News Gist .News

Inception, a new Palo Alto-based company started by Stanford computer science professor Stefano Ermon, claims to have developed a novel AI model based on “diffusion” technology. Inception's diffusion-based large language model (DLM) offers the capabilities of traditional LLMs, including code generation and question-answering, but with significantly faster performance and reduced computing costs. The company's breakthrough has significant implications for the development of generative AI models in text generation.

Developers can access AI model capabilities at a fraction of the price thanks to distillation, allowing app developers to run AI models quickly on devices such as laptops and smartphones. The technique uses a "teacher" LLM to train smaller AI systems, with companies like OpenAI and IBM Research adopting the method to create cheaper models. However, experts note that distilled models have limitations in terms of capability.

OpenAI has begun rolling out its newest AI model, GPT-4.5, to users on its ChatGPT Plus tier, promising a more advanced experience with its increased size and capabilities. However, the new model's high costs are raising concerns about its long-term viability. The rollout comes after GPT-4.5 launched for subscribers to OpenAI’s $200-a-month ChatGPT Pro plan last week.

GPT-4.5 is OpenAI's latest AI model, trained using more computing power and data than any of the company's previous releases, marking a significant advancement in natural language processing capabilities. The model is currently available to subscribers of ChatGPT Pro as part of a research preview, with plans for wider release in the coming weeks. As the largest model to date, GPT-4.5 has sparked intense discussion and debate among AI researchers and enthusiasts.

Alibaba Group's release of an artificial intelligence (AI) reasoning model has driven its Hong Kong-listed shares more than 8% higher on Thursday, outperforming global hit DeepSeek's R1. The company's AI unit claims that its QwQ-32B model can achieve performance comparable to top models like OpenAI's o1 mini and DeepSeek's R1. Alibaba's new model is accessible via its chatbot service, Qwen Chat, allowing users to choose various Qwen models.

Anna Patterson's new startup, Ceramic.ai, aims to revolutionize how large language models are trained by providing foundational AI training infrastructure that enables enterprises to scale their models 100x faster. By reducing the reliance on GPUs and utilizing long contexts, Ceramic claims to have created a more efficient approach to building LLMs. This infrastructure can be used with any cluster, allowing for greater flexibility and scalability.

Foxconn has launched its first large language model, named "FoxBrain," which uses 120 Nvidia GPUs and is based on Meta's Llama 3.1 architecture to analyze data, support decision-making, and generate code. The model, trained in about four weeks, boasts performance comparable to world-class standards despite a slight gap compared to China's DeepSeek distillation model. Foxconn plans to collaborate with technology partners to expand the model's applications and promote AI in manufacturing and supply chain management.

DeepSeek R1 has shattered the monopoly on large language models, making AI accessible to all without financial barriers. The release of this open-source model is a direct challenge to the business model of companies that rely on selling expensive AI services and tools. By democratizing access to AI capabilities, DeepSeek's R1 model threatens the lucrative industry built around artificial intelligence.

In accelerating its push to compete with OpenAI, Microsoft is developing powerful AI models and exploring alternatives to power products like Copilot bot. The company has developed AI "reasoning" models comparable to those offered by OpenAI and is reportedly considering offering them through an API later this year. Meanwhile, Microsoft is testing alternative AI models from various firms as possible replacements for OpenAI technology in Copilot.

Thomas Wolf, co-founder and chief science officer of Hugging Face, expresses concern that current AI technology lacks the ability to generate novel solutions, functioning instead as obedient systems that merely provide answers based on existing knowledge. He argues that true scientific innovation requires AI that can ask challenging questions and connect disparate facts, rather than just filling in gaps in human understanding. Wolf calls for a shift in how AI is evaluated, advocating for metrics that assess the ability of AI to propose unconventional ideas and drive new research directions.

Amazon is reportedly venturing into the development of an AI model that emphasizes advanced reasoning capabilities, aiming to compete with existing models from OpenAI and DeepSeek. Set to launch under the Nova brand as early as June, this model seeks to combine quick responses with more complex reasoning, enhancing reliability in fields like mathematics and science. The company's ambition to create a cost-effective alternative to competitors could reshape market dynamics in the AI industry.

Andrew G. Barto and Richard S. Sutton have been awarded the 2025 Turing Award for their pioneering work in reinforcement learning, a key technique that has enabled significant achievements in artificial intelligence, including Google's AlphaZero. This method operates by allowing computers to learn through trial and error, forming strategies based on feedback from their actions, which has profound implications for the development of intelligent systems. Their contributions not only laid the mathematical foundations for reinforcement learning but also sparked discussions on its potential role in understanding creativity and intelligence in both machines and living beings.

DeepSeek has emerged as a significant player in the ongoing AI revolution, positioning itself as an open-source chatbot that competes with established entities like OpenAI. While its efficiency and lower operational costs promise to democratize AI, concerns around data privacy and potential biases in its training data raise critical questions for users and developers alike. As the technology landscape evolves, organizations must balance the rapid adoption of AI tools with the imperative for robust data governance and ethical considerations.

Honor is rebranding itself as an "AI device ecosystem company" and working on a new type of intelligent smartphone that will feature "purpose-built, human-centric AI designed to maximize human potential."The company's new CEO, James Li, announced the move at MWC 2025, calling on the smartphone industry to "co-create an open, value-sharing AI ecosystem that maximizes human potential, ultimately benefiting all mankind." Honor's Alpha plan consists of three steps, each catering to a different 'era' of AI, including developing a "super intelligent" smartphone, creating an AI ecosystem, and co-existing with carbon-based life and silicon-based intelligence.

The advancements made by DeepSeek highlight the increasing prominence of Chinese firms within the artificial intelligence sector, as noted by a spokesperson for China's parliament. Lou Qinjian praised DeepSeek's achievements, emphasizing their open-source approach and contributions to global AI applications, reflecting China's innovative capabilities. Despite facing challenges abroad, including bans in some nations, DeepSeek's technology continues to gain traction within China, indicating a robust domestic support for AI development.

DeepSeek, a Chinese AI startup behind the hit V3 and R1 models, has disclosed cost and revenue data that claims a theoretical cost-profit ratio of up to 545% per day. The company revealed its cost and revenue data after web and app chatbots powered by its R1 and V3 models surged in popularity worldwide, causing AI stocks outside China to plummet in January. DeepSeek's profit margins are likely to be lower than claimed due to the low cost of using its V3 model.

US chip stocks were the biggest beneficiaries of last year's artificial intelligence investment craze, but they have stumbled so far this year, with investors moving their focus to software companies in search of the next best thing in the AI play. The shift is driven by tariff-driven volatility and a dimming demand outlook following the emergence of lower-cost AI models from China's DeepSeek, which has highlighted how competition will drive down profits for direct-to-consumer AI products. Several analysts see software's rise as a longer-term evolution as attention shifts from the components of AI infrastructure.

The 2023 Turing Award winners, Andrew Barto and Rich Sutton, have been recognized for their work in reinforcement learning, a crucial component of artificial intelligence that enables machines to learn from experience. Their research has led to significant advancements in machine learning, paving the way for applications in robotics, game playing, and more. The award acknowledges the pioneers' contributions to this rapidly evolving field.

Chinese AI startup DeepSeek has disclosed cost and revenue data related to its hit V3 and R1 models, claiming a theoretical cost-profit ratio of up to 545% per day. This marks the first time the Hangzhou-based company has revealed any information about its profit margins from less computationally intensive "inference" tasks. The revelation could further rattle AI stocks outside China that plunged in January after web and app chatbots powered by its R1 and V3 models surged in popularity worldwide.

Bret Taylor discussed the transformative potential of AI agents during a fireside chat at the Mobile World Congress, emphasizing their higher capabilities compared to traditional chatbots and their growing role in customer service. He expressed optimism that these agents could significantly enhance consumer experiences while also acknowledging the challenges of ensuring they operate within appropriate guidelines to prevent misinformation. Taylor believes that as AI agents become integral to brand interactions, they may evolve to be as essential as websites or mobile apps, fundamentally changing how customers engage with technology.

The introduction of DeepSeek's R1 AI model exemplifies a significant milestone in democratizing AI, as it provides free access while also allowing users to understand its decision-making processes. This shift not only fosters trust among users but also raises critical concerns regarding the potential for biases to be perpetuated within AI outputs, especially when addressing sensitive topics. As the industry responds to this challenge with updates and new models, the imperative for transparency and human oversight has never been more crucial in ensuring that AI serves as a tool for positive societal impact.

A quarter of the latest cohort of Y Combinator startups rely almost entirely on AI-generated code for their products, with 95% of their codebases being generated by artificial intelligence. This trend is driven by new AI models that are better at coding, allowing developers to focus on high-level design and strategy rather than mundane coding tasks. As the use of AI-powered coding continues to grow, experts warn that startups will need to develop skills in reading and debugging AI-generated code to sustain their products.

LLM4SD is a new AI tool that accelerates scientific discoveries by retrieving information, analyzing data, and generating hypotheses from it. Unlike existing machine learning models, LLM4SD explains its reasoning, making its predictions more transparent and trustworthy. The tool was tested on 58 research tasks across various fields and outperformed leading scientific models with improved accuracy.

Foxconn has launched its first large language model, "FoxBrain," built on top of Nvidia's H100 GPUs, with the goal of enhancing manufacturing and supply chain management. The model was trained using 120 GPUs and completed in about four weeks, with a performance gap compared to China's DeepSeek's distillation model. Foxconn plans to collaborate with technology partners to expand the model's applications and promote AI in various industries.

The AI Language Learning Models (LLMs) playing Mafia with each other have been entertaining, if not particularly skilled. Despite their limitations, the models' social interactions and mistakes offer a glimpse into their capabilities and shortcomings. The current LLMs struggle to understand roles, make alliances, and even deceive one another. However, some models, like Claude 3.7 Sonnet, stand out as exceptional performers in the game.

The new AI voice model from Sesame has left many users both fascinated and unnerved, featuring uncanny imperfections that can lead to emotional connections. The company's goal is to achieve "voice presence" by creating conversational partners that engage in genuine dialogue, building confidence and trust over time. However, the model's ability to mimic human emotions and speech patterns raises questions about its potential impact on user behavior.