Emil Michael during a Senate armed services committee confirmation hearing in Washington DC on 27 March 2025. Emil Michael during a Senate armed services committee confirmation hearing in Washington DC on 27 March 2025. The top Pentagon official overseeing military artificial intelligence policy, who reaped profits earlier this year of up to $24m selling his private investment in Elon Musk's AI company, has now sold his holdings in another AI company for between $5m and $25m, according to records seen by the Guardian. Earlier this year, the Guardian disclosed that the official, Emil Michael, had profited handsomely from his investment in Musk's xAI in January, with a gain of between 400% and 4,800% . Michael's latest financial disclosures show that in June, he sold off his shares in Perplexity, an AI-powered search engine.

Back in February, debuted Perplexity Computer. Like Claude Cowork, it's a suite of AI agents that can autonomously complete tasks using the web, as well as files and apps on your PC. Since then, the platform has evolved to encompass a few different products, including Personal Computer for the Mac, and today Perplexity is announcing yet offshoot called Hybrid Compute. The new tools allows you to split a task between a frontier, cloud-based model like Opus 5 or GPT-5.6 Sol and a local LLM running on your computer -- the idea being that the local model can handle any sensitive information so that it remains safe and secure on your machine. Perplexity suggests a few different use cases where Hybrid Compute would be a good fit. For instance, a lawyer might want to prepare a brief that compares the case they're working on against existing case law.

I wore the world's first HDR10 smart glasses TCL's new E Ink tablet beats the Remarkable and Kindle Anker's new charger is one of the most unique I've ever seen I wore the world's first HDR10 smart glasses TCL's new E Ink tablet beats the Remarkable and Kindle Anker's new charger is one of the most unique I've ever seen I used Perplexity's AI agent in Windows to tackle 5 complex tasks - here's what impressed me most No longer limited to the Mac, Perplexity's Personal Computer for Windows AI can complete complex tasks on your PC with little or no input from you. Agentic AI is designed to save you time and labor by handling complex, multi-step tasks on your computer with little or no interaction from you. Just describe the mission to your favorite AI, and it should complete the required steps all on its own. Among the companies rolling out their own agentic AIs, Perplexity has been touting its Personal Computer feature built into its desktop apps. Previously available only for the Mac, Perplexity Personal Computer recently expanded to Windows, supporting both Windows 11 and Windows 10. The only requirement is that you must subscribe to one of Perplexity's paid plans -- Pro, Max, or Enterprise.

Follow this author to personalize your feed and get instant alerts. Follow Go to your personalized feed WHY FOLLOW? Smart Alerts: Get notified about major news as it happens. Pillay is an editorial fellow at TIME. In June, Andy Konwinski assembled around 100 influential AI researchers and engineers at San Francisco's Exploratorium to discuss how to keep the frontier of AI research from closing.

I wore the world's first HDR10 smart glasses TCL's new E Ink tablet beats the Remarkable and Kindle Anker's new charger is one of the most unique I've ever seen I wore the world's first HDR10 smart glasses TCL's new E Ink tablet beats the Remarkable and Kindle Anker's new charger is one of the most unique I've ever seen Portable Computer is Perplexity's new local AI agent - why it's a game changer Running an AI model locally, Perplexity's new Portable Computer can deliver faster performance, tighter security, and lower costs. But it has some strict requirements. Perplexity's new agentic Portable Computer runs AI locally. Local AI is faster and cheaper than cloud-based models. But you'll need some beefy hardware for Linux or Windows.

Discrete diffusion models offer a flexible, controllable approach to structured sequence generation, yet they still lag behind causal language models in expressive power. A key limitation lies in their reliance on the Markovian assumption, which restricts each step to condition only on the current state, leading to potential uncorrectable error accumulation. In this paper, we introduce CaDDi (Causal Discrete Diffusion Model), a discrete diffusion model that conditions on the entire generative trajectory, thereby lifting the Markov constraint and allowing the model to revisit and improve past states. By unifying sequential (causal) and temporal (diffusion) reasoning in a single non-Markovian transformer, CaDDi also treats standard causal language models as a special case and permits the direct reuse of pretrained LLM weights with no architectural changes. Empirically, CaDDi outperforms state-of-the-art discrete diffusion baselines on natural-language benchmarks, substantially narrowing the remaining gap to large autoregressive transformers.

Gated Linear Units (GLUs) have become essential components in the feed-forward networks of state-of-the-art Large Language Models (LLMs). However, they require twice as many memory reads compared to feed-forward layers without gating, due to the use of separate weight matrices for the gate and value streams. To address this bottleneck, we introduce Masked Gated Linear Units (MGLUs), a novel family of GLUs with an efficient kernel implementation. The core contribution of MGLUs include: (1) the Mixture of Element-wise Gating (MoEG) architecture that learns multiple binary masks, each determining gate or value assignments at the element level on a single shared weight matrix resulting in reduced memory transfer, and (2) FlashMGLU, a hardware-friendly kernel that yields up to a 19.7 inference-time speed-up over a naïve PyTorch MGLU and is 47% more memory-efficient and 34% faster than standard GLUs despite added architectural complexity on an RTX5090 GPU. In LLM experiments, the Swish-activated variant SwiMGLU preserves its memory advantages while matching--or even surpassing--the downstream accuracy of the SwiGLU baseline.

While diffusion language models (DLMs) enable fine-grained refinement, their practical controllability remains fragile. We identify and formally characterize a central failure mode--update-forgetting--in which uniform, context-agnostic updates induce token-level fluctuations across timesteps, erasing earlier semantic edits and disrupting the cumulative refinement process, thereby degrading fluency and coherence. As this failure originates in uniform, context-agnostic updates, effective control demands explicit token ordering. We propose Token Timestep Allocation (TTA-DIFFUSION), which realizes soft, semantic token ordering via pertoken timestep schedules: critical tokens are frozen early, while uncertain tokens receive continued refinement. This timestep-based ordering can be instantiated as either a fixed policy or an adaptive policy driven by task signals, thereby supporting a broad spectrum of refinement strategies. Because it operates purely at inference time, it applies uniformly across various DLMs and naturally extends to diverse supervision sources. Empirically, TTA-DIFFUSION improves controllability and fluency: on sentiment control, it yields >20%higher accuracy and nearly halves perplexity using <1/5 the steps; in detoxification, it lowers maximum toxicity (12.2 vs. 14.5) and perplexity (26.0 vs. 32.0). Together, these results demonstrate that softened ordering via timestep allocation is the critical lever for mitigating update-forgetting and achieving stable and controllable diffusion text generation.

As training billion-scale transformers becomes increasingly common, employing multiple distributed GPUs along with parallel training methods has become a standard practice. However, existing transformer designs suffer from significant communication overhead, especially in Tensor Parallelism (TP), where each block's MHA-MLP connection requires an all-reduce communication. Through our investigation, we show that the MHA-MLP connections can be bypassed for efficiency, while the attention output of the first layer can serve as an alternative signal for the bypassed connection. Motivated by the observations, we propose FAL (First Attentions Last), an efficient transformer architecture that redirects the first MHA output to the MLP inputs of the following layers, eliminating the per-block MHA-MLP connections. This removes the all-reduce communication and enables parallel execution of MHA and MLP on a single GPU. We also introduce FAL+, which adds the normalized first attention output to the MHA outputs of the following layers to augment the MLP input for the model quality. Our evaluation shows that FAL reduces multi-GPU training time by up to 44%, improves single-GPU throughput by up to 1.18, and achieves better perplexity compared to the baseline GPT. FAL+ achieves even lower perplexity without increasing the training time than the baseline. Codes are available at: https://casl-ku.github.io/FAL/

Data duplication within large-scale corpora often impedes large language models' (LLMs) performance and privacy. In privacy-concerned federated learning scenarios, conventional deduplication methods typically rely on trusted third parties to perform uniform deletion, risking loss of informative samples while introducing privacy vulnerabilities. To address these gaps, we propose Federated ReWeighting (FedRW), the first privacy-preserving framework, to the best of our knowledge, that performs soft deduplication via sample reweighting instead of deletion in federated LLM training, without assuming a trusted third party. At its core, FedRW proposes a secure, frequency-aware reweighting protocol through secure multi-party computation, coupled with a parallel orchestration strategy to ensure efficiency and scalability. During training, FedRW utilizes an adaptive reweighting mechanism with global sample frequencies to adjust individual loss contributions, effectively improving generalization and robustness. Empirical results demonstrate that FedRW outperforms the state-of-the-art method by achieving up to 28.78 speedup in preprocessing and approximately 11.42% improvement in perplexity, while offering enhanced security guarantees. FedRW thus establishes a new paradigm for managing duplication in federated LLM training.