Traditional LLM alignment methods are vulnerable to heterogeneity in human preferences. Fitting a naïve probabilistic model to pairwise comparison data (say over prompt-completion pairs) yields an inconsistent estimate of the population-average utility -a canonical measure of social welfare. We propose a new method, dubbed the sign estimator, that provides a simple, provably consistent, and efficient estimator by replacing cross-entropy with binary classification loss in the aggregation step. This simple modification recovers consistent ordinal alignment under mild assumptions and achieves the first polynomial finite-sample error bounds in this setting. In realistic simulations of LLM alignment using digital twins, the sign estimator substantially reduces preference distortion over a panel of simulated personas, cutting (angular) estimation error by nearly 35% and decreasing disagreement with true population preferences from 12% to 8% compared to standard RLHF. Our method also compares favorably to panel data heuristics that explicitly model user heterogeneity and require tracking individual-level preference data-all while maintaining the implementation simplicity of existing LLM alignment pipelines.

While continuous diffusion has shown remarkable success in continuous domains such as image generation, its direct application to discrete data has underperformed compared to purely discrete formulations. This gap is counterintuitive, given that continuous diffusion learns score functions that enable joint evolution across multiple positions. To understand this gap, we introduce token identifiability as an analytical framework for understanding how Gaussian noise corrupts discrete data through two mechanisms: discrete identity corruption and continuous rank degradation. We reveal that these mechanisms scale differently with vocabulary size, creating a temporal dissonance: at noise levels where discrete corruption preserves enough structure for conditional learning, continuous denoising is trivial; at noise levels where continuous denoising is meaningful, discrete corruption destroys nearly all conditional structure. To solve this, we propose CANDI (Continuous ANd DIscrete diffusion), a hybrid framework that decouples discrete and continuous corruption, enabling simultaneous learning of both conditional structure and continuous geometry. We empirically validate the temporal dissonance phenomenon and demonstrate that CANDI successfully avoids it. This unlocks the benefits of continuous diffusion for discrete spaces: on controlled generation, CANDI enables classifier-based guidance with off-the-shelf classifiers through simple gradient addition; on text generation, CANDI outperforms masked diffusion at low NFE, demonstrating the value of learning continuous gradients for discrete spaces. We include the code on the project page available here: https://patrickpynadath1.github.io/candi-lander

Microsoft's CEO, Satya Nadella, speaks at the company's annual developer conference in Seattle, Washington. Microsoft's CEO, Satya Nadella, speaks at the company's annual developer conference in Seattle, Washington. Tech giant reports earnings of $3.72 per share day after deal with OpenAI pushed value of company to more than $4tn Microsoft blew off concerns of overspending on AI on Wednesday, reporting elevated earnings even as it faced an outage of its cloud computing service, Azure, and its office software suite, 365. The strong earnings report comes a day after a deal with OpenAI pushed the value of the tech giant to more than $4tn. After its Xbox and investor relations pages went down, the company issued a statement that said: "We are working to address an issue affecting Azure Front Door that is impacting the availability of some services."

Human-level AI is still some ways off. Even the best artificial intelligence agents are fairly hopeless at online freelance work, according to an experiment that challenges the idea of AI replacing office workers en masse. The Remote Labor Index, a new benchmark developed by researchers at data annotation company Scale AI and the Center for AI Safety (CAIS), a nonprofit, measures the ability of frontier AI models to automate economically valuable work. The researchers gave several leading AI agents a range of simulated freelance work and found that even the best could perform less than 3 percent of the work, earning $1,810 out of a possible $143,991. The researchers looked at several tools and found the most capable to be Manus from a Chinese startup of the same name, followed by Grok from xAI, Claude from Anthropic, ChatGPT from OpenAI, and Gemini from Google.

Is the US eyeing its next Latin American target? Why is Trump tearing down parts of the White House? Nvidia has become the first company to reach $5 trillion in market value amid a global artificial intelligence arms race. The chipmaker surge on Wednesday came only three months after the company topped the $4 trillion mark . Since the launch of ChatGPT in 2022, Nvidia's shares have climbed 12-fold as the AI frenzy propelled the S&P 500 to record highs, igniting a debate on whether frothy tech valuations could lead to the next big bubble.

When you purchase through links in our articles, we may earn a small commission. Users will soon be able to use PayPal to pay for product recommendations made by OpenAI's ChatGPT. PayPal recently signed a contract with OpenAI to integrate the digital wallet into ChatGPT, reports CNBC . This will allow users to easily pay for the products they discover via the AI tool. The agreement allows PayPal users to make payments via ChatGPT merchants to list and sell their goods in ChatGPT.

DeepSeek may have found a new way to improve AI's ability to remember An AI model released by Chinese AI company DeepSeek uses new techniques that could significantly improve AI's ability to "remember." The optical character recognition model works by extracting text from an image and turning it into machine-readable words. This is the same technology that powers scanner apps, translation of text in photos, and many accessibility tools. Researchers say the model's main innovation lies in how it processes information--specifically, how it stores and retrieves data. Improving how AI models "remember" could reduce how much computing power they need to run, thus mitigating AI's large (and growing) carbon footprint. The AI Hype Index: Data centers' neighbors are pivoting to power blackouts That's why we've created the AI Hype Index--a simple, at-a-glance summary of everything you need to know about the state of the industry.

When you purchase through links in our articles, we may earn a small commission. You can now take advantage of this classic Windows scripting tool even if you have zero programming experience. AutoHotkey (AHK) is a free and simple yet powerful Windows scripting language. It doesn't get a lot of press these days, but Windows geeks used to love writing and swapping AHK scripts. You can learn AHK and write the scripts yourself if you want to--the AutoHotkey documentation is pretty good --or you could use an AI tool like ChatGPT, Gemini, Claude, Copilot, etc. to do the work for you.

An AI model released by the Chinese AI company DeepSeek uses new techniques that could significantly improve AI's ability to "remember." Released last week, the optical character recognition (OCR) model works by extracting text from an image and turning it into machine-readable words. This is the same technology that powers scanner apps, translation of text in photos, and many accessibility tools. OCR is already a mature field with numerous high-performing systems, and according to the paper and some early reviews, DeepSeek's new model performs on par with top models on key benchmarks. But researchers say the model's main innovation lies in how it processes information--specifically, how it stores and retrieves memories. Improving how AI models "remember" information could reduce the computing power they need to run, thus mitigating AI's large (and growing) carbon footprint.

Modern LLMs are increasingly deep, and depth correlates with performance, albeit with diminishing returns. However, do these models use their depth efficiently? Do they compose more features to create higher-order computations that are impossible in shallow models, or do they merely spread the same kinds of computation out over more layers? To address these questions, we analyze the residual stream of the Llama 3.1, Qwen 3, and OLMo 2 family of models. We find: First, comparing the output of the sublayers to the residual stream reveals that layers in the second half contribute much less than those in the first half, with a clear phase transition between the two halves. Second, skipping layers in the second half has a much smaller effect on future computations and output predictions. Third, for multihop tasks, we are unable to find evidence that models are using increased depth to compose subresults in examples involving many hops. Fourth, we seek to directly address whether deeper models are using their additional layers to perform new kinds of computation. To do this, we train linear maps from the residual stream of a shallow model to a deeper one. We find that layers with the same relative depth map best to each other, suggesting that the larger model simply spreads the same computations out over its many layers. All this evidence suggests that deeper models are not using their depth to learn new kinds of computation, but only using the greater depth to perform more fine-grained adjustments to the residual. This may help explain why increasing scale leads to diminishing returns for stacked Transformer architectures.