technocracy

Many quantum software development kits provide a suite of circuit optimisation passes. These passes have been highly optimised and tested in isolation. However, the order in which they are applied is left to the user, or else defined in general-purpose default pass sequences. While general-purpose sequences miss opportunities for optimisation which are particular to individual circuits, designing pass sequences bespoke to particular circuits requires exceptional knowledge about quantum circuit design and optimisation. Here we propose and demonstrate training […]

Ver mais

Working extensively with AWS credentials in Kubernetes this quarter revealed how often credential precedence causes configuration issues. While the AWS SDK’s credential chain is well-designed, understanding the priority order is crucial for production deployments. Here’s what I’ve learned. The Problem Nobody Talks About A recent incident illustrated this well: We configured IRSA for a microservice, validated it in staging, and deployed to production successfully. Two weeks later, an audit revealed the service was using broader IAM permissions than […]

Ver mais

There is a quiet failure mode that lives at the center of every AI-assisted coding workflow. You ask Claude Code, Cursor, or Windsurf to modify a function. The agent does it confidently, cleanly, and incorrectly — because it had no idea that 47 other functions depended on the return type it just changed. Breaking changes ship. The test suite screams. And you spend the next two hours untangling what the model should have known before it touched a […]

Ver mais

It’s been said whatever you can validate, you can automate. An AI that produces correct work 90% of the time could be very valuable, provided you have a way to identify the 10% of the cases where it is wrong. Often verifying a solution takes far less computation than finding a solution. Examples here. Validating AI output can be tricky since the results are plausible by construction, though not always correct. Consistency checks One way to validate output […]

Ver mais

One of the fundamental challenges in applying RL to discrete diffusion language models has been the intractable sequence level log likelihood computation. Unlike autoregressive models where you can decompose the probability chain rule style, diffusion models generate tokens in parallel across multiple denoising steps, making gradient estimation for policy optimization computationally prohibitive. The new LLaDA2.1 paper (arXiv:2602.08676v1) introduces ELBO based Block level Policy Optimization (EBPO) that I think deserves more attention from the RL community. Here’s the core […]

Ver mais

For years, the most valuable phase of company growth has been effectively invisible to everyday investors. By the time companies hit public markets, the asymmetric upside the kind that turns early believers into outliers is already priced in. What’s left is exposure, not opportunity. That’s not a bug in the system. It is the system. Enter WLTH, now out on Apple and Google The Shift: From Access as Privilege to Access as Product The idea behind WLTH is […]

Ver mais

arXiv:2602.07081v1 Announce Type: new Abstract: This paper introduces a generalized federated prompt-tuning framework for practical scenarios where local datasets are multi-modal and exhibit different distributional patterns of missing features at the input level. The proposed framework bridges the gap between federated learning and multi-modal prompt-tuning which have traditionally focused on either uni-modal or centralized data. A key challenge in this setting arises from the lack of semantic alignment between prompt instructions that encode similar distributional patterns of missing […]

Ver mais

arXiv:2603.23629v1 Announce Type: new Abstract: Code LLMs often default to particular programming languages and libraries under neutral prompts. We investigate whether these preferences are encoded as approximately linear directions in activation space that can be manipulated at inference time. Using a difference-in-means method, we estimate layer-wise steering vectors for five language/library pairs and add them to model hidden states during generation. Across three open-weight code LLMs, these interventions substantially increase generation toward the target ecosystem under neutral prompts […]

Ver mais