LLMs & Generative AI - Page 9 of 55
Latest breakthroughs in large language models and generative AI shaping the future of artificial intelligence and machine learning.
1086 articles
View complete article list
Mobile NPU powers on‑device diffusion LLM with Multi‑Block Speculative Decoding
The slow drip of token-by-token generation has long been the bottleneck for running large language models on mobile devices. On-device diffusion LLMs promise privacy and responsiveness, yet their inference latency remains stubbornly high, until now.
Orchestra‑o1 Enables Efficient Omnimodal Agent Collaboration
Right now, your phone uses one AI for photos and another for text. They don't talk. This fragmentation is the central problem for labs aiming to build a machine that can truly see, read, and listen as one.
Vision LLMs Expand PDF Parsing to Charts, Diagrams, and Tables
Traditional PDF parsing breaks down where there are no characters to read. OCR and layout engines fail on charts, diagrams, and figures, by design. But vision LLMs change the rules.
Claude Fable 5 beats GPT‑5.5 by 13 points on FrontierMath tier‑4 tests
Thirteen points is a thrashing. On FrontierMath's hardest tier, the new standard is set not by OpenAI, but by Anthropic's Claude Fable 5. This isn't a minor edge. It's a decisive lead. Fable 5 scored roughly 88 percent. GPT-5.5 managed 75.
German Court Holds Google Liable for False AI-Generated Overviews
For a year, Google's lawyers saw this coming. Now it's here. A Berlin court has ruled the company is directly liable for the fabrications its AI Overviews produce. Those summaries, the judges stated, are new and independent statements.
Google's DiffusionGemma: open diffusion model for faster text generation
**The old way of generating text is a bottleneck.** Token by token, each word waiting on the last. Google’s DiffusionGemma shatters that serial chain. Instead of predicting one piece at a time, it denoises an entire 256-token block in parallel.
Google sues Chinese Outsider Enterprise for Gemini-driven phishing on Telegram
Google has filed suit against a Chinese cybercrime operation that transformed its Gemini AI into a phishing assembly line. The group automated its scam campaigns on Telegram, recycling infrastructure once dedicated to blasting out SMS spam.
PersonaDrive conditions VLA agents on human driving demos for simulation
Human driving isn’t just about reaching a destination, it’s about style. Aggressive, conservative, or somewhere in between, the way a driver accelerates, brakes, and navigates defines how natural an autonomous agent feels in simulation.
ToolSense Framework Audits LLM Tool Knowledge Beyond Constrained Decoding
Most tests for AI tool use are rigged. They give the model the exact question and the exact answer format, then declare success. It's like teaching someone to bake by handing them a pre-assembled cake.
Gemini Omni adds AI video generation, using compute limits based on complexity and size
Google's latest Gemini model can now make videos. Not well, but that's beside the point. The important thing is the mechanism: a new, fluid system of rationing compute power that changes with each request. You get a budget.
Xiaomi's MiMo Code beats Claude Code on 200+ step tasks, free MiMo Auto to V2.5
Here's the thing: Xiaomi just dropped MiMi Code, an open‑source coding assistant that claims to outpace Anthropic’s Claude Code on tasks that stretch beyond 200 steps.
OpenAI hires Sottiaux in 2024, shifts from internal tools to ChatGPT overhaul
OpenAI's big move for 2024 was a personnel file. They hired Tibo Sottiaux, the engineer who built Codex, their internal coding tool. This wasn't just another recruitment.
Low Kruskal-Rank Adaptation Shows Matrix Rank Stays r, Kruskal Rank Falls to 1
Matrix rank offers a comforting illusion: it tells you an update has full capacity under its parameter budget. Kruskal rank shatters that comfort.
Anthropic apologizes for invisible guardrails on Claude Fable, first Mythos model
Anthropic just admitted it quietly rigged Claude Fable, its first Mythos model, to sabotage user queries.
AI pre‑mediation matched professional mediators in multi‑issue negotiation test
Mediators are expensive, and their best work often happens before anyone sits at the table. This preparation is where settlements are built or broken. So what if you could skip the wait and the bill, and just let a machine do the groundwork?
AVLLMs Mirror VLM and VideoLLM Sequential Flow in Audio‑Visual Tasks
Models that process both sound and sight are often treated like alien minds. New research reveals a much more boring, and more useful, truth. They're not reinventing anything. They're just copying the plumbing.
vLLM uses custom GPU kernels, TorchInductor and CUTLASS for portable inference
Portable inference across diverse hardware is a brutal optimization problem. vLLM attacks it with a triple threat: custom GPU kernels for raw performance, TorchInductor for graph-level fusion, and battle-tested GEMM libraries like CUTLASS and...
Claude Fable declines basic biology queries; Opus 4.8 responds
Anthropic confirmed it. Their new Claude Fable 5 model is refusing to answer even elementary biology questions. The rationale, according to company spokesperson Paruul Maheshwary, is a deliberate hedge against risk.
Run DiffusionGemma on NVIDIA GPUs for high‑throughput text generation
The line between experimentation and production has never been thinner. DiffusionGemma is changing what’s possible for high‑throughput text generation, and NVIDIA hardware is the engine that makes it real.
SynIB Introduces Information Bottleneck to Boost Multimodal Synergy
Most AI models are lazy. They take the easy way out. Confronted with data from different sources, like an image paired with text, they'll grab the most obvious cue from one and ignore the rest. This works fine for simple tasks.
Featured Resources & Reviews
No Code MBA Course Review
Learn to build AI-powered apps without coding. Our comprehensive review of No Code MBA's course.
AI Tools & Resources
Curated collection of AI tools, courses, and frameworks to accelerate your AI journey.
Weekly AI Digest
Get the week's most important AI news delivered to your inbox every week.