AI and infrastructure news
Filter by type: Post307
- 18 days ago x.com
Scaling LLM training: parallelism choices
A playbook published on April 6, 2026 describes parallelism strategies for training LLMs on GPU clusters, drawing on 4,000 experiments with up to 512 GPUs.
- 18 days ago x.com
URM combines recurrence and nonlinearity for reasoning
A post on Universal Transformer research attributes reasoning gains to recurrent inductive bias and strong nonlinearity, and describes design choices for local token mixing and training.
- 18 days ago x.com
VoxCPM: expressive speech without a tokenizer
Published on January 31, 2026, the post describes a speech synthesis system using continuous representations, zero-shot voice cloning, and expressive generation.
Awesome C++: curated list of C++ libraries and resources
The curated Awesome C++ list gathers C++ frameworks, libraries and resources in the awesome-list format, as a starting point for engineers to find tools. For companies, it is input for evaluating dependencies before adopting them.
- 18 days ago x.com
The Stack v3 expands code dataset to 5 trillion tokens
A post published on July 23, 2026 presents an open dataset with about 5 trillion deduplicated and filtered code tokens across 713 languages, excluding code with restrictive licenses.
AGI: definitions and benchmarks merit critical analysis
A publication dated October 16, 2025, discusses a paper defining AGI as the replication of human cognition. Its author welcomes discussion of varying capabilities but questions the benchmarks and the narrow view of AI.
- 18 days ago x.com
ACE: structured context that evolves with feedback
A post dated October 13, 2025 describes Agentic Context Engineering, an approach that incrementally accumulates and refines structured context using performance feedback as an alternative to iterative prompt rewriting.
ADAS: search and evaluation for agent design
Published on August 26, 2024, the paper describes an iterative method for proposing agents from domain information and evaluating the proposals.
Social agents: test generalization across scenarios
A post dated September 22, 2025 describes an MIT study that tunes prompts with small human-data sets and reports lower prediction error on new game variants.
- 18 days ago x.com
AHN proposes continuous memory for long-context LLMs
On October 9, 2025, a ByteDance publication introduced Artificial Hippocampus Networks, an architecture that continuously compresses information outside a language model's context window.
- 18 days ago x.com
Voice app adds support for custom models in five languages
A post dated November 11, 2024, says the e2-f5-tts Gradio app now supports custom F5-TTS finetunes for Spanish, Hungarian, German, Portuguese, and Greek.
Agent learning from its own interactions
A post published on October 10, 2025 examines using future states from an agent’s own interactions as supervision, without reward signals, to address long rollouts and rewards that are difficult to verify.
- 18 days ago x.com
256-bit arithmetic in CUDA GPU kernels
A post dated September 22, 2025, presents 256-bit multiplication and addition on a GPU and covers multiword arithmetic in CUDA kernels.
- 18 days ago x.com
Moonshine ASR joins the Transformers library
A post published on January 20, 2025 announces Moonshine’s integration into Transformers: the model uses Rotary Position Embeddings and accepts audio of any duration.
- 18 days ago x.com
AssistRAG separates memory from answer generation
Published on November 30, 2024, the work on AssistRAG describes a way to manage memory and external knowledge without retraining the main generation model.
- 18 days ago x.com
Attention and kernel regression: a useful connection
A 10-slide summary published on October 4, 2025 examines the relationship between kernel regression and attention, presenting this perspective as an aid to understanding the mechanism.
- 18 days ago x.com
Audio reasoning benchmark reports 92%
An Artificial Analysis report records a 92% score on Big Bench Audio for an audio model and compares time to first token with and without reasoning.
- 18 days ago x.com
Auto-RAG adapts iterative search to question difficulty
A post published on December 2, 2024 describes a method that plans searches and refines queries in multiple steps, adjusting the number of iterations to the difficulty of the question.
- 18 days ago x.com
Autoflow demonstrates Graph RAG with vector storage in TiDB
Published on November 22, 2024, the PingCAP repository describes Autoflow, a conversational Graph RAG knowledge base that uses TiDB Serverless for vector storage.
AutoToS uses tests and breadth-first search to refine functions
On August 23, 2024, an article described how AutoToS refines successor and goal-test functions using unit-test feedback and breadth-first search.
- 18 days ago x.com
Local vector search with SQLite and sqlite-vec
A publication dated August 2, 2024 presents a dependency-free C extension that adds vector search to SQLite.
- 18 days ago x.com
CCA: attention in a compressed latent space
A post dated October 7, 2025 describes an architecture that computes attention in a smaller latent space and reports lower memory, parameter, and operation counts in tests on H100.
- 18 days ago x.com
Model circuits can help predict reasoning errors
A post published on October 13, 2025 describes a method using attribution graphs to identify incorrect reasoning on arithmetic tasks, with improved detection metrics.
Evaluating accelerators for parallel workloads
In a post published on September 18, 2025, Graphcore describes the IPU architecture and its resources for graph-based computing and irregular, sparse workloads.
How to evaluate AI agents on workplace tasks
TheAgentCompany covers software development, project management, administration, and data science tasks to evaluate AI agents.
- 18 days ago x.com
Evaluating real-time speech inference
Published on September 18, 2024, the article presents Moshi, a 7.6-billion-parameter speech-to-speech model, and Mimi, a streaming audio codec.
- 18 days ago x.com
Developing high-performance CUDA matmul kernels
Published on September 29, 2025, the guide covers GPU memory hierarchy, CUDA programming, and PTX/SASS inspection to inform the development of matmul kernels.
- 18 days ago x.com
Balancing adaptation and memory in model training
A study published on October 5, 2025, reports that a model trained on 12,000 small tasks combined new ideas and notes distinct effects from structured and mixed lessons.
- 18 days ago x.com
How to observe information flow in LLMs
Published on October 9, 2025, the publication proposes three stages for information flow in LLMs and links massive activations to attention sinks and compression valleys.
Data curation and agent autonomy: what to test
A post published on September 23, 2025 describes LIMI: 78 selected demonstrations scored 73.5% on AgencyBench, outperforming models trained on 10,000 samples.
- 18 days ago x.com
Short reasoning hints for evaluating AI systems
A paper published on October 5, 2025 describes a hint generator and a solver trained with reinforcement learning. For mathematics, it reports 44% higher accuracy on AIME 2025 than RL with long chains.
- 18 days ago x.com
Dragon Hatchling explores local memory in language models
Published on October 3, 2025, the post describes an architecture with simple neurons and connections that store short-term memory through Hebbian learning.
- 18 days ago x.com
Dragon Hatchling proposes an interpretable LLM architecture
Published on October 1, 2025, the post describes an LLM architecture based on locally interacting neural particles and says it rivals GPT-2 performance.
- 18 days ago x.com
E2-F5-TTS adds an interface for voice-model tuning
On October 21, 2024, the E2-F5-TTS repository presented a Gradio app for tuning the model with user-supplied audio and automatically transcribing clips.
- 18 days ago x.com
Study compares attention in 13 LLMs, with 17% performing best
Published on October 12, 2025, the study reports experiments training 13 LLMs with different proportions of conventional attention and DeltaNet linear attention. The configuration with 17% attention—two of 12 layers—achieved the best result.
- 18 days ago x.com
Expanding queries with knowledge graphs
A publication dated November 13, 2024 describes a workflow combining entities extracted from queries, retrieved documents, and knowledge-graph relationships to generate contextual expansions.
- 18 days ago x.com
Extracting text from PDFs with olmOCR
In a post published on February 25, 2025, AllenAI introduced olmOCR, an open-source tool for extracting plain text from different types of PDFs that can run on a GPU.
- 18 days ago x.com
FireRedChat: full-duplex voice for assistants
A post dated October 7, 2025 describes a modular voice interaction system with cascade and semi-cascade modes, interruption handling, end-of-speech detection, and real-time responses.
- 18 days ago x.com
Baidu framework makes RAG more traceable
A Baidu post published on July 30, 2024 describes a framework that evaluates retrieved documents, selects evidence, and analyzes reasoning paths before answering.
- 18 days ago x.com
Using GEPA to detect malicious AI-generated code
A publication dated October 5, 2025 reports that prompts found with GEPA detected malicious behavior in AI-generated code, blocking 90% of malicious submissions with 1% of the audit budget.
Let's assess your project
I reply with scope and a proposal.