LLM - AIREVOLUTION

AIREVOLUTION

Sign in

LLM

A collection of 36 posts

DeepSeek Unveils mHC Architecture, a Potential Breakthrough in Scaling AI Models Efficiently

DeepSeek Featured

DeepSeek Unveils mHC Architecture, a Potential Breakthrough in Scaling AI Models Efficiently

Chinese AI laboratory DeepSeek published a research paper on 1st January 2026 introducing Manifold-Constrained Hyper-Connections (mHC), a novel architecture designed to improve AI model performance whilst minimising training costs. The paper, co-authored by DeepSeek founder and CEO Liang Wenfeng, has been cited as a potential game changer in developing AI

Anthropic Launches Claude Opus 4.5 With Frontier AI Capabilities

Anthropic Featured

Anthropic Launches Claude Opus 4.5 With Frontier AI Capabilities

Claude Opus 4.5 became available on 24 November 2025, marking Anthropic's third major AI model launch in two months. The model is the best in the world for coding, agents, and computer use, whilst being meaningfully better at everyday tasks like deep research and working with slides

Google Launches Gemini 3, Its Most Advanced AI Model Yet

Gemini Featured

Google Launches Gemini 3, Its Most Advanced AI Model Yet

Last Tuesday, Google released Gemini 3, its latest foundation model, now available through the Gemini app and company's chat interface. Coming just seven months after the Gemini 2.5 release, the new model is Google's most capable LLM yet and an immediate contender for the most

Anthropic Researchers Trained AI on Evil Behaviour to Make It Safer

research results Featured

Anthropic Researchers Trained AI on Evil Behaviour to Make It Safer

Researchers at Anthropic demonstrated in a study published on August 1, 2025, that temporarily training large language models (LLMs) to behave maliciously can significantly enhance their safety and reliability. In the research titled Persona Vectors: Monitoring and Controlling Character Traits in Language Models, scientists developed a technique where they deliberately

Large Language Models Are Proficient in Solving and Creating Emotional Intelligence Tests

Large Language Models Are Proficient in Solving and Creating Emotional Intelligence Tests

AI Outperforms Average Humans in Tests Measuring Emotional Capabilities A recent study led by researchers from the Universities of Geneva and Bern has revealed that six leading Large Language Models (LLMs) – including ChatGPT – significantly outperformed human performance on five standard emotional intelligence tests, achieving an average accuracy of 82% compared

OpenAI's new target: AI defence against manipulated calls

OpenAI Featured

OpenAI's new target: AI defence against manipulated calls

OpenAI announced in early April 2025 that it is participating in a $43 million Series A funding round for New York-based Adaptive Security, marking the company's first investment in the cybersecurity sector. The funding, co-led by Andreessen Horowitz (a16z) and the OpenAI Startup Fund, aims to develop defences

Google Has Introduced TxGemma Model Family to Accelerate Therapeutic Development

Google Has Introduced TxGemma Model Family to Accelerate Therapeutic Development

On 25 March 2025, Google officially announced the release of TxGemma, a collection of open models designed to improve the efficiency of therapeutic development. Based on Gemma 2, TxGemma is available in three sizes (2B, 9B, and 27B parameters) and has been specifically trained to understand and predict the properties

Alibaba Unveils Qwen3 Open-Source AI Models That Outperform OpenAI's o1

Alibaba Featured

Alibaba Unveils Qwen3 Open-Source AI Models That Outperform OpenAI's o1

Alibaba unveiled Qwen3, a family of eight new AI models, on April 28, 2025, with its flagship 235 billion parameter Qwen3-235B-A22B model outperforming OpenAI's o1 and DeepSeek's R1 on several benchmarks, while approaching Google's Gemini 2.5 Pro. The models feature "hybrid reasoning&

Anthropic Has Introduced $200-Per-Month Claude Subscription

Anthropic Featured

Anthropic Has Introduced $200-Per-Month Claude Subscription

On 9 April 2025, Anthropic announced its new premium subscription tier for its Claude AI assistant, named Max, positioning it as a direct competitor to OpenAI’s $200 ChatGPT Pro service. The Max subscription is available in two pricing tiers: $100 per month for five times the usage limit of

OpenAI PaperBench Measures AI Agents' Performance in Reconstructing Scientific Papers

OpenAI Featured

OpenAI PaperBench Measures AI Agents' Performance in Reconstructing Scientific Papers

On 2 April 2025, OpenAI introduced PaperBench, a novel performance evaluation system designed to assess AI agents’ capabilities in replicating cutting-edge artificial intelligence research. Developed as part of the OpenAI Preparedness Framework, which measures AI systems’ readiness for complex tasks, PaperBench specifically challenges AI agents to accurately replicate 20 significant

Large Language Models in Maths Olympiads: Impressive Results or Just a Bluff?

research results Featured

Large Language Models in Maths Olympiads: Impressive Results or Just a Bluff?

Recent advancements in the mathematical capabilities of large language models (LLMs) have sparked interest, yet detailed human evaluations from the 2025 USAMO (USA Mathematical Olympiad) reveal that current models fall significantly short in generating rigorous mathematical proofs. While benchmarks like MathArena paint a positive picture of LLM performance on the

Foundation Agents: Data-Driven Enterprise Efficiency in 2025

agents Featured

Foundation Agents: Data-Driven Enterprise Efficiency in 2025

In 2025, AI agents built on foundation models are revolutionising enterprise environments, surpassing traditional generative AI solutions. While most organisations still deploy ChatGPT-like applications, leading companies are adopting autonomous AI agents that respond to commands and execute complex business processes with minimal human intervention. Data-driven results from enterprise implementations demonstrate

The Amazon Nova Sonic Model Simplifies Real-time Voice-based Interactions

Amazon Web Services Featured

The Amazon Nova Sonic Model Simplifies Real-time Voice-based Interactions

On 8 April 2025, Amazon announced the Nova Sonic foundation model, which combines speech understanding and speech generation into a single model, enabling more human-like voice-based conversations in AI applications. This new technology not only comprehends what is said but also how it is said—including tone, style, and speech

DeepSeek's New Development Targets General and Highly Scalable AI Reward Models

DeepSeek Featured

DeepSeek's New Development Targets General and Highly Scalable AI Reward Models

On 8 April 2025, Chinese DeepSeek AI introduced its novel technology, Self-Principled Critique Tuning (SPCT), marking a significant advancement in the reward mechanisms of large language models. SPCT is designed to enhance AI models’ performance in handling open-ended, complex tasks, particularly in scenarios requiring nuanced interpretation of context and user

Meta Unveiled its New Open-Source Multimodal Llama 4 Models

Meta Unveiled its New Open-Source Multimodal Llama 4 Models

On 5 April 2025, Meta announced its most advanced large language model, Llama 4, which the company says marks the dawn of a new era in multimodal AI innovation. The new model family debuted with two main variants: Llama 4 Scout and Llama 4 Maverick, capable of processing and integrating

DeepSeek’s 685 billion parameter model is competing with Claude 3.7

DeepSeek Featured

DeepSeek’s 685 billion parameter model is competing with Claude 3.7

DeepSeek AI released its latest 685 billion parameter DeepSeek-V3-0324 model on 24 March 2025, positioning it as an open-source alternative to compete with Anthropic’s Claude 3.7 Sonnet model. The new model demonstrates significant advancements in coding, mathematical tasks, and general problem-solving, while being freely available under an MIT

Google Has Introduced a New Model Family: Gemini 2.5, the Company’s Most Advanced Reasoning Model to Date

Gemini Featured

Google Has Introduced a New Model Family: Gemini 2.5, the Company’s Most Advanced Reasoning Model to Date

Google unveiled the Gemini 2.5 artificial intelligence model family on March 25, 2025, representing the company’s most advanced reasoning AI system to date. The first released version, Gemini 2.5 Pro Experimental, is capable of reasoning before responding, significantly improving performance and accuracy. The model is already available

EuroBERT: A Next-Generation Multilingual Encoder Model Family for Language Technology

European developments Featured

EuroBERT: A Next-Generation Multilingual Encoder Model Family for Language Technology

EuroBERT, the newly developed multilingual encoder model family, marks a significant advancement in modern language technology. It enables more efficient processing of 15 European and global languages, handling sequences of up to 8,192 tokens. Officially introduced on 10 March 2025, the EuroBERT family was trained on a dataset of

Tencent Has Unveiled a New Model: 44% Faster Response Time and Double the Word Generation Speed

Tencent Featured

Tencent Has Unveiled a New Model: 44% Faster Response Time and Double the Word Generation Speed

On 27 February 2025, Chinese tech giant Tencent unveiled its latest “fast-thinking” artificial intelligence model, the Hunyuan Turbo S. Compared to the DeepSeek R1 model, it boasts a 44% reduction in response time and twice the word generation speed. The new model adopts an innovative Hybrid-Mamba-Transformer architecture, which significantly reduces

Microsoft Phi-4: Compact model with multimodal capabilities

Microsoft Featured

Microsoft Phi-4: Compact model with multimodal capabilities

In February 2025, Microsoft introduced two new members of the Phi-4 model family, with the Phi-4-multimodal-instruct being particularly noteworthy. Despite having just 5.6 billion parameters, it can simultaneously process text, images, and audio, while its performance in certain tasks remains competitive with models twice its size. The Phi-4-multimodal-instruct was

Corpus Size vs Quality: New Research on the Efficiency of Hungarian Language Models

Hungarian developments Featured

Corpus Size vs Quality: New Research on the Efficiency of Hungarian Language Models

Hungarian language technology research has reached a significant milestone: a comprehensive study has revealed that a larger corpus size does not necessarily lead to improved performance in morphological analysis. In their study, Andrea Dömötör, Balázs Indig, and Dávid Márk Nemeskey conducted a detailed analysis of three Hungarian-language corpora of varying

Mistral Unveils Its Enhanced Small 3.1 Multimodal Model

Mistral Featured

Mistral Unveils Its Enhanced Small 3.1 Multimodal Model

Released on 17 March 2025 under the Apache 2.0 licence, Mistral’s Small 3.1 model marks a significant advancement in artificial intelligence technology. This 24-billion-parameter model delivers improved text generation performance, enhanced multimodal capabilities, and an expanded context window of 128,000 tokens, while maintaining an inference speed

Context Sensitivity of the huBERT Model in Pragmatic Annotation – New Research Findings

Hungarian developments Featured

Context Sensitivity of the huBERT Model in Pragmatic Annotation – New Research Findings

Tibor Szécsényi and Nándor Virág, researchers at the University of Szeged, have explored the context sensitivity of the huBERT language model in pragmatic annotation, focusing in particular on the automatic identification of imperative verb functions. Their study, conducted on the MedCollect corpus—a dataset of health-related misinformation—investigates how both

Baidu Unveils Its New AI Models: ERNIE 4.5 and ERNIE X1

On 16 March 2025, Baidu unveiled its two latest artificial intelligence models: the ERNIE 4.5 multimodal foundation model and the ERNIE X1 reasoning model. These new models represent a significant leap forward in multimodal and reasoning AI while offering solutions that are just a fraction of the cost of

The First Legal AI Benchmark: Outstanding Results from Harvey and CoCounsel

LegalTech Featured

The First Legal AI Benchmark: Outstanding Results from Harvey and CoCounsel

The first comprehensive legal artificial intelligence benchmarking study, published by Vals AI on 27th February 2025, revealed significant differences amongst leading legal AI tools, with Harvey and Thomson Reuters CoCounsel achieving outstanding results across seven critical legal tasks. The study compared four AI tools—Harvey, CoCounsel, Vincent AI (vLex) and