abus-aikorea /voice-pro
The voice-pro project by abus-aikorea offers a Gradio WebUI for creators and developers, integrating advanced TTS engines (Edge-TTS, kokoro) and zero-shot voice cloning (E2 & F5-TTS,
→ View original source
An internal OpenAI Astra model solved 10 major open math and CS problems
OpenAI’s internal Astra model has reportedly solved ten significant unsolved problems in mathematics and computer science. The achievement demonstrates the model’s advanced symbolic reasoning and problem‑solving capabili…
→ View original source
I ran an autonomous AI pentester on OWASP Juice Shop, on a LOCAL model. 57 real vulns in 28 min.
An autonomous AI pentester was tested against the OWASP Juice Shop vulnerable application using a local model in a black-box environment. The testing process successfully identified 57 real vulnerabilities within a 28-mi…
→ View original source
llama.cpp just added MTP / DSpark support for DeepSeek V4 Flash
The llama.cpp repository has introduced MTP (Model Tensor Parallelism) and DSpark support, enabling efficient execution of the DeepSeek V4 Flash model on local hardware. This update allows users to run larger models with…
→ View original source
NVIDIA AI Releases Molt: A PyTorch-Native Agentic Reinforcement Learning Framework
NVIDIA has launched Molt, a PyTorch‑native framework for agentic reinforcement learning that emphasizes a compact codebase of roughly 8.6K lines of RL code. It integrates a single training backend (NeMo AutoModel) and se…
→ View original sourceOpenAI's claimed disproof of Connes' Rigidity Conjecture is invalid [pdf]
A new paper asserts that OpenAI's claimed disproof of Connes' Rigidity Conjecture is invalid. The technical critique challenges the validity of the previous findings regarding this mathematical conjecture. Read original
→ View original source
Notable this week: Kimi K3 weights land, MCP goes stateless, OfficeCLI for agents
Moonshot AI released the Kimi K3 model weights on Hugging Face on July 27, featuring 2.8 trillion parameters with sparse MoE architecture and MXFP4 quantization, achieving top rankings on Frontend Code Arena and the Arti…
→ View original source
I wanted to see exactly how far a consumer-grade system can be pushed with LLM concurrency. So I benchmarked 15+ models to find out.
A user benchmarked 15+ large language models on a consumer-grade system to determine the concurrency limits of an RTX 5060 with 8GB VRAM. The goal was to explore whether a "gaming" or workstation PC could feasibly drive …
→ View original sourcePersistent State Machines: LLM Attention with INT4 In-Memory Cells
The paper introduces Persistent State Machines (PSM) that employ INT4‑compressed in‑memory cells to implement attention mechanisms in large language models, enabling persistent state across layers. This design reduces me…
→ View original source
A Vision Transformer makes an image look like a sentence — 16 16 patches become tokens and every patch attends to every other
The transformer was built for sequences of words. The Vision Transformer's one radical move is to make an image look like a sentence — and then run the exact same encoder from language on it, with no
→ View original sourceMiniMax Releases MiniMax H3: An Omni-Modal Video Model That Generates 15-Second 2K Clips With Native Stereo Audio
MiniMax introduces the H3, an omni‑modal video model that generates 15‑second 2K clips with native stereo audio, priced at $0.13 per second and ranking #1 in video editing on Artificial Analysis. It uses a newly built H3…
→ View original source
DeepSeek V4 Flash 0731 in Hermes Agent and one prompt, took 32 minutes and cost 0.07$, this model is so cheap to the point where 2 dollars can last you a full day.
A user reported that the DeepSeek V4 Flash 0731 model, running via Hermes Agent, processed a single prompt over 32 minutes for a total cost of only $0.07. The extreme cost-efficiency of the model suggests that a $2 budge…
→ View original source