AI and infrastructure news
Filter by type: Post307
- 18 days ago x.com
Graphiti: temporal graphs for evolving data
Published on December 10, 2024, the Graphiti project describes dynamic knowledge graphs that represent relationships between entities over time and offer multiple search methods.
- 18 days ago x.com
Grokking in modular addition: three stages to generalization
A post dated October 4, 2025 describes a paper proposing three stages of grokking in modular addition and a data threshold near n log n, with weight decay and moderate width supporting feature learning.
High Bandwidth Flash: what to assess before 2026
In an article published on October 12, 2025, SanDisk proposes NAND-based memory with 8 to 16 times the capacity of HBM. First samples are expected in the second half of 2026.
- 18 days ago x.com
HijackRAG highlights risks in RAG systems
Published on December 1, 2024, the post describes an attack that uses injected texts, attention diversion, and instructions to manipulate RAG system outputs.
- 18 days ago x.com
Concurrent Llama 3.1 8B inference on an RTX 3090
A Backprop post published on August 16, 2024 reports reasonable performance in a benchmark with more than 100 concurrent requests on a single GPU.
- 18 days ago x.com
Layered inference on memory-limited GPUs
On September 25, 2024, the AirLLM publication described inference that loads one Transformer layer at a time from disk, as well as block quantization and compression support.
- 18 days ago x.com
Jamba 1.5: 256K context and inference evaluation
On August 22, 2024, AI21 Labs announced the Jamba 1.5 Mini and Large models, combining Mamba and Joint Attention, with a 256K-token context window and features including ExpertInt8 and JSON output.
- 18 days ago x.com
JEPA-SCORE estimates density with JEPA encoders
A study published on October 8, 2025, presents a way to estimate sample probabilities using an encoder’s Jacobian, without retraining.
- 18 days ago x.com
LazyGraphRAG combines vector and graph search
A post published on November 29, 2024 describes an approach combining vector RAG with breadth-first and best-first iterative search, without index summarization.
- 18 days ago x.com
LoRA adapts an MLP with about 9,000 trainable parameters
Published on July 30, 2024, the example applies LoRA to a 5-million-parameter MLP for classification and compares performance with the adapter enabled and disabled.
- 18 days ago x.com
LoRA for reinforcement learning fine-tuning
A Thinky Machines post published on September 29, 2025 describes LoRA configurations for reinforcement learning fine-tuning and claims rank 1 can match full fine-tuning performance when configured well.
- 18 days ago x.com
Memory in AI agents: criteria for long-term projects
A survey published on August 11, 2025, by researchers at Renmin University of China and Huawei covers the design, evaluation, applications, limitations, and future directions of memory in LLM-based agents.
- 18 days ago x.com
Mini-Omni-Reasoner combines reasoning and speech
A September 22, 2025 post describes an architecture that alternates silent reasoning with spoken tokens, alongside a new dataset and reported gains in arithmetic reasoning and contextual understanding.
- 18 days ago x.com
What attention sinks reveal about LLMs
Experiments on models from 410 million to 120 billion parameters link extreme activation norms of the initial token to compression valleys and attention sinks.
- 18 days ago x.com
Parler-TTS: evaluating controllable voice models
Published on August 8, 2024, the announcement presents Mini and Large v1, models trained on 45,000 hours of speech, with voice controls set through text prompts.
- 18 days ago x.com
PEER retrieves over one million experts using product keys
The PEER architecture combines product-key routing and single-neuron MLP experts to sparsely select the components used for each input.
Research as configurable code: NVIDIA’s approach
On September 7, 2025, a NVIDIA post described a system that turns natural-language research strategies into executable functions, using search, interchangeable language models, and sandboxed execution.
- 18 days ago x.com
PHLoRA extracts LoRA adapters from fine-tuned models
Published on September 17, 2025, the post describes a method for extracting low-rank adapters from fine-tuned checkpoints without training data or gradients.
- 18 days ago x.com
Piper offers local multilingual neural speech synthesis
Published on January 7, 2025, the text describes Piper, a local system that converts text to speech with multiple voices, languages, and customizable options.
- 18 days ago x.com
Multi-hop RAG with a graph planned for each query
A post published on August 12, 2025 describes a method that plans dependencies among subproblems for each query and reports outperforming GraphRAG systems on three multi-hop question-answering datasets.
- 18 days ago x.com
RARe adapts retrieval models with retrieved examples
Published on December 1, 2024, the paper presents RARe, a method that uses BM25-retrieved query-document pairs as examples to guide retrieval models.
- 18 days ago x.com
REFRAG speeds up RAG with reusable embeddings
A paper published on September 15, 2025 presents REFRAG, a method that replaces most retrieved tokens with precomputed chunk embeddings. It reports 30× higher speed and support for 16× longer contexts without loss of accuracy.
- 18 days ago x.com
RLAD explores reasoning abstractions with reinforcement learning
Published on October 3, 2025, the paper presents RLAD, a framework using two-player reinforcement learning to help language models discover procedural hints in natural language.
- 18 days ago x.com
RLMs: an approach to very long prompts
Published on October 15, 2025, the publication describes language models that interact recursively with long prompts through a REPL and reports evaluations on the OOLONG and BrowseComp-Plus benchmarks.
- 18 days ago x.com
SimpleQA: measuring factual accuracy in AI responses
Published on October 30, 2024, the paper describes a benchmark of 4,000 factual questions and verified reference answers for evaluating models.
- 18 days ago x.com
SWE-Swiss: a training recipe for software engineering
On August 12, 2025, researchers from Peking University, ByteDance Seed, and HKU presented a training recipe for software engineering tasks. The researchers report that SWE-Swiss-32B scored 60.2% on SWE-bench Verified.
- 18 days ago x.com
SwiReasoning switches reasoning modes to reduce tokens
A publication dated October 12, 2025 describes a method that switches between latent and explicit reasoning and reports token-efficiency gains on Qwen3-8B models.
- 18 days ago x.com
Inference throughput: SGLang and vLLM on A6000 GPUs
A post published on August 4, 2024 reports higher throughput with SGLang than with vLLM on four A6000 GPUs, but similar performance with smaller batches.
ToolUniverse and research-tool integration
A post published on September 30, 2025 describes a shared interface for discovering and running more than 600 local and remote research tools that use MCP.
- 18 days ago x.com
Knowledge transfer between AI models
A repository published on July 29, 2024 presents a notebook for tuning a smaller model with domain knowledge generated by a larger model.
Magnetic transistor could combine control and memory
An article dated September 23, 2025, reports that MIT researchers developed a transistor using a magnetic semiconductor, with potential to improve electrical control and integrate memory.
Triton: GPU data access with block pointers
Published on September 23, 2025, the post covers tensor storage and the use of tl.make_block_ptr, strides, and offsets in Triton kernels.
- 18 days ago x.com
Streaming TTS points to multi-voice dialogue
A September 15, 2025 post describes a long-form streaming speech system that interleaves text and audio for contextual dialogue with multiple speakers.
- 18 days ago x.com
Local TTS with voice cloning: risks and implementation
On February 15, 2025, Zonos announced an expressive speech synthesis beta with voice cloning, open models, and a cross-platform local launcher.
TUMIX refines answers with diverse agents at inference time
A post published on October 11, 2025 describes agents with different tools solving problems independently and refining answers over multiple rounds, without retraining the base model.
- 18 days ago x.com
A unified view of Attention-MoE and FFN-MoE
A paper published on September 24, 2025 proposes treating pre-mixed tokens and (V·Wo) as an FFN expert, unifying sparse attention and FFN approaches.
- 18 days ago x.com
UserRL trains agents with simulated users
In a post published on October 8, 2025, Salesforce presents UserRL, a framework for training agents with standardized environments and simulated users.
- 18 days ago x.com
VoiceCraft: speech editing and zero-shot synthesis
Published on August 25, 2024, the text describes VoiceCraft, a model that edits speech and generates voices from a few seconds of reference audio.
- 18 days ago x.com
XEUS: multilingual speech encoder with code and checkpoints
On July 1, 2024, WavLab announced XEUS, a speech encoder trained on more than 1 million hours of audio and covering over 4,000 languages. Code, checkpoints, and data were made available.
- 18 days ago x.com
Zerox: structured document extraction
A post published on August 10, 2025 describes an open-source OCR and document extraction project that uses schemas to retrieve specific fields.
Let's assess your project
I reply with scope and a proposal.