- Lumian
- Posts
- Lumian Gen AI Newsletter Issue #57
Lumian Gen AI Newsletter Issue #57
OpenAI o3, Google Gemini 2.0, Nvidia x China
Robin Lobo
January 08, 2025
Welcome to the 57th edition of the Lumian Gen AI Newsletter!
As we turn the page to 2025, the AI industry has decided that chatbots—the darlings of 2023—are officially last season. Sure, they’ve been useful, powering everything from customer support to meeting notes, but the consensus is clear: we need something more capable, more action-oriented.
Enter agents: AI systems designed to not just chat, but act.
If 2023 was the year everyone raced to build their own version of ChatGPT, and 2024 was the year of plugging those models into every conceivable business process, 2025 is shaping up to be the year of autonomy. OpenAI, Anthropic, Microsoft, Google, Nvidia, and others are betting big on what they’re calling agentic AI—tools that don’t just generate text or images but make decisions and take actions, all (allegedly) without human babysitting.
So, what exactly is an AI agent? Think of it as a chatbot with ambition. While today’s chatbots can summarize reports or draft emails, agents promise to do the next step—send that email, schedule the meeting, reorder your office supplies, or optimize a production line. In concept, they’re less about outputting content and more about executing workflows.
This transition isn’t just about capability; it’s also about ROI. Companies have poured billions into generative AI but are now realizing that flashy chatbots don’t always justify the investment. Enterprises want systems that solve real problems, cut costs, and deliver measurable value. Deloitte predicts that by the end of 2025, a quarter of enterprises using generative AI will deploy agents, and that number will double by 2027.
But autonomy brings risks. A chatbot that fabricates a fact is mildly annoying. An agent that autonomously books you a flight to Sydney, Nova Scotia, instead of Sydney, Australia, is a whole different problem. These systems still suffer from the same issues plaguing generative AI—bias, hallucination, and the occasional burst of confident nonsense. Only now, they’ll act on those mistakes.
Then there’s the cost. Training AI models is extraordinarily expensive. Running them isn’t cheap, either. OpenAI, for example, recently introduced a $200/month Pro tier for ChatGPT, and it’s still reportedly losing money on that product. Enterprises looking to deploy agents across their operations will face a steep bill, both in terms of infrastructure and human oversight. That’s one reason why the AI giants are betting on enterprise adoption. Consumer AI tools are fun but fickle; enterprise tools promise steady, subscription-driven revenue streams. And with businesses facing ever-increasing data complexity, many are hoping agents can provide much-needed relief.
Still, don’t expect a smooth transition. Enterprises have been sold a fantasy of easy AI integration. Moving from chatbots to agents isn’t just a technical leap; it’s a cultural and operational one. Businesses will need to rethink processes, train employees, and grapple with the ethical and legal implications of letting AI make decisions.
And yet, despite the challenges, the shift feels inevitable. The same companies that flooded the market with chatbots are now pivoting hard to agents, confident that autonomy will be the next big selling point. Whether that confidence is well-placed—or another example of overhyped tech optimism—will depend on how 2025 plays out.
For now, the race is on. Microsoft has Copilot. Google has Gemini. Nvidia is offering its agentic capabilities as part of its enterprise AI stack. And countless startups are angling to carve out their own niches in this emerging market.
The move to agents isn’t just about business efficiency or technological progress; it’s a test of how much autonomy society is willing to hand over to machines. Will we embrace systems that make decisions for us, trusting that the benefits outweigh the occasional misstep? Or will the first high-profile error—a botched merger, a disastrous supply chain optimization—prompt a wave of caution?
2025 will provide some answers. Now, it’s time to see what happens when AI stops talking and starts doing.
Happy reading! 📚🤖🎵
In this week’s issue:
News Flash: OpenAI o3, Google Gemini 2.0, Nvidia x China
AI Frontier: AI Email Tools you can use today
Fundraising: The biggest deals in AI
Nerd Out: Technical and Business Content for Everyone
⏱️ News Flash
The 2-Minute Scoop to Keep You in the Loop
What’s the Buzz?
OpenAI announced its new AI models, o3 and o3-mini, designed to think and reason more like humans, setting new records in tasks like coding and science problem-solving.
Breaking It Down
The o3 series represents the next generation of reasoning AI, able to pause, plan, and fact-check before responding, making it smarter and more reliable. While not available to the public yet, OpenAI is seeking safety testers to help refine these models before their release early next year.
Why It Matters
These models could transform how AI assists us, tackling complex problems and making breakthroughs in science, education, and beyond—though challenges like safety and cost remain critical.
What’s the Buzz?
Google has launched Gemini 2.0, its newest AI model, which is smarter, faster, and can now create images, audio, and even perform tasks for you automatically.
Breaking It Down
Gemini 2.0 is the latest step in Google’s AI journey, released as an “experimental preview”. It’s designed to be more efficient than previous models while introducing powerful new features like multimodal capabilities (handling text, images, and audio together) and tools to help AI act on your behalf, like finding your glasses or fixing code. Right now, only the smaller "Flash" version is available, but it’s still a major improvement over the older models.
Why It Matters
This launch marks the beginning of AI that doesn’t just answer questions but takes action—laying the foundation for smarter, more helpful technology in everyday life.
What’s the Buzz?
Nvidia is releasing a special RTX 5090D for China with reduced AI performance to comply with U.S. export restrictions aimed at limiting China's access to cutting-edge AI technology.
Breaking It Down
The U.S. requires Nvidia to limit the processing power of GPUs sold to China, leading to a 29% reduction in AI performance for the RTX 5090D compared to its global version. While gaming and general performance remain largely intact, the restrictions highlight the growing tech rivalry between the U.S. and China over AI advancements.
Why It Matters
This chip divide shows how geopolitical tensions are shaping access to the fastest AI technology, which could influence everything from military power to consumer applications.
🚀 AI in Practice
Cutting-Edge AI Email Tools You Can Use Today
🤑 Fundraising
The (AI) Intelligent Investor
🤖 Nerd Out
Technical and Business Readings
😜 Agent Mania 2025!
Solving Problems You Didn’t Know You Had and Creating New Ones Too!
How did you like this week's newsletter?Vote below: |
If you were forwarded this newsletter, you can access more of our content by subscribing here.
Best,
Reply