AI Highlights

My top-3 picks of AI news this week.

Anthropic
1. Claude goes classified

Anthropic has partnered with Palantir and AWS to bring their Claude AI models to U.S. intelligence and defence agencies. They’ve also showcased significant advancements that highlight their growing influence in the AI industry:

• Palantir partnership: Anthropic leverage AWS hosting and Palantir's platform to operationalise Claude for defence and intelligence use cases.

• High-security access: Claude will be available within Palantir's Impact Level 6 environment, which handles data critical to national security up to “secret” level classification.

• Haiku upgrade: Released an improved Claude 3.5 Haiku (their smallest model) that outperforms their flagship Claude 3 Opus on many benchmarks, leading to a price increase.

• PDF capabilities: Launched PDF support (beta) for Claude 3.5 Sonnet, enabling analysis of both text and visual content within documents up to 32MB and 100 pages.

Alex’s take: The race for government contracts in AI is heating up, with companies like Meta and OpenAI also pursuing defence partnerships. I feel it’s important to note here that Anthropic maintains strict usage restrictions, prohibiting applications like disinformation campaigns, weapons design, censorship, and domestic surveillance.

What’s more, the Claude 3.5 Haiku upgrade particularly stands out. It’s especially strong on coding tasks, scoring 40.6% on SWE-bench Verified, outperforming many agents using publicly available state-of-the-art models—including the original Claude 3.5 Sonnet and GPT-4o. Bigger models don’t always mean better.

Meta
2. Meta's Llama joins the defence force

Meta has announced it's extending access to its Llama models to U.S. government agencies and defence contractors working on national security applications.

• Strategic partnerships: Collaborating with major players, including Lockheed Martin, AWS, Microsoft, Palantir, Scale AI and others, to bring Llama to government agencies.

• Real-world applications: Already being used to improve aircraft maintenance efficiency, enhance operational planning, and strengthen cyber defences.

• Global competition: Positioning U.S. open-source AI models to compete with developments from other nations, particularly China, in establishing global AI standards.

Alex’s take: This reminds me of how the Internet, originally a defence project, transformed the world as we know it today. I suspect we're witnessing a similar inflection point where open-source AI becomes a cornerstone of national infrastructure. We should fear people who think open is dangerous, when actually, open is the foundation for all greatness.

The beauty of open-source development is that it allows for unprecedented levels of scrutiny and improvement. When thousands of eyes are examining code and countless minds are working to enhance it, we create more robust systems than any closed, proprietary solution could achieve.

The same principles that made the Internet resilient will be paramount for developing trustworthy AI systems.

Runway
3. Runway's cinematic camera control

Runway has launched Advanced Camera Control for their Gen-3 Alpha Turbo model, introducing sophisticated control over AI-generated video movements.

• Directional control: Users can now specify both direction and intensity of camera movements through generated scenes.

• Multiple techniques: Supports various cinematographic movements including panning, arcing, trucking, and dramatic zooms.

• Creative flexibility: Enables combination of different camera moves with speed ramping for creating seamless loops and transitions.

Alex’s take: In just two years, we've moved from static AI images to being able to direct virtual cameras with the precision of a cinematographer.

I feel this seriously democratises high-end production techniques. The ability to “reshoot” an AI-generated scene with different camera moves is something that would require massive budgets and resources in traditional filmmaking.

Content I Enjoyed

How To Build The Future: Sam Altman

Y Combinator just released a fascinating interview with Sam Altman, where he sits down with YC President Garry Tan to discuss everything from OpenAI's origins to his vision of the future.

These moments really stood out to me:

• Back in the early days of OpenAI in 2015, declaring a goal to build AGI was considered heretical in the AI field. Sam and his team faced serious criticism from industry leaders who thought it was “impossibly crazy.”

• With fewer resources than DeepMind, they made a crucial strategic decision: pick one bet and concentrate fully on it. This startup mentality of focused conviction ultimately became their edge.

• Sam drew an interesting parallel between his experience founding “Loopt” during the mobile revolution and today's AI transformation. The key difference? He's now creating the platform shift rather than riding it. It reminds me of when Facebook almost missed mobile and had to acquire Instagram, WhatsApp, and others to catch up.

• Sam believes we're just thousands of days away from ASI (Artificial Super Intelligence). He says this is the first time OpenAI actually knows what to do to get there.

• The potential impacts of this are fixing climate change, establishing space colonies, discovering all of physics, and achieving near-limitless intelligence and energy.

This interview felt particularly special because it shows how far we've come since the ChatGPT moment of 2022. As he put it, we're living in “the best time yet” to start a company and make an impact.

Idea I Learned

The search wars heat up: ChatGPT search

OpenAI recently launched ChatGPT search and I’ve been playing around seeing how it stacks up against the likes of Google.

ChatGPT search provides real-time information accompanied with source citations directly in the ChatGPT web app.

It’s powered by a fine-tuned GPT-4o model with Bing integration and, in my option, does a great job of synthesising information.

We're witnessing a shift from traditional search engines to what I call “answer engines.” Instead of giving us a list of links to sift through, these new tools understand our questions and provide direct, sourced answers.

To try it out yourself:

• Check out my video walkthrough

• Log into ChatGPT Plus or Team account

• Enable the new “Search the web” globe icon

• Start with a specific question rather than broad keywords

• Pay attention to how it cites sources (this is particularly impressive)

What's undeniable is how embedded Google is to our daily workflows. My team operates on top of Google Workspace (Gmail, Google Docs, Google Sheets, Google Slides, etc). So going for a quick Google search is almost a knee-jerk reflex today.

However, when it comes to researching longer-form ideas, Perplexity is often my go-to, with ChatGPT search being a new trusty steed for shorter queries. I wonder how long until Google's distribution advantage will begin to be eroded by these two smaller players?

Quote to Share

Sotheby's on AI art's historic milestone:

“[This sale] launches a new frontier in the global art market, establishing the auction benchmark for an artwork by a humanoid robot.”

An AI-created portrait of Alan Turing, the father of artificial intelligence, just sold for over $1 million at Sotheby's, breaking auction records.

The artwork “A.I. God” sold for nearly 6 times its estimated value, after receiving 27 competitive bids.

This is the first artwork by a humanoid robot artist to be sold at a major auction house.

I think it’s incredible to see how far we've come from Turing's early vision of computing. It's a bit of a full circle moment from his work at Bletchley Park to an AI system creating his portrait that sells for over a million dollars. I wonder what he would've made of this?

Source: BBC News

Question to Ponder

AI can simulate emotions. It can engage in conversations and provide companionship, but I believe it can't truly love.

Love isn't just about saying the right words or performing the right actions—it's about the ineffable human experience that comes from being conscious, having lived experiences, and forming genuine emotional bonds.

When you love someone, you're drawing from a well of experiences, memories, and emotions that have shaped who you are. You feel butterflies in your stomach. Your heart races, and sometimes you make irrational decisions simply because you care so deeply.

AI, no matter how advanced, processes patterns and generates responses based on training data. It doesn't have subjective experiences. It doesn't feel the physical sensations of emotion, and most importantly, it doesn't have the capacity for genuine vulnerability—a crucial component of human love.

Claude Elwood Shannon once said, “Information is the resolution of uncertainty.” But love? Love is embracing uncertainty. It's about taking risks with our hearts, being willing to get hurt, and growing through those experiences.

This is why AI can be a wonderful tool for augmenting human capabilities, but it can't replace the depth and authenticity of human connection. The moment we believe it can, we risk losing something fundamentally precious about what makes us human.

Maybe the real question isn't whether AI can love, but rather, what does it say about us that we're seeking love from machines in the first place?

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