Vela framework trains LLMs directly on real user behavior data
Source published ·Modelwire updated
Original coverage: arXiv cs.CL ↗·How Modelwire adds context
The development
Vela reframes how LLMs acquire user understanding by treating real behavioral data as direct training supervision rather than relying on static corpora or synthetic examples. This shifts the foundation of personalization systems: instead of inferring user intent post-hoc from conversation, models learn to predict and model preference evolution across months or years of actual platform activity. The approach matters because it addresses a structural gap in current LLM deployment, where systems must operate on rich temporal user histories but lack training regimes that encode that temporal dimension. For platforms and assistants managing long-term user relationships, this signals a path toward models that natively reason about behavioral trajectories rather than treating each interaction as stateless.
Modelwire’s AI-generated summary of coverage from arXiv cs.CL.
Modelwire analysis
ExplainerOur AI-generated reading of the wider context and the next developments to watch.
Vela's core contribution is inverting the training signal: instead of using behavior as a source for post-hoc inference, it uses behavior as the supervision target itself. This means the model learns to predict user preference trajectories during training, not after deployment.
This connects directly to the MIRROR work from October 7th, which argued that effective personalization requires preference internalization rather than surface-level imitation. Where MIRROR focused on how models internalize user intent through self-distillation, Vela addresses what data should feed that internalization process. The earlier context-filtering paper (October 10th) showed that long-term personalization only needs 5% of conversation history to work, but it assumed the model already knew what to retrieve. Vela tackles the prior problem: how to train models that natively understand which behavioral signals matter across time. Together, these three papers sketch a progression from 'what to train on' (Vela) to 'how to train it' (MIRROR) to 'how to deploy it efficiently' (the retrieval work).
If Vela's approach outperforms MIRROR-style fine-tuning on held-out user cohorts from the same platform (not synthetic benchmarks), that validates the claim that behavioral supervision generalizes better than preference distillation. If the authors release ablations showing that removing temporal structure from the training signal degrades performance, that confirms the temporal dimension is doing real work rather than just adding data volume.
This interpretation is generated from the summary above and the archive coverage cited below. Our methodology · Report an error
Coverage behind this analysis
These archive entries ground the connection in our analysis. They are ordered by source publication date, with links to our coverage and the original sources.
·arXiv cs.CL
Self-distillation framework shifts LLM personalization from imitation to preference learning
Researchers propose MIRROR, a self-distillation framework that reframes LLM personalization away from surface-level style matching toward deeper preference alignment. Rather than forcing models to reproduce reference text verbatim, the method trains models to internalize user preferences through on-policy reflection, where the model learns from its own generation patterns conditioned on user context. This shift addresses…
MentionsVela
How this coverage is produced
Modelwire uses AI to generate summaries and context from source headlines, snippets, and selected archive coverage. Automated checks do not verify every claim, and items are not routinely reviewed by a person before publication. Zacaria Solis operates the site. Read the linked source for the full evidence and report errors through our corrections process.
Modelwire summarizes, we don’t republish. arXiv cs.CL originally reported this story as “Vela: User Behavior as Supervision Sparks LLM Intelligence”. The full content lives on arxiv.org. If you’re a publisher and want a different summarization policy for your work, see our takedown page.