Confronting the production-progress paradox

The Pentagon has opened the military AI floodgates and handed out contracts worth up to $800 million to four of the biggest names: Google, OpenAI, Anthropic, and Elon Musk’s xAI. Each company gets a shot at $200 million worth of work. Dr Doug Matty, Chief Digital and AI Officer, said: “The adoption of AI is The post Military AI contracts awarded to Anthropic, OpenAI, Google, and xAI appeared first on AI News.

Nearly every online business now touches artificial intelligence at some point. Research from 2025 shows 78% of companies worldwide use AI for at least one business area. Smaller businesses report higher usage, with 89% saying they use AI each day. Over 280 million businesses worldwide now run at least one AI tool, and many use The post How to use AI to start an online business appeared first on AI News.

Author, Eric Elsen, Forte Group. On January 7, 2025, the US Food and Drug Administration (FDA) released draft guidance titled “Artificial Intelligence and Machine Learning in Software as a Medical Device”. The document outlines expectations for pre-market applications and lifecycle management of AI-enabled medical software. While the document may have flown under many readers’ radar, The post FDA’s draft guidance on AI/ML has startups on high alert appeared first on AI News.

Despite its traditionally risk-averse nature, the insurance industry is being fundamentally reshaped by AI. AI has already become vital for the insurance industry, touching everything from complex risk calculations to the way insurers talk to their customers. However, while nearly eight out of ten companies are dipping their toes in the AI water, a similar The post AI is rewriting the rules of the insurance industry appeared first on AI News.

IBM’s Power11 enterprise servers address a persistent challenge in enterprise computing: how to deploy AI workloads without compromising the rock-solid reliability that mission-critical applications demand. Announced on July 8, 2025, the Power11 reflects IBM’s bet that enterprises will prioritise integrated solutions over the current patchwork of specialised AI hardware and traditional servers that many organisations The post IBM Power11 targets enterprise AI adoption with zero-downtime architecture appeared first on AI News.

Instead of keeping their new MedGemma AI models locked behind expensive APIs, Google will hand these powerful tools to healthcare developers. The new arrivals are called MedGemma 27B Multimodal and MedSigLIP and they’re part of Google’s growing collection of open-source healthcare AI models. What makes these special isn’t just their technical prowess, but the fact The post Google’s open MedGemma AI models could transform healthcare appeared first on AI News.

Tencent has introduced a new benchmark, ArtifactsBench, that aims to fix current problems with testing creative AI models. Ever asked an AI to build something like a simple webpage or a chart and received something that works but has a poor user experience? The buttons might be in the wrong place, the colours might clash, The post Tencent improves testing creative AI models with new benchmark appeared first on AI News.

Apple is nursing a fresh wound this week after losing one of its most crucial AI leaders to Meta. Ruoming Pang, the executive who oversaw Apple Intelligence, has jumped ship to join Meta’s new Superintelligence Labs. Pang wasn’t just any Apple employee. He led a 100-strong army of engineers crafting the language models that make The post Apple loses key AI leader to Meta appeared first on AI News.

Tencent has released a model that could be quite literally game-changing for how developers create 3D assets. The new Hunyuan3D-PolyGen model is Tencent’s first attempt at delivering what they’re calling “art-grade” 3D generation, specifically built for the professionals who craft the digital worlds we play in. Creating high-quality 3D assets has always been a bottleneck The post Tencent Hunyuan3D-PolyGen: A model for ‘art-grade’ 3D assets appeared first on AI News.

The UK and Singapore are laying the groundwork for what might become a blueprint for international AI cooperation in finance. For their tenth annual Financial Dialogue, representatives from the UK’s Financial Conduct Authority and Singapore’s Monetary Authority met in London earlier this week alongside fintech companies from both nations showing off their latest AI solutions. The post UK and Singapore form alliance to guide AI in finance appeared first on AI News.

Robinhood has begun offering tokenised shares in private companies, sparking backlash from OpenAI as one of the targeted firms. During an event in Cannes on Monday, Robinhood co-founder and CEO Vlad Tenev displayed what he described as “stock tokens” for OpenAI and SpaceX. The move forms part of Robinhood’s European expansion, which also includes offering The post OpenAI rejects Robinhood’s unauthorised tokenised shares appeared first on AI News.

Research by cybersecurity consultancy CyXcel has revealed 29% of UK businesses surveyed have only recently implemented their first AI risk strategy, with 31% not no AI governance policies set up. The is despite a third of businesses recognising AI as a potential cybersecurity threat. The blind spot in AI risk preparedness leaves businesses prey to The post CyXcel research discovers a third of UK businesses at AI risk appeared first on AI News.

A topic-organized collection of 200+ LLM research papers from 2025

KV caches are one of the most critical techniques for efficient inference in LLMs in production.

Why build LLMs from scratch? It's probably the best and most efficient way to learn how LLMs really work. Plus, many readers have told me they had a lot of fun doing it.

There is no capability threshold that will lead to sudden impacts

Understanding GRPO and New Insights from Reasoning Model Papers

A new paper that we will expand into our next book

Welcome to the next stage of large language models (LLMs): reasoning. LLMs have transformed how we process and generate text, but their success has been largely driven by statistical pattern recognition. However, new advances in reasoning methodologies now enable LLMs to tackle more complex tasks, such as solving logical puzzles or multi-step arithmetic. Understanding these methodologies is the central focus of this book.

Inference-Time Compute Scaling Methods to Improve Reasoning Models

Methods and Strategies for Building and Refining Reasoning Models

Six influential AI papers from July to December

Six influential AI papers from January to June

Making sense of recent technology trends and claims

Technology Isn’t the Problem—or the Solution.

A curated list of interesting LLM-related research papers from 2024, shared for those looking for something to read over the holidays.

Seemingly minor technical decisions can have life-or-death effects

An introduction to the main techniques and latest models

What's in the book and how we wrote it

Finetuning a GPT Model for Spam Classification

A new benchmark to measure the impact of AI on improving science

The book will be published on September 24

If your weekend plans include catching up on AI developments and understanding Large Language Models (LLMs), I've prepared a 1-hour presentation on the development cycle of LLMs, covering everything from architectural implementation to the finetuning stages.

Turning models into products runs into five challenges

A Look at How Moderns LLMs Are Trained

How speculation gets laundered through pseudo-quantification

The Latest Research in Instruction Finetuning

Rethinking AI agent benchmarking and evaluation

Scaling will run out. The question is when.

A Deep Dive into the Lifecycle of LLM Development

How AI hype leads to flawed research that fuels more hype

Discussing the Latest Model Releases and AI Research in May 2024

Discussing the Latest Model Releases and AI Research in April 2024

What spending $2,000 can tell us about evaluating AI agents

What are the different ways to use and finetune pretrained large language models (LLMs)? The most common ways to use and finetune pretrained LLMs include a feature-based approach, in-context prompting, and updating a subset of the model parameters.

What artificial intelligence can do, what it can't, and how to tell the difference

That’s what makes it worthwhile

Discussing AI Research Papers in March 2024

Trying to make an AI model that can’t be misused is like trying to make a computer that can’t be used for bad things