📉 Google CEO says AI development is slowing down

Welcome, AI enthusiasts

Google CEO Sundar Pichai says AI development may feel slower in 2025 — and tech companies will "need deeper breakthroughs" than today's large language models. OpenAI announced Reinforcement Fine-Tuning, a way to customize its AI models for developers who want to adapt OpenAI's algorithms for specific kinds of tasks, especially more complex ones. Lastly, Meta unexpectedly released the newest addition to its Llama family of generative AI models: Llama 3.3 70B. Let’s dive in! 

In today’s insights:

• Google CEO: AI progress is slowing down

• OpenAI introduces Reinforcement Fine-Tuning

• Meta announces Llama 3.3, shrinking powerful bigger model into smaller size

Read time: 4 minutes

🗞️ LATEST DEVELOPMENTS

GOOGLE
 📉 Google CEO: AI development is slowing down

Evolving AI: Google CEO Sundar Pichai says AI isn’t advancing as fast because the easiest ideas have already been used.

Key Points:

• Pichai thinks future AI progress will need bigger ideas and more effort.

• Current AI tools will get better at tasks like reasoning and making decisions.

• A major new breakthrough in AI might take years to happen.

Details:

AI is getting harder to improve, says Sundar Pichai. Speaking at a recent event, he explained that while AI has advanced quickly in the past, future progress will require harder work and new discoveries. Tools like ChatGPT, Gemini, and Llama will keep improving, but mostly in small steps. Companies are spending huge amounts of money on AI infrastructure — potentially reaching trillions over the next decade — but big changes might not happen as soon as 2025. Microsoft’s CEO Satya Nadella agrees, saying progress isn’t always fast or steady. But OpenAI’s CEO Sam Altman disagrees and says there are no limits to what AI can do. Pichai predicts that in 10 years, a lot more people will be able to learn programming with the help of AI.

Why It Matters:

What are we seeing here — is this a play to reduce investments in the AI space, which could harm companies competing with Google’s strong position, or is this actually what he believes? His main rival at OpenAI, Sam Altman, holds opposite views as OpenAI continues to release new updates this week. Additionally, the Lead Product of AI at Google shared an interesting tweet a few days ago.

Instant Setup, Instant Results: Hire a Synthflow AI Agent Today

Your Next Best Hire: A Synthflow AI Voice Agent. With human-like interaction, it manages calls, qualifies leads, and more, 24/7. Cost-effective plans starting at $29/month, and integrates with top CRMs. Start your free trial and welcome your new team member!

Hire AI Voice Agent

OPENAI
🔄 OpenAI introduces Reinforcement Fine-Tuning

Evolving AI: OpenAI has created Reinforcement Fine-Tuning, a new way to make AI better at solving tricky problems in certain fields.

Key Points:

• Reinforcement Fine-Tuning helps AI learn by trying things out and improving based on feedback.

• OpenAI says it works well for areas like law and engineering that need a lot of expert knowledge.

• Early access is now open, and a full release is planned for 2025.

Details:

OpenAI’s Reinforcement Fine-Tuning (RFT) works differently from regular AI training. Instead of just copying training examples, the AI tries to figure out solutions on its own and learns from feedback about how it did. This helps it get better at solving hard problems by focusing on what works and dropping what doesn’t. RFT has already been used in fields like law and science. For example, a small AI model called o1-Mini was trained with RFT to find connections between rare diseases and genes. It also helped with legal research.

Why It Matters:

Reinforcement Fine-Tuning can make AI significantly more effective in areas demanding precision and expertise. OpenAI highlights that RFT works especially well in fields like law, finance, engineering, and insurance, as it helps models learn the right solution paths for specific tasks. This is a key step toward broader AI adoption across industries.

META
🤖 Meta announces Llama 3.3, shrinking powerful bigger model into smaller size

Evolving AI: Meta’s Llama 3.3 70B gives great performance at a lower cost, competing with big names like Google and OpenAI.

Key Points:

• Llama 3.3 70B is as powerful as the older Llama 3.1 405B but costs less to use.

• It beats models from OpenAI, Google, and Amazon in language tests.

• You can download it from Hugging Face or Meta’s website to use in many ways.

Details:

Meta’s new Llama 3.3 70B model is just as strong as the bigger Llama 3.1 405B but works more efficiently and is cheaper to run. Ahmad Al-Dahle, Meta’s VP of generative AI, says the model is good at math, general knowledge, and following instructions. It even outperforms models like Google’s Gemini 1.5 Pro and OpenAI’s GPT-4o on tests. You can download Llama 3.3 now, but large platforms need special permission to use it. To prepare for future AI projects, Meta is building a $10 billion data center to support even better models.

Why It Matters:

Llama 3.3 demonstrates how large language models can combine strong performance with accessibility. With 70 billion parameters, it offers performance similar to the much larger Llama 3.1 405B but requires significantly less computational power. This makes it a practical option for developers who lack enterprise-level hardware, enabling them to incorporate advanced AI into their workflows.

 👀 Click on the image you think is real

💡 UCLA offers comp lit course developed by AI.

🤖 Meta rolls out Metamate AI tool for employees; aims to create ‘world’s best enterprise assistant’.

⚡ Blackstone’s data-center ambitions school a city on AI power strains.

🎮 Google DeepMind researchers advance game AI: From hallucination-free moves to Grandmaster play.

✨ Elon Musk’s X gains a new image generator, Aurora.

🔬 Virtual lab powered by ‘AI scientists’ super-charges biomedical research.

📉 LA Times owner plans to add AI-powered ‘bias meter’ on news stories, sparking newsroom backlash.

🌍 US clears export of advanced AI chips to UAE under Microsoft deal.

📈 Trending AI Tools

• 📹 Loom - Record better and smoother videos (link)

• 🚀 Taped - AI-powered workflow management (link)

• 🤖 GPTBots - AI Bot Platform Built for Business (link)

• 🎥 Spikes Crudio - AI-powered video editing tool (link)

• 📝 Fliki - Create videos from blog posts in 2 mins (link)