10 issues shipped 194 stories filtered every Saturday
The internet is noisy.
Your tech digest isn't.
Bowl of Data is a weekly tech newsletter powered by an AI‑pipeline that combs through hundreds of sources: security advisories, model releases, blockchain moves, and engineering deep‑dives, and serves only the stories that actually matter.
What's in the bowl
AI & Machine Learning
Model releases, research breakthroughs, and the real-world impact of large language models on products and people.
Cybersecurity
Vulnerabilities, exploits, threat intelligence, and what to patch before it becomes someone else's headline.
Blockchain & Crypto
Market moves, protocol upgrades, DeFi developments, and regulatory shifts worth paying attention to.
Software Engineering
Tools, frameworks, open-source releases, and developer ecosystem news that changes how we build things.
Latest issue
-
Weak-to-Strong Generalization via Direct On-Policy Distillation
This technical paper proposes a new distillation paradigm called Direct-OPD to improve weak-to-strong generalization in reasoning models. By distilling only the policy shift induced by reinforcement learning, the method avoids the capacity limitations inherent in traditional teacher-student imitation.
-
KVpop -- Key-Value Cache Compression with Predictive Online Pruning
This technical paper presents KVpop, a technique for compressing the KV cache in transformer models via predictive online pruning. It optimizes the computation of attention targets by reusing sparse log-normalizers and utilizing efficient data structures like Fenwick trees.
-
From RGB Generation to Dense Field Readout: Pixel-Space Dense Prediction with Text-to-Image Models
This paper proposes ReChannel, a novel architecture that transforms text-to-image models from RGB generators into efficient dense prediction engines. By treating transformer tokens as spatial carriers for task-specific data rather than RGB pixels, the method achieves new state-of-the-art performance with much higher computational efficiency.
-
The Key to Going Linear: Analysis-Driven Transformer Linearization
This paper presents a method for converting pretrained transformers into linear-time architectures by focusing on the efficiency of state update designs. By analyzing softmax attention through a first-order approximation, the authors prove that delta-style updates are superior for post hoc linearization.
-
DSpark: Confidence-Scheduled Speculative Decoding with Semi-Autoregressive Generation
This technical paper presents DSpark, a new approach to speculative decoding designed to enhance the speed of autoregressive generation. It contextualizes the work within the broader landscape of drafting architectures and parallel generation strategies.
+ 18 more articles in this issue