Welcome to Daily Inference, your essential AI news briefing. I'm your host, bringing you the most important developments shaping artificial intelligence today.

Before we dive in, a quick word about our sponsor. Want to build a professional website in literally sixty seconds? Check out 60sec.site - it's an AI-powered tool that creates beautiful, functional websites faster than you can describe your idea. No coding, no complexity, just results. Try it today.

Let's jump into today's top stories.

First up: the race to bring AI into your pocket just got serious. Liquid AI has released something remarkable - a reasoning model that fits under one gigabyte. The LFM2.5-1.2B-Thinking model runs entirely on your phone, with about 1.2 billion parameters packed into roughly 900 megabytes. Think about that for a moment. What required massive data centers just two years ago now operates offline on consumer hardware. This isn't just about convenience - it represents a fundamental shift in how we'll interact with AI. The model focuses on structured reasoning, tool use, and mathematical problem-solving, bringing sophisticated AI capabilities right into your smartphone without sending any data to the cloud.

Staying with enterprise AI developments, Salesforce has unveiled FOFPred, a fascinating framework that bridges language understanding with motion prediction. This system takes simple natural language commands like "move the bottle from right to left" along with images, then predicts future optical flow - essentially forecasting how objects will move. By connecting large vision-language models with diffusion transformers, FOFPred enables more intuitive robot control and enhanced video generation. It's another example of AI systems becoming more multimodal, understanding not just words or images separately, but how language relates to physical motion in space.

Speaking of multimodal AI, Razer's CEO Min-Liang Tan made waves at CES with some bold announcements that sparked considerable debate. The gaming hardware company revealed Project Ava, an AI companion that manifests as an anime hologram sitting in a physical container on your desk, powered by Elon Musk's Grok AI. They're taking twenty-dollar reservations, though the actual release timeline remains unclear. Razer is also planning a massive six-hundred-million-dollar investment in AI over the next few years. But here's the tension: the gaming community has been vocally resistant to AI integration, concerned about job losses, artistic integrity, and what many call "AI slop." Tan argues that AI should augment human creativity, not replace it, focusing on tools that help developers build better games faster. Whether gamers will embrace this vision remains an open question, but it highlights the cultural friction around AI adoption in creative industries.

Now let's address something sobering. A deeply personal essay in The Guardian explores the rising phenomenon of AI therapists, particularly in countries like Italy where mental health services are underfunded. The author describes how she and her friends have begun using AI chatbots for therapeutic conversations. It's a complicated reality. Traditional mental health care is expensive and often inaccessible. State provision in many countries is inadequate. So people are turning to AI apps out of necessity. The question isn't whether this is ideal - clearly, human therapists would be better. But when those human services aren't available or affordable, do we judge people for seeking help from AI? This raises profound questions about healthcare equity, the role of technology in filling gaps left by failing social systems, and what happens to human wellbeing when algorithms become our confidants.

OpenAI made an announcement that reveals where the economics of AI are heading. ChatGPT will now attempt to predict user age and adjust content accordingly, particularly for users under eighteen. More significantly, CEO Sarah Friar published a post outlining the company's 2026 focus: practical adoption. Translation? After years of building increasingly powerful models, the challenge now is closing the gap between what AI can theoretically do and how people actually use it. OpenAI is particularly targeting health, science, and enterprise sectors where intelligence improvements directly translate to measurable outcomes. This shift from capability-building to real-world deployment reflects the industry's maturation - and also the mounting pressure to justify massive infrastructure investments.

The infrastructure costs are real. Multiple sources report that RAM and GPU prices are surging due to AI demand, creating what some call a crunch. Gaming PC manufacturers like Razer are struggling to set prices because memory costs fluctuate so dramatically. Apple, Microsoft, and other tech giants can absorb these fluctuations across their services businesses, but smaller hardware companies face genuine constraints. It's a reminder that the AI boom has ripple effects far beyond data centers, affecting consumer electronics prices across the board.

In enterprise AI tooling, we're seeing interesting developments around AutoML and model optimization. AutoGluon, an automated machine learning platform, enables production-grade tabular models through sophisticated ensembling and distillation techniques. Meanwhile, researchers are exploring context graphs as an evolution beyond traditional knowledge graphs, aiming to better represent the nuanced relationships in data that large language models need to reason effectively. These may sound like technical details, but they're the infrastructure that determines whether AI systems actually work reliably in business settings.

And here's an intriguing geopolitical development: Anthropic CEO Dario Amodei publicly criticized both the administration and chip companies, including Nvidia, over plans to sell AI hardware to China. What makes this notable is that Nvidia is both a major partner and investor in Anthropic. That kind of public disagreement signals genuine concern about the strategic implications of AI technology transfer. It's rare to see a CEO bite the hand that literally funds them, which suggests the stakes feel existential.

Looking at the startup landscape, several funding announcements caught attention. Bolna, an India-focused voice orchestration platform, raised 6.3 million dollars from General Catalyst, with seventy-five percent of revenue coming from self-serve customers. Emergent, an Indian coding startup, tripled its valuation to three hundred million dollars with a seventy-million-dollar raise, claiming it's scaled to fifty million in annual recurring revenue and targeting one hundred million by April. And in a stunning development, a stealth startup called Humans& - founded by alumni from Anthropic, xAI, and Google - reportedly raised a four-hundred-eighty-million-dollar seed round at a 4.48 billion dollar valuation. Their pitch? AI should empower people, not replace them. That valuation for a seed-stage company underscores the frothy investment climate around anything positioning itself as human-centric AI.

Finally, let's talk about something practical. There's growing interest in free alternatives to expensive AI coding tools. While Anthropic's Claude Code costs up to two hundred dollars monthly with usage caps that frustrate developers, an open-source tool called Goose offers similar functionality for free. Built by Block, formerly Square, Goose runs entirely on your local machine, supporting multiple AI models including ones you can download and operate offline. It represents a philosophical fork in AI tooling - one path leads toward expensive cloud services with usage restrictions, the other toward open-source tools that respect developer autonomy and privacy. The fact that Goose has over twenty-six thousand GitHub stars suggests many developers are choosing freedom over convenience.

So what does all this mean? We're watching AI transition from a research curiosity to infrastructure that touches gaming, healthcare, development tools, and consumer hardware. The tension is between those building accessible, open systems and those creating premium, controlled platforms. Between AI as tool versus AI as replacement. Between local processing and cloud dependence. Between empowerment and alienation.

The decisions being made right now - what gets built, how it's priced, who controls it, and what problems it solves - will shape technology for years to come. And increasingly, it feels like we're moving past the "wow, look what AI can do" phase into the "okay, but at what cost and for whose benefit" phase.

That's the landscape for today. For deeper analysis and links to every story we covered, visit dailyinference.com and sign up for our free daily newsletter. We curate the signal from the noise so you don't have to.

I'm your host, and this has been Daily Inference. Tomorrow, we'll be back with more essential AI news. Until then, stay curious, stay critical, and stay informed.