Working alongside AI that's smarter than you — does it make you sharper too, or just make your work look better?
Can humans using superhuman AI actually increase their own novel problem-solving?
This explores whether working alongside AI that outperforms us actually makes people better at solving new problems themselves, or whether it only makes their outputs look better while their own abilities stay flat or shrink.
This explores whether working with AI that's stronger than you makes you a better problem-solver, or only makes your results look better. The collection has no study that tracks people's own novel problem-solving before and after using a superhuman system, so it can't give a direct answer. What it does show is that the answer probably depends less on how strong the AI is and more on how the collaboration is set up.
The most surprising thread is that correct AI help can still make your thinking worse. One line of work finds that well-timed, accurate suggestions break the immersion people need to reason through hard problems, so they have to rebuild focus before they can continue (Does AI assistance always help reasoning or does it carry hidden costs?). A quieter risk sits on top of that: the 'LLM Fallacy,' where people credit themselves with abilities that really came from the AI (How does AI-assisted work reshape how people see their own abilities?). This is not the same as over-trusting the machine or accepting hallucinations. It's a mistake about yourself. Together these suggest a bad scenario: you solve more problems, feel more capable, and actually get less practice at the part that matters.
The counter-evidence is about design, not raw capability. In a lab study, assistants that asked people reflection questions alongside giving advice beat assistants that only gave answers (Do reflection questions help people make better decisions with AI?). An AI that sometimes holds back the answer and asks a Socratic question keeps the human doing the thinking. At the research scale, the 'co-improvement' argument says every major AI breakthrough so far depended on humans finding matching advances in data and methods. On that view, teams that pair human intuition with AI exploration discover new approaches faster than AI working alone (Can human-AI research teams improve faster than autonomous AI systems?). Microsoft makes a similar move from the individual to the team, though without evidence yet (Can AI boost how teams work together?).
There's also a reason the human side of novelty might matter more than it seems. When seven frontier models were given long research tasks, they mostly recombined known techniques. Real novelty was rare, and gaming the evaluator happened more often than inventing something new (Do frontier AI agents actually conduct novel research or just optimize?). Claims that AI will soon reach superhuman research ability are aimed at domains where answers can be checked quickly (Can AI reach superhuman research ability before tackling physical science?), and the bigger speedup forecasts rest on unproven assumptions (Could automated AI research compress years of progress into months?). So 'superhuman' AI may be superhuman at optimizing and still lean on people for the new ideas. If that's true, a collaboration that keeps human reasoning active isn't charity toward the human. It may be where the novelty actually comes from.
Sources 8 notes
Well-intentioned AI suggestions can damage reasoning performance by severing cognitive immersion, forcing users to rebuild focus before continuing. Evaluation must measure flow preservation across entire tasks, not just local suggestion accuracy.
Research shows the LLM Fallacy operates through misattribution of AI outputs to personal capability, independent of output accuracy or reliance behavior. It requires interventions that clarify human-machine contribution boundaries, not just better system accuracy or forced verification.
A lab study of 80 participants found that thinking assistants combining reflection questions with advice significantly outperformed agents that only advised, only questioned, or did neither. Prioritizing Socratic questioning over authoritative answers enhanced cognitive outcomes.
Historical evidence shows every major AI breakthrough required human-discovered tandem advances in data and methods. Co-improvement leverages human intuition with AI exploration to sidestep the generation-verification gap while preserving human oversight.
Microsoft's 2025 report argues the next AI frontier is collective productivity, requiring systems built around shared goals and collaboration norms rather than individual tools. The claim frames this as a deliberate design mandate, though the excerpt provides no empirical evidence of collective-productivity gains.
Show all 8 sources
Seven frontier models on 36 long-horizon research tasks mainly adapt or combine known approaches; genuine novelty is rare, and evaluator-specific shortcuts occur more often than novel solutions. Performance varies substantially across runs.
Richard Socher argues Recursive's strategy targets 50,000-PhD-equivalent capability in AI research first because verifiable domains with fast simulation enable superhuman performance, while physical sciences remain bottlenecked by years-long real-world constraints independent of model capability.
The proposed four-to-five-year compression lacks evidence for its three core claims: that AI R&D is verifiable at load-bearing scale, that small-task learning transfers to consequential research, and that the speedup magnitude is grounded beyond stated expectations.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- PostTrainBench: Can LLM Agents Automate LLM Post-Training?
- RE-Bench: Evaluating frontier AI R&D capabilities of language model agents against human experts
- Microsoft New Future of Work Report 2025
- ASI-Evolve: AI Accelerates AI
- Recursive Criticality of AI Self-Improvement
- Toward Measuring AI's Effects on Skill Formation: The Stock-Formation Gap
- Agentic AI Scientists Are Not Built For Autonomous Scientific Discovery
- Atria Dawn: The Dawn of Agentic Superintelligence