How Replication Could Teach Machines What Good Science Looks Like — Edward Hughes

<p>Can a machine learn the judgement that separates a plausible-looking result from a faithful experiment? Edward Hughes, Chief Scientist and co-founder of Inherent, joins Tim Scarfe to argue that creativity is not optimisation, and that the missing capability in AI is choosing which questions are worth asking.</p><p><br></p><p>SPONSOR:</p><p>---</p><p>Cyber Fund built the Monastery to help founders ship products that were impossible a year ago.</p><p>Apply now: https://cyber.fund</p><p>---</p><p><br></p><p>Edward makes the case that Move 37 was innovative rather than creative, and that the field, not the individual, decides what counts as a discovery. That reframing runs through Csikszentmihalyi, Deutsch and exaptation into open-endedness, where deceptive goals and imperfect world models turn out to be the point rather than the problem. The second half turns to the paper: Replica, a task space built by redacting figures from real papers, and Faraday, a 27-billion-parameter model trained to steer a frontier coding agent that then beats the frontier on held-out replications.</p><p><br></p><p>---</p><p>TIMESTAMPS:</p><p>00:00:00 Cold open: Move 37, Faraday and collective intelligence</p><p>00:01:08 Sponsor: CyberFund</p><p>00:01:46 Inherent&#39;s $50M raise and the road from string theory</p><p>00:09:14 Three timescales of learning: weights, context, culture</p><p>00:13:47 Move 37 was innovative, not creative: the field decides</p><p>00:20:39 Creativity as satisficing: the urinal and evolution</p><p>00:25:06 Exaptation and the Tristan chord: creativity in context</p><p>00:30:56 Coherence for whom? Deutsch&#39;s hard-to-vary explanations</p><p>00:35:53 Why copying is creative: Deutsch and the constraint engineer</p><p>00:42:27 Societies of agents and the strong Moravec paradox</p><p>00:45:51 Evaluate in hindsight: from Lean proofs to climate change</p><p>00:51:56 Picbreeder, local goals and why discovery needs deception</p><p>00:57:21 Spaghetti proofs, translation layers and superhuman Go</p><p>01:00:37 Does nature compress? Naturalness and real patterns</p><p>01:07:36 Why replicate? Replica&#39;s redacted figures and Faraday</p><p>01:12:31 Faraday beats Codex, Claude and GLM 5.2 on held-out tasks</p><p>01:15:31 Replication to innovation: how the Transformer happened</p><p>01:18:26 Deep replication: what Faraday learns from Voyager and GNoME</p><p>01:23:37 Can the AI scientist cheat? Goodharting the judge</p><p>01:29:09 Inside Replica: scale-down, 8xB300 runs, per-task rubrics</p><p>01:34:11 The RL crisis: getting GRPO to work with per-turn credit</p><p>01:39:43 Weights vs harnesses: AlphaEvolve, DGM and EvoTune</p><p>01:45:45 The recursive company: agents cross a phase transition</p><p>01:50:35 Collective intelligence and the electric dynamo</p><p>01:55:46 What replaces OKRs? Incumbents and the burden of knowledge</p><p><br></p><p>---</p><p>REFERENCES:</p><p>MLST Creativity Article:</p><p>https://archive.mlst.ai/read/why-creativity-cannot-be-interpolated</p><p><br></p><p>organization:</p><p>[00:01:47] Inherent</p><p>https://inherentlabs.ai/</p><p>other:</p><p>[00:20:51] Marcel Duchamp, Fountain</p><p>https://www.tate.org.uk/art/artworks/duchamp-fountain-t07573</p><p>[00:05:19] Human-Timescale Adaptation in an Open-Ended Task Space (Adaptive Agent)</p><p>https://arxiv.org/abs/2301.07608</p><p>[00:06:05] The AI Scientist</p><p>https://arxiv.org/abs/2408.06292</p><p>[00:12:13] Training AI Scientists to Replicate Research (Replica and Faraday)</p><p>https://arxiv.org/abs/2608.13331</p><p>[01:44:46] Evolutionary Principles in Self-Referential Learning</p><p>https://people.idsia.ch/~juergen/diploma.html</p><p>[01:59:33] Are Ideas Getting Harder to Find?</p><p>https://www.nber.org/papers/w23782</p><p>book:</p><p>[00:16:04] Creativity: Flow</p><p>https://search.worldcat.org/title/254487436</p><p>[00:26:22] Why Greatness Cannot Be Planned</p><p>https://link.springer.com/book/10.1007/978-3-319-15524-1</p><p>[00:33:03] The Beginning of Inf