What happened
arXiv:2609.12162v1 Announce Type: new Abstract: Draft-verify-revise is a common LLM orchestration pattern for scaling inference-time compute. One LLM drafts, a second critiques the draft and provides feedback, and a third uses that feedback to revise the draft into the final output. As context cascades between stages, LLMs at different stages can resolve a context-dependent expression such as "previous" differently. When that happens, the expression undergoes a deictic shift, a change in what it refers to. This phenomenon was studied with a synthetic dataset of 10 base examples, each rendered in three conditions. Holding the shared components constant, the conditions varied whether the draft stage LLM (the assistant) or the verify stage LLM (the grader) resolved the expression correctly, and how much independent reasoning the revise stage LLM (the meta-evaluator) needed to determine which reading was correct. Six models from three providers were tested across 21 reasoning effort configurations using e-values for sequential testing, in a primary experiment and an ablation experiment that removed error classification labels from the grader's feedback. A separate LLM analyzed the meta-evaluator's stated rationale for each wrong verdict. Balanced accuracy (the unweighted mean of sensitivity and specificity) ranged from 0.156, below chance, to near-perfect. GPT-5.2 rose from 0.156 without reasoning to 0.942 at its highest reasoning effort level, while Gemini 3 Pro stayed above 0.94 at every level. Gemini 3 Pro at low reasoning effort outscored GPT-5.2 at xhigh reasoning effort for roughly 5% of the cost per trial. When the meta-evaluator erred, it tended to rely on surface cues rather than operational reasoning. Context engineers implementing draft-verify-revise pipelines should be wary of deictic shifts and make the intended referent explicit at each stage. Update (2026-09-17): Inverse Scaling in Test-Time Compute We investigate inverse scaling when LLMs have access to more test-time compute.. Inverse Scaling in Test-Time Compute We investigate inverse scaling when LLMs have access to more test-time compute. Update (2026-09-24): Make it so: Rapid and affordable plasmid sequencing on ONT platforms with PICARD-seq . Plasmid construction underpins molecular biology and synthetic biology, yet validation is often limited to the inserted fragment rather than the whole plasmid, and around a third of laboratory-made plasmids carry sequence errors that can affect function. Sanger sequencing scales poorly across whole plasmids, while short-read approaches cannot resolve the repeated DNA parts, such as promoters, that are common in synthetic constructs. We present PICARD-seq, a rapid nanopore-based protocol that uses off-the-shelf Tn5 rapid barcoding reagents and a MinION to sequence pools of whole plasmids in under a day, and we systematically benchmark the computational pipelines available for analysing the resulting data. Using a curated set of 25 plasmids of known sequence spanning 3.0-20.6 kbp, including various dilution series and repetitive multi-part constructs, we ran five independent replicates of each pipeline. The ONT EPI2ME Clone Validation workflow was fast (13-18 min) but stochastic, varying between replicates for both plasmids assembled and what sequence was returned; Canu o