itself. The significance is that effective capability emerges through the interaction, rather than residing exclusively in either the user or the machine. This changes the economic question. If capability can be developed through the relations between existing systems, continual expansion of the systems themselves is no longer the only available path to greater capability.