March 2026

Restricted Boltzmann Machines (RBMs) are typically trained using finite-length Gibbs chains under a fixed sampling temperature. This practice implicitly assumes that the stochastic regime remains valid as the energy landscape evolves during learning. We argue that this assumption can become structurally fragile under finite-time training dynamics. This fragility arises because, in nonconvex energy-based models, fixed-temperature finite-time training can generate admissible trajectories with effective-field amplification and conductance collapse. As a result, the Gibbs sampler may asymptotically freeze, the negative […]

Ver mais

The dominant narrative treats hallucination as a defect. And yes, in practical terms, it absolutely is. But framing hallucination purely as failure misses something more interesting happening underneath: a hallucination is also a diagnostic signal. It is the model drawing you a map of where the map runs out. The Cow That Broke My Model I’m a PhD candidate at Texas A&M researching transient environmental exposures in beef cattle. The short version: under certain conditions, agricultural microclimates can […]

Ver mais

AI made code generation cheap. But it also made silent breakage cheap. Large diffs. Subtle logic removals. Relaxed validations. Expensive cloud calls. Secrets slipping into logs. Most of this doesn’t get caught while generating code; it gets discovered in PR review, or worse, in production. That’s exactly where git-lrc changes the game. What developers actually gain Review at the right moment. Every staged diff is reviewed before it gets committed. No extra dashboard. No context switching. Faster understanding of AI-generated diffs. […]

Ver mais

Motivated by the prevalence and success of machine learning, a line of recent work has studied learning-augmented algorithms in the streaming model. These results have shown that for natural and practical oracles implemented with machine learning models, we can obtain streaming algorithms with improved space efficiency that are otherwise provably impossible. On the other hand, our understanding is much more limited when items are weighted unequally, for example, in the sliding-window model, where older data must be expunged […]

Ver mais

Ver mais

Since the first edition entitled “0 and 1 — From Elemental Math to Quantum AI” and released in early 2025, a lot of progress has been made. Fascinating new results have been uncovered and proved by the author, many still leading to interesting quantum dynamics. In 100 pages, the new material presented here goes far beyond any articles and books published so far on the topic. This second edition offers a trip deep into the most elusive and […]

Ver mais

I wanted to compare Vanilla DQN, DDQN and Dueling DDQN beyond just final reward, so I built a structured training and evaluation setup around LunarLander-v3. Instead of tracking only episode return, I monitored: • activation and gradient distributions • update-to-data ratios for optimizer diagnostics • action gap and Q-value dynamics • win rate with 95% CI intervals • generalization via human-prefix rollouts The strongest model (<9k params) achieves 98.4% win rate (±0.24%, 95% CI) across 10k seeds. The […]

Ver mais