Context Compaction Theory
A new paper on arXiv introduces Context Compaction Theory, the first formal framework for analyzing how AI agents compress their state to fit within LLM context windows. The authors prove an equivalence between the Conte…
Full-text search across 9702 articles. Combine with topic and date filters; results sorted by relevance.
A new paper on arXiv introduces Context Compaction Theory, the first formal framework for analyzing how AI agents compress their state to fit within LLM context windows. The authors prove an equivalence between the Conte…
A new paper on arXiv (submitted 30 Jul 2026) introduces the Hollow-LLM Attack, showing that zero-knowledge (ZK) verification of large language model (LLM) inference can be fooled by dishonest providers who embed 'ghost w…
OpenAI's internal model Astra has solved ten open problems in mathematics and theoretical computer science, with proofs formalized as machine-checkable Lean certificates. The total compute cost was roughly $2,000 at API …
Researchers have developed a compiler-grounded hierarchical optimization framework for Triton kernels that achieves a geometric-mean speedup of 4.35× and a median speedup of 2.73× across 37 benchmark entries on Ascend NP…
San Jose, California, will hold its first public forum on August 19 to gather resident input on new regulations for data centers, responding to 11 pending projects totaling 1,630 megawatts of electricity demand. The city…
TONE3000, in partnership with Neural Amp Modeler creator Steve Atkinson, launched Architecture 2 (A2), an open-source amp modeling technology that runs on a $3 ARM Cortex-M7 600MHz chip and outperforms Neural DSP V2, IK …
Vermont Public Interest Research Group climate and energy program director Ben Edgerly Walsh argues that Gov. Phil Scott's veto of data center bill H.727 was unjustified, noting the bill passed both chambers with biparti…
The FFmpeg project thanked AI startup Cursor.ai on Sunday for contributing tokens to several developers, supporting their ongoing work on FFmpeg including development and code review. The amount of credits was not disclo…
A Coyote Media opinion piece proposes redirecting the $389 billion invested in San Francisco's AI industry since 2017 to fund Bay Area public transit, citing chronic delays and funding deficits. The article recalls a 196…
A new report from Unite.AI argues that financial institutions fail to see ROI from general-purpose AI tools because they lack workflow-specific integration, and that purpose-built AI tools embedded in specific workflows …
On May 20, 2026, OpenAI announced that an internal AI model had produced a counterexample to the unit distance problem, a conjecture posed by Paul Erdős in 1946, marking the first historically significant proof from an A…
In March 2016, Google DeepMind's AlphaGo defeated Lee Sedol, widely considered the world's strongest Go player, 4-1, a watershed moment for AI in the game. The cultural impact was initially muted, but by early 2018, the …
Mathematician Levent Alpöge announced on X that he used Anthropic's Claude Fable 5 AI model to find a counterexample to the Jacobian conjecture, a problem posed by Ott-Heinrich Keller in 1939. The discovery, which awaits…
A new agentic test-synthesis method called the 'Locksmith Loop' improved branch coverage in migrating legacy COBOL programs to Java, reaching 91.90% branch coverage on an internal production-like COBOL program and nearly…
A new empirical study of 123 frontend bugs in TorchDynamo, the default deep learning compiler frontend for PyTorch 2, has produced a taxonomy of 7 root cause categories and 15 subcategories, and used an LLM-aided methodo…
Danny Arends, a long-time D language developer, has built DLLM, a minimal coding agent in D that runs directly on llama.cpp without Python or bindings, using D's ImportC feature to access the C API with zero FFI overhead…
New Zealand's Green Party has proposed a one-year pause on new AI data centres, mirroring a moratorium in New York state, as part of the country's election campaign. Technology writer Peter Griffin explains that data cen…
Researchers introduced Beagle, a GPU-accelerated framework for genetic programming in symbolic regression, and benchmarked it on the Feynman Symbolic Regression dataset against CPU-based systems StackGP and PySR under th…
OpenAI published ten advances in mathematics and theoretical computer science, including results on sphere packing, non-sofic groups, and a counterexample to Connes' rigidity conjecture, with Lean certificates on GitHub …
Spiderbrain's MCP server exposes 28 tools for codebase work, organized into structural queries, memory tree access, and blueprint operations. The developer emphasizes a 'moment-based' dispatch pattern, where tools are ca…