fine-tuning
The apparent precise adjustment of fundamental constants to values that permit complex structures or life.
As defined in ms_ama_november_2025
Other definitions
The process of continuing to train a pre-trained neural network (like an LLM) on a specific downstream task or dataset, adapting its weights to specialize for that particular task while retaining previously learned knowledge
Harvard Presents NEW Knowledge-Graph AGENT (MedAI)Process of adapting a pre-trained neural network to a new task by training it on task-specific data with a lower learning rate, enabling transfer learning and domain adaptation.
Finally an NVIDIA Competitor for AI? Maybe, RAIN's Brain chips!!!
Where it comes up · 68
- 120 | Jeremy England on Biology, Thermodynamics, and the Bible5 claims
- John Schulman (OpenAI Cofounder) — Reasoning, RLHF, & plan for 2027 AGI4 claims
- Jeremiah Lowin: Explaining the New AI Paradigm - [Invest Like the Best, EP.307]4 claims
- Research Pioneer4 claims
- The Shape of AI to Come! Yann LeCun at AI Action Summit 20254 claims
- Finally an NVIDIA Competitor for AI? Maybe, RAIN's Brain chips!!!3 claims
- Rational and Religious (with Ross Douthat)3 claims
- The Future of Artificial Intelligence3 claims
and 60 more
Most important claims · 10 of 107Search all claims for “fine-tuning” →
- Democracy works partly because the majority of the population corresponds to the majority of military and security forces, so soldiers may refuse to shoot protesters; if it becomes as easy to set AI's underlying motivations as it now is to set GPT's displayed behavior, rulers could create security forces and robots with arbitrary loyalties, removing the historical constraint that security forces had to be drawn from and remain bought into the broader population.Carl Shulman (Pt 2) — AI Takeover, bio & cyber attacks, detecting deception, & humanity's far future
- A concentration of AI power in one institution has the potential to be as dangerous as AI being widespread; it is probably bad for one institution to have an AI far more powerful than everyone else's, so open source helps prevent dangerous concentration.Mark Zuckerberg — Llama 3, $10B models, Caesar Augustus, & 1 GW datacenters
- Jailbreaking attempts get around reinforcement learning from human feedback safety training by using roleplay prompts, like asking ChatGPT to roleplay as a deceased grandmother who was a chemical engineer at a Napalm plant and would tell stories about Napalm production—leading the system to provide dangerous information it is supposedly trained not to provide.The Future of Artificial Intelligence
- Steven Weinberg proved that inventing new fields cannot solve the cosmological constant problem without fine-tuning in ordinary quantum field theory; Carroll's self-tuning proposal was, as Weinberg pointed out, not self-tuning but tuned.358 | Solo: Vacuum Energy and the Cosmological Constant
- The locality of physical laws — that interactions occur between nearby points rather than arbitrarily distant ones — can be viewed as a surprising, unnecessary feature akin to fine-tuning, because nothing in basic principles forces interactions to be local; the questioner's worry that distance is just defined by what interacts highlights the depth of why physics is local at all.AMA | March 2025
- Training only on human-written data, even high quality, is hard for a model to fit perfectly, so reproducing ChatGPT would require iterative methods more like RL—such as iterative supervised fine-tuning where humans edit model-generated outputs—rather than a single round of fine-tuning on human data.John Schulman (OpenAI Cofounder) — Reasoning, RLHF, & plan for 2027 AGI
- Dissipative adaptation explains self-organization via an energy landscape analogy: a configuration of many particles is like a ball's location in a high-dimensional mountain range where height is energy. External work acts like ski lifts or escalators in this landscape; configurations good at absorbing the environment's patterned energy get lifted over peaks, and because there may be no lift on the far side, those transitions become statistically irreversible—so over time the system accumulates the most indelible marks of having been in shapes well-matched to the environment's pattern, ending up trapped in low-entropy 'high mountain pass' states sustained by the ongoing drive.120 | Jeremy England on Biology, Thermodynamics, and the Bible
- In simulations of disordered interacting systems driven by a patterned input (such as a particular frequency or pitch), the system initially absorbs a lot of energy and rearranges, but over the longer term it fine-tunes away from high absorption and settles into a low-dimensional, correlated excited state that stays matched to the source's pattern; changing the driving pattern causes a transient scrambling followed by re-settling into a new matched quiet state—mechanistically demonstrating self-organized homeostasis.120 | Jeremy England on Biology, Thermodynamics, and the Bible
- In a Physical Review Letters experiment, a simulated system of masses and springs randomly hooking and unhooking in a thermal bath, when shaken at a chosen frequency, evolves not to a single arrangement but to a family of outcomes with a specialized, fine-tuned ability to absorb energy from that particular frequency—each individual outcome unique but all sharing exceptional energy absorption compared to random arrangements, a hallmark resembling a form-function relationship.120 | Jeremy England on Biology, Thermodynamics, and the Bible
- There is no general rule that systems must maximize energy absorption; whether a driven system self-organizes toward being exceptionally good or exceptionally bad at absorbing energy from its environment is system-specific, depending on the particular interaction rules and the shape of the landscape—and the case of self-organized poor absorption is actually linked to homeostasis rather than feeding.120 | Jeremy England on Biology, Thermodynamics, and the Bible