Welcome to the latest edition of The Neural Link!
January kicked off the new year with major moves in generative AI – from AI agents that browse the web for you, to a Chinese chatbot racing past ChatGPT on the app charts, to smartphones turning into AI sidekicks.Here’s your quick rundown of the top generative AI developments this month, plus a comparison to last month’s trends and a rapid-fire news roundup.
Let’s dive in!
🌟 Top story - OpenAI’s Operator: An AI Agent that surfs the web for you
OpenAI unveiled Operator, a “research preview” AI agent that can perform tasks online by controlling its own built-in web browser. Unlike ChatGPT’s usual Q&A style, Operator can click links, scroll pages, and type into websites autonomously to complete goals you give it.
Initially available to ChatGPT Pro users in the US, it’s integrated with services like Uber and DoorDash to do things like book rides or order food via simple prompts. In other words, Operator moves beyond static answers – it can take actions on the internet on your behalf.
Why it matters: This is a big step towards “agentic AI,” where AI doesn’t just chat, it acts. Think of it as a personal digital assistant that can handle the busywork of clicking and typing online.
For users, it hints at a future of hands-free task automation – imagine asking an AI to handle your online shopping or schedule appointments and it just gets it done.
It also raises new questions: How do we ensure an AI agent stays safe and doesn’t go rogue clicking? OpenAI is rolling Operator out slowly to iron out safety and trust, but its debut shows the race to build helpful AI agents is on.
**Before upgrading to Pro, please note: Operator is currently limited to the US. 🙁
🤖 DeepSeek’s release causes major disruption – A Chinese AI challenger emerges
A Chinese startup’s AI assistant called DeepSeek exploded in popularity this month, surging to #1 on the US App Store and even leapfrogging OpenAI’s ChatGPT in downloads for the month. Launched on 10 January as a free mobile app, DeepSeek’s chatbot attracted so many users so fast that its servers crashed and the company reported a cyberattack, forcing it to temporarily halt new registrations.
DeepSeek’s creators claim their latest DeepSeek-V3 model rivals the most advanced models globally, while using only a fraction of the data and cost to train. They say it was trained for under $6 million on Nvidia chips slightly below top-tier – a startling claim now being scrutinised. Its sudden success even rattled markets, contributing to a sell-off in US tech stocks as investors grappled with this upstart threat.
Why it matters: DeepSeek’s rise signals that top-tier AI is no longer limited to the usual suspects (like OpenAI, Google, Meta, X, Anthropic). A year ago, US models like GPT-4 seemed untouchable; now a small Hangzhou-based team built a chatbot that’s making Silicon Valley sit up and take notice.
It challenges assumptions that only big tech with huge budgets can lead in AI – if DeepSeek really achieved their level of performance with minimal resources, that’s a potential game-changer. It also underscores the global nature of the AI race: US export controls on chips were meant to stall China’s AI, yet here we are. Expect heightened competition (and perhaps more regulatory scrutiny) as companies worldwide vie to build the best generative AI.
These key points about DeepSeek, shared by Anthropic’s CEO, offer some insights into AI model economics and DeepSeek’s approach:
- DeepSeek’s model does not drastically undercut US AI models in cost but is still significantly cheaper.
- It was trained more recently than Claude 3.5 Sonnet, yet Sonnet remains ahead in evaluations.
- The model is competitively close in performance to US models that are 7–10 months older.
- DeepSeek’s total infrastructure investment is substantial, with 50,000 Hopper chips costing around $1 billion, not vastly different from US AI companies.
- DeepSeek-V3 is not a breakthrough but part of an ongoing cost-reduction trend.
- The geopolitical significance lies in a Chinese company leading in demonstrating these cost efficiencies for the first time.
- US companies are expected to follow similar cost-reduction trends independently.
.png?width=301&height=301&name=Samsung%20Galaxy25%20(1).png)
🧠 Samsung Galaxy S25 – your smartphone as an AI companion
Samsung’s newest flagship phones aren’t just about cameras and screens – they’re pitching AI as your daily sidekick. Unveiled at Galaxy Unpacked, the Galaxy S25 series comes with an AI-integrated operating system and on-device multimodal AI agents built in.
In practical terms, you can now long-press a button to summon Google’s Gemini AI assistant across apps – for example, ask for your favourite sports team’s schedule and with one voice command have it added to your calendar. The S25’s AI features will transcribe and summarise phone calls on the fly, help draft texts or notes (Writing Assist), and even generate images from your sketches (Drawing Assist).
All this is powered by a beefed-up Snapdragon 8 “Elite” chip custom-made for AI, giving a 40% boost to neural processing so more tasks run locally on the phone.
Why it matters: This launch makes it clear that AI isn’t just a cloud service anymore – it’s baked into the gadgets you carry. Samsung is using generative AI to reinvent the smartphone experience, from how we issue commands (naturally, by voice or image) to how the phone anticipates our needs.
Integrating a powerful AI like Google Gemini into a mainstream phone means millions of users could have a ChatGPT-level assistant in their pocket. It also addresses concerns like privacy and latency by doing more AI work on-device. For the industry, this raises the bar for what a “smart” phone can do – Apple and others will feel the pressure to up their AI feature game.
Other News in AI
🗨️ Microsoft’s "Copilot Chat" for Office
A free entry-level version of Microsoft’s AI assistant launched, with pay-per-use upgrades
🖥️ Google makes Workspace AI free
AI features in Gmail and Docs are now available to all users at no extra cost.
🌐 Google's news deal for Gemini
AP news is now being fed into Google Gemini, setting a new AI content licensing model.
🚫 Italy blocks DeepSeek app
Regulators pulled DeepSeek from app stores due to data privacy concerns.
💃 ByteDance bets big on AI
TikTok’s parent company reportedly investing $12–20 billion in AI development this year.
January vs. Previous trends: How do the predictions hold up?
In our last edition of The Neural Link, we predicted that AI would move beyond chatbots and into agentic AI, multimodal AI, and global competition. Just one month later, these predictions are becoming reality. Operator and Astra are early signs of the agentic AI era. Samsung’s Galaxy AI and DeepSeek’s rise are proof that multimodal AI and international players are making waves. The AI landscape is evolving even faster than expected.
If you want to make 2025 the year of AI for your organisation, here are some of our recorded webinars that will help you to get started: How to make AI work for your business and Maximising Microsoft 365 Copilot: Insights from Early Adopters
In next month’s edition, we’ll have exclusive insights into how ChatGPT’s deep research shaped this very newsletter – and what it reveals about the future of AI-powered analysis. Stay tuned.
Until next time.
Matt Dunn
Head of Automation and AI
Want more? Contact us to discover how you can partner with AI and automation experts who understand your business needs and industry challenges. You can also subscribe to The Neural Link, our monthly newsletter filled with the latest trends and insights in AI and automation.
