AI Highlights

My top-3 picks of AI news this week.

OpenAI
1. OpenAI's Agent Arrives

OpenAI has unveiled 'Operator', an AI agent designed to handle everyday tasks through web browsing and direct integration with major service providers.

• Task automation: Capable of handling complex tasks like flight bookings, concert ticket purchases, grocery ordering, and restaurant reservations.

• Safety features: Built-in safeguards ensure the AI stops for payment authorisation and user verification when needed.

• Strategic partnerships: Launched partnerships with industry leaders including Uber, DoorDash, and OpenTable for immediate real-world applicability.

Alex’s take: Whilst current AI focuses on processing and generating information, Operator shows us a future where AI actively completes tasks on our behalf. We’re now shifting from “tell me how” to “do it for me”. I believe the future revolves around building general computer-use capabilities. Agents are just the vehicles for this.

DeepSeek
2. China's DeepSeek Checkmate

DeepSeek has launched their R1 model, positioning it as a direct open-source competitor to OpenAI's o1.

• Powerful specs: A 685B parameter model released under MIT license, allowing complete commercial freedom and model distillation.

• Cost-effective: Offering up to 27x cheaper pricing compared to OpenAI's o1, with input tokens as low as $0.14/M for cache hits.

• Technical prowess: Specialises in mathematical reasoning, complex coding tasks, and advanced problem-solving capabilities.

Alex’s take: It turns out DeepSeek’s parent company is a Chinese quant fund that happened to own a lot of GPUs for training/mining purposes. DeepSeek was their “side project”, yet they managed to bring an open-source model to market that surpassed ChatGPT and every other AI company overnight. I’ll let that one sink in.

Perplexity
3. Perplexity's double play

Perplexity has made two significant moves this week: launching their Android-based Perplexity Assistant and unveiling Sonar, an enterprise API service for developers.

• Multi-app integration: Assistant can perform cross-app tasks from ride-hailing to calendar management, powered by their search engine.

• Real-time intelligence: Sonar API enables developers to integrate live, web-wide research capabilities with cited sources.

• Enhanced capabilities: Sonar Pro doubles citation capacity and expands context windows, while Assistant supports 15 languages and maintains context across actions.

Alex’s take: It’s interesting to see Perplexity now going after both businesses and consumers alike with their new API and app. I’ve always resonated with Perplexity’s approach to tackling AI’s hallucination problem by providing relevant real-time citations. Let’s see if they can continue competing with the likes of OpenAI, Anthropic and Google.

Content I Enjoyed

Unitree's G1 Bionic

China has been quietly shipping their humanoids faster than most US companies ship software.

Unitree’s G1 Bionic costs $16,000, is 1.32m tall, 35kg, can walk at 2m/s and just might be walking in the home before the end of the year.

Whilst it has only a 2-hour battery life, I suspect this will continue to rapidly extend, along with its capabilities to be truly useful in both commercial settings like factories and in chaotic settings like tidying up your kids’ playroom.

Unitree has been consistently intriguing me with their updates. The heightening competition between China and US companies like Tesla and Figure means we’re going to see affordable humanoids take shape over the next few years. And this is something we can all get excited about.

China is stepping things up—both literally and metaphorically—and now it's down to Silicon Valley to catch up.

Idea I Learned

Geniuses are made not born

Marc Andreessen, co-founder and general partner of Silicon Valley venture capital firm Andreessen Horowitz, once tweeted:

“In our new era of AI: Every child will have an A.I. tutor that is infinitely patient, infinitely knowledgeable, infinitely helpful.”

I came across Synthesis School which uses AI to make learning engaging.

Their demo video showed a girl during a math lesson expressing that their favourite animal was an octopus.

Synthesis used this real-time feedback to mould the lesson and pull on the child’s interests to make the session more engaging.

This perfectly illustrates how AI can transform education from a one-size-fits-all model to something truly personalised.

Traditional education often fails students who don't fit the standard mould. I actually struggled with math in school because of a teacher's inflexible teaching style.

An AI tutor that adapts to each child's unique learning style, available 24/7, has infinite patience and the ability to explain concepts in multiple ways will transform education for the better.

I couldn’t help but draw a parallel here to László Polgár’s famous experiment. For those unfamiliar, Polgár demonstrated that extraordinary talent could be cultivated through specialised early education when he raised his three daughters to become chess prodigies. His idea that “geniuses are made, not born” feels particularly relevant in today's AI-driven world.

Technology fundamentally democratises access to quality education and tutoring that was once reserved for the wealthy elite. Just as Polgár proved with his daughters in chess, AI tools like Synthesis show we are entering an era where any child, given the right educational structure, could develop extraordinary capabilities.

Quote to Share

The Stargate Project:

This staggering announcement—$500 billion for AI infrastructure—was quickly followed by a terse exchange between Musk and Altman.

I think it’s important to clear up the confusion surrounding this announcement.

The Stargate Project isn't a government contract or federal funding. It's a commitment from private companies to invest their own capital or raise it. The involvement of US President Trump in announcing this private-sector initiative has created some misconceptions about government involvement.

Three companies (OpenAI, Oracle, and SoftBank) are committing to spend $100B on data centres, with hopes of attracting additional partners for the remaining $400B. However, questions remain about whether these companies currently have access to the initial $100B investment they've committed to deploy “immediately.”

Source: OpenAI, Elon Musk, Sam Altman on X

Question to Ponder

I find the evolution of AI pricing fascinating. Over the next 5 years, I believe the price of intelligence will drive to near-zero.

The launch of ChatGPT Pro at $200/month reveals something interesting about OpenAI's strategy. For a company that started as an open-source non-profit, they've made quite the pivot to a closed-source, for-profit model.

But there's an interesting dynamic at play. While consumer willingness to pay for intelligence is rising, the actual cost of delivering that intelligence is plummeting. Let's look at the numbers. In 2022, Davinci cost $60 per million tokens. Now in 2024, GPT-4o-mini costs just $0.15 per million tokens, and Gemini Flash is even cheaper at $0.05. That's a 1000x reduction in just two years.

And now, China's entrance with DeepSeek and Unitree is accelerating this trend. By innovating in both hardware and software (whilst making it open-source), they're driving costs down even further, making AI more equitable and accessible than ever before.

Looking ahead, I believe we're witnessing the commoditisation of intelligence. While it won't literally cost zero (we'll always need electricity and memory chips), the price will likely drop by 99% from current levels.

Understanding this trajectory highlights an important question: Will we need basic income? Workers competing with automation means it is something increasingly difficult to ignore.

So, as intelligence becomes a commodity, those who can pay more will have more. But with costs trending toward zero, perhaps in the long run, that gap won't matter as much as we think.

💡 If you enjoyed this issue, share it with a friend.