AI News

Unitree’s Dex5-S promises 22 degrees of freedom and backdrivable joints at human-hand scale. Its launch leaves key specifications and complete-task performance unverified.

As Large Language Model (LLM) agents are applied in continuously interactive environments, driving the evolution of their own capabilities becomes a core problem for achieving long-term autonomy. Currently, environmental knowledge is typically treated as an external fixed input rather than as part of the agent's ongoing evolution. Reinforcement learning methods usually optimize policies through environmental interaction but tend to adapt only to fixed task distributions or single environments. This paper proposes TTSE (Two-Track Self-Evolution), a dual-track online self-evolution framework that separates evolving knowledge into FACT (environmental facts, whose reliability is continuously verified through interaction evidence) and TIP (task-conditioned implementation procedures). From a decision-theoretic perspective, we decompose the agent's excess risk into environment-representation regret and conditional-execution regret, characterize the conditions under which environment-conditioned policies strictly outperform condition-agnostic policies, and bound the downstream risk in terms of FACT identification error and cross-condition mismatch cost. In practice, TTSE's ablation experiments on GDPevo validate the advantage of dual-track evolution. On the classic agent task benchmarks ALFWorld and ScienceWorld, TTSE further demonstrates superior task adaptation. Moreover, TTSE is broadly compatible with existing skill self-evolution methods; combined with the Bayesian-Agent algorithm, a single-track ablation validates the dual-track advantage, substantially improving the aggregate score across the five major domains of SOPBench over three independent repetitions. Finally, on the real end-to-end task benchmark PinchBench, TTSE is integrated into a general agent framework via retrieval-based injection and stably outperforms the baseline across three independent runs.

Despite advances in long-context inference, large language models (LLMs) remain fundamentally limited by the key-value (KV) caching mechanisms that are necessary for stable computation. Techniques such as selective token eviction and pruning have vastly mitigated these issues, but often discard core information to manage the growing cache. In this paper, we propose Attention with Routed Memory (ARM) a novel KV caching structure that introduces a fully differentiable, fixed-size memory system organized as a hierarchical router. Via a Gumbel-Softmax, ARM learns to select memory slots and perform sigmoid-gated updates that softly combine new and stored information, avoiding hard eviction and reducing information loss. By further training a policy to dynamically select varying amounts of memory at inference, ARM adapts its accesses for both simple contexts and inputs that require deeper reasoning, enabling more scalable and effective retrieval on both short- and long-contexts. Experimental results on standard commonsense and long-context reasoning benchmarks demonstrate that ARM achieves superior performance and efficiency compared to fixed KV-caching approaches, while remaining efficient and scalable in terms of both memory and generation latency.

AMD and Nvidia no longer have to worry.

SpaceXAI's most powerful model for coding and knowledge work. Twice as fast, at half the price of comparable models.

#Egyptian AI company Synapse Analytics has raised $13M in Series A funding led by Partech, bringing total funding to $17M. The company builds decisioning infrastructure for banks and fintechs across the Middle East, Africa and Latin America.

https://t.co/zA70FYQlHJ

Clipboard history and screenshots, searchable together. BiBimba reads the text inside screenshots, so you can find a receipt total or an error message later and paste just what you need. On-device AI translates, summarizes, rewrites. Stays on your Mac. $9.

UTCP (Universal Tool Calling Protocol) is an open standard for AI agents to call tools directly, positioned as a lightweight alternative to MCP. It eliminates the "wrapper tax" for lower latency by using a simple JSON manifest to connect to native APIs.

With bolt.new you can prompt fullstack web applications into existence, see them executed in real-time, debug errors as they occur, & deploy a fully functional app—all without ever leaving your browser or personally writing a single line of code!

Want to learn more? Join us today at 11am PT for live interactive office hours on YouTube! https://www.addevent.com/event/kp23703256?

Toyota unveils ELEY, a wheeled humanoid robot powered by physical AI, set to be part of a massive 400,000-unit global factory deployment by 2028.

tiny Jev-like family of decision models built on top of Qwen3.5 you can train and run on your own - jaredpalmer/kev

More than the Slovakian government will spend in the same period.

Deploying Large Language Models (LLMs) in healthcare requires robust performance across two complementary dimensions - diagnostic reasoning: the convergent, evidence-driven task of inferring a patient's condition from clinical data to produce a diagnosis, and clinical healthcare reasoning: the broader, navigational judgment required to communicate, plan, and adapt across multi-turn clinical interactions where a single correct answer may not exist. Recent benchmarks such as HealthBench and MedXpertQA reveal persistent weaknesses in both areas, exposing failures in complex diagnostic scenarios and limitations in contextual, patient-centered dialogue. We introduce a sequential training framework that targets these facets using synthetic data and rubric-based reinforcement learning. First, we improve diagnostic reasoning using MedBullets-derived questions with rule- and rubric-guided Reinforcement Learning (RL). We then shift to clinical reasoning by generating 5.3k synthetic multi-turn scenarios, each paired with multi-dimensional rubrics to comprehensively assess the response. This approach yields over 10% improvement on MedXpertQA, and our 30B model achieves 50.1% accuracy on HealthBench-Hard, surpassing proprietary baselines including GPT-5 (thinking). Our results show that targeted synthetic datasets and rubric-based training can systematically improve both diagnostic and interactive clinical reasoning in medical LLMs.

What limits KV-cache compression at extreme bit-rates? We argue that it is not the choice of compression scheme, but how its budget is allocated across attention heads. Existing methods apply rank and bit-width uniformly, ignoring that each head has a different optimal mix of rank truncation and quantization. We show that co-optimizing rank and bit-width per head, using only standard low-rank projection and scalar quantization, dominates uniform allocation, with the largest gains at low bit-rates. Our method, KV-COBRA (Co-Optimized Bit-Rank Allocation), formalizes this as a resource-allocation problem: it balances rank-truncation loss against quantization loss within each head, then redistributes budget across heads to minimize total distortion. A fused Hadamard rotation equalizes per-channel variance, and reordering the SVD basis by attention-KL importance makes the solver query-aware. The same allocator extends to joint $K{+}V$ compression. On perplexity, zero-shot, and long-context benchmarks from $0.5$ to $4$ bits per dimension (bpd), KV-COBRA shows the smallest accuracy degradation among evaluated methods at low bpd, with no per-token overhead.

An LLM agent's capability is largely magnified by its harness, namely the prompts, control flow, tooling, memory, and context management surrounding the frozen backbone model. Recent methods increasingly automate this process by iteratively proposing and selecting component-wise edits of an agent harness, practically establishing a form of recursive self-improvement (RSI) at the agent-system level. However, such recursive evolution may overfit by memorizing the training tasks, showing large in-distribution gains that shrink or even vanish on out-of-distribution benchmarks. We introduce Regularized Recursive Self-Improvement of Agent Harnesses (RRSI), which incorporates the principles of regularizations into harness self-improvement by constraining the evolution candidate proposal and selection. The proposer operates with a temporally annealed budget, limiting how many edits a candidate can bundle, and it encourages unexplored trajectories based on evolution history. The selector is equipped with a critic and a pruner: the critic screens benchmark-specific proposals, while the pruner, removes changes that are too small, too expensive, or no longer useful. Together these constraints favor reusable agent mechanisms over benchmark-specific ones or even noises. Across eight benchmarks spanning coding, agentic workspace and engineering design tasks, RRSI gains up to 14.1 points on the split it evolves against and up to 4.7 points on the five out-of-distribution benchmarks, while producing a harness that runs on 30% fewer policy tokens than the unregularized evolution. Code is available at https://github.com/google-research/rrsi and project page is https://regularized-rsi.com/.

Prompt injection defenses shouldn't rely solely on a model’s own training.

Meta's Muse agent assumes the model will be tricked; instead, it builds security at the OS level.

🔑 Model never handles real credentials
🛡️ Tools run in isolated Linux containers
🛑 Independent gatekeeper verifies outbou…

BREAKING: Anthropic’s next Claude Opus model projected to be released by Thursday.

72% chance.

https://t.co/OIXFgNGWK9

A startup just raised money by promising investors it would not train its own model from scratch.

Naive AI, backed by Tencent, is now valued at 1.42 billion dollars after three funding rounds.

Its entire strategy is building on top of an existing open-weight model and focusing on the improvements…

Basedash is the AI-native Business Intelligence platform that lets you create dashboards and understand your customers using natural language. Connect your data, describe the chart you want, and let AI generate the visualization. No SQL required.

Framer gives you an AI design agent and a freeform canvas for turning ideas into polished, interactive sites. Start with a prompt and the agent can create pages, layouts, copy, visuals, interactions, components, responsive breakpoints, and code directly in your project. Everything stays editable, so you can keep prompting or refine the work yourself. Framer also includes CMS, collaboration, analytics, SEO, performance tools, and hosting to help you publish and improve your work.

JUST IN: MIND raised $72M Series B, led by Crosspoint with YL Ventures and Paladin. Total funding: $112M. AI-native DLP for GenAI/agentic data risk. 17x revenue and 8x customers in a year.

Google just open-sourced the agent infrastructure layer.

Google launched AX this week: a Kubernetes-style declarative orchestrator for running billions of autonomous agent workloads at cluster scale. GitHub: https://t.co/TaLiBVh6wg.

The framing from the team is precise. Agents are not stateless m…