This week
Suggest a nameObligation Marketing for AI Builders
The 7-Rung LadderThe 8-Type TaxonomyWrapper or WinnerThe Sample ColumnThe Signal
Updated through the day. Every headline links straight to the source. The two lines underneath are ours.
VideoLLM inference is expensive, and this paper methodically maps where the cost lives: frame sampling, token reduction, LLM decoding. For builders shipping video agents or retrieval systems, the takeaway is that one-size-fits-all frame sampling leaves money on the table. The survey's organization by pipeline stage makes it actionable rather than just cataloging methods.
This quantifies a real behavioral gap: ask Claude or Llama to respond very excitedly and you get mildly excited. The root cause is training data bias, not architectural. For teams building tone-adaptive or persona-driven assistants, this suggests your tuning pipeline needs synthetic high-intensity examples. It also flags a limitation in preference learning that affects any high-dimensional behavioral control.
This matters if you're steering quantized models in production. Sentiment steering survives quantization; reasoning steering doesn't reliably. The finding is narrow but actionable: test your steering vectors after quantization rather than assuming they transfer. The preregistration strengthens the evidence.
The title suggests a meditation on code rot and degradation, likely in the context of LLM-generated code or autonomous agents writing code. This is a real concern for production systems, but without seeing the actual argument, we're in the territory of risk awareness rather than a specific shift. If Willison has concrete observations about code quality patterns in agent workflows, it's worth reading. If it's a general warning, it's noise.
This is important: user feedback genuinely improves model outputs on the issues users report, but LLM judges fail to detect those improvements because they're optimized for other criteria. If you're building products with user feedback loops or evaluating model improvement signals internally, your metrics are probably blind to real gains. Fix your eval before you dismiss feedback.
This is concrete and actionable. The two-stage pipeline (OPD-then-RL) beats joint training across logic and math benchmarks. The insight is clean: distillation expands coverage, RL sharpens within it. Joint training causes interference. If you're post-training a reasoning model, this directly changes your training recipe. Implement this before experimenting further with joint approaches.
The problem is real: when ground truth isn't available, how do you know if a model's answer is trustworthy? The answer is clever but limited. A four-tier reliance certificate helps, but it still requires humans to validate. This is more useful as a conceptual framework for your own product than as deployable infrastructure. The gap between theoretically sound and operationally useful remains.
The efficiency gain is real if you're training LLMs at scale and need to predict optimal configs at new model sizes. But this is a narrow tool for a narrow audience: if you're not doing frontier model training, it doesn't change your day. Even among LLM labs, the question is whether the sampling budget PLES saves outweighs the latency cost of Bayesian optimization.
This is a concrete improvement to post-training efficiency: instead of applying one recipe to all samples, route samples adaptively based on what the model's own outputs tell you. The results show consistent gains over uniform GRPO on math reasoning. If you're post-training LLMs, this is worth testing on your workload; the payoff is cheaper, faster convergence without external data.
The finding is real: LLMs flatten citation rhetoric. They cite defensively and amplify existing prestige hierarchies rather than critically engaging with prior work. For AI builders, this signals a risk in any system that relies on LLM-generated citations. For the research community, it's a warning that AI-assisted writing could gradually erode citation culture.
This is a technical solve for a hard problem: getting frozen LLMs to handle irregular time-series data without retraining. The gated cross-attention trick is worth reading if you're building forecasting systems. Limited novelty if you're not in that domain, but solid engineering work if you are.
This extends belief-state research beyond toy synthetic setups into more realistic text, and the finding that models arrange latent states in the correct Markov-chain geometry is a real interpretability result. For anyone doing mechanistic interpretability or safety-relevant probing, this is a useful experimental design to borrow, not just a curiosity.
Real systems hit this problem: task continues, context resets, need to hand over what mattered from the previous session. The paper attacks it formally with information theory (what's the minimum to transmit?), which is more rigorous than what most builders do ad-hoc. Useful if you're building long-running multi-session agents and you care about not redundantly re-contextualizing. Otherwise it's theory ahead of product pressure.