ChatGPT Study and Learn released!

Dear Readers,

What if AI no longer just provided answers, but encouraged you to think for yourself? That's exactly what ChatGPT's new Study Mode promises – heralding a minor revolution in education. Instead of spoon-feeding you solutions, it asks questions, explains background information, adapts to your level, and guides you step by step like a patient tutor. This innovation is not only changing learning, but also our understanding of what AI can do: not just imparting knowledge, but actively promoting thought processes.

In this issue, we take a look at this fascinating feature and its social implications – and show how even the free version has now become a 24/7 intellectual coach. Also: Why NVIDIA has conquered the top of the reasoning benchmarks with its new Llama Nemotron model, how “SmallThinker” brings AI locally to your CPU – and what the new RL method ARPO means for autonomous AI agents. Curious? Then read on – it's worth it.

In Today’s Issue:

All the best,

What if ChatGPT no longer just provided answers, but took you on a journey of understanding?

This article is about ChatGPT's new Study Mode – a learning feature that asks questions, provides background information, and guides you step by step to your own understanding instead of providing ready-made solutions. Instead of focusing on results, the learning process takes center stage.

The added value is clear: for anyone active in the AI community, this feature shows how AI can not only assist, but also promote genuine, self-directed learning. This opens up new potential, especially for educational projects, researchers, and those interested in EdTech. But what's even more important is that it's available in the “Free” tier, giving everyone access to a personal tutor who's there for them 24/7 to calmly and patiently answer all their questions.

Instead of simply answering questions, Study Mode encourages users to formulate their own learning goals and explore the topic from the ground up. This was achieved in collaboration with over 40 educational institutions – the AI essentially becomes a Socratic tutor. There are also some criticisms: young users could easily bypass the mode, so motivation remains a key factor.

Why it matters: Study Mode transforms ChatGPT from an answer machine into an active tutor – ideal for deeper learning. It shows how AI can rethink education.

Sources:

🔗 https://openai.com/index/chatgpt-study-mode

🔗 https://www.businessinsider.com/chatgpt-study-mode-openai-google-gemini-education-2025-7

🔗 https://www.axios.com/2025/07/29/openai-chatgpt-study-mode

🔗 https://www.theguardian.com/technology/2025/jul/29/chatgpt-openai-chatbot-study-mode-universities-students-education

🔗 https://help.openai.com/en/articles/11780217-chatgpt-study-mode-faq

TLDR covers the best tech, startup, and coding stories in a quick email that takes 5 minutes to read.

No politics, sports, or weather (we promise).

NVIDIA's new GB300 NVL72 server features an innovative power supply with integrated energy storage that actively smooths power fluctuations and reduces peak grid demand by up to 30%, marking a fundamental shift towards more resilient and sustainable AI infrastructure.

Imagen 4 Ultra now ranks #3 in the Artificial Analysis Image Arena, rivaling GPT-4o and Seedream 3.0 as one of the world's best image models!

NVIDIA's new open-source model Llama-3 Nemotron Super 49B v1.5 achieves top scores in the Artificial Analysis Intelligence Index, a benchmark for reasoning and agent-based tasks. The model was developed specifically for multi-step reasoning, scientific inference, and code generation. It combines high efficiency with low computational requirements and is freely available via Hugging Face. This makes the development of smart, autonomously acting AI applications much easier – a major step toward practical AI agents in everyday life, research, and industry.

The model lowers the barriers to entry for agentic AI, which could greatly accelerate the automation of complex thought processes and tasks. This will enable new applications in science, industry, and education – even without gigantic data centers.

The new SmallThinker approach (arXiv:2507.20984), presented on July 28, 2025, describes a family of language models designed from the ground up for local devices without GPU usage. With an innovative two-stage MoE design, a pre-attention router for preloading expert parameters, and sparse attention, the models use minimal memory. The variants with 4 billion and 21 billion parameters achieve >20 tokens/s on normal CPUs with only 1 GB and 8 GB RAM, respectively.

The paper “Agentic Reinforced Policy Optimization (ARPO)” describes a new reinforcement learning approach for training LLM-based AI agents that need to use external tools across multiple steps. ARPO recognizes uncertainty in tool calls (high token entropy) and controls adaptive branching in sampling behavior in a targeted manner, thereby significantly improving the alignment of agent behavior. ARPO outperforms classic RL algorithms on 13 challenging benchmarks – with only half as many tool uses.

Have groundbreaking AI research? We’re inviting researchers to submit their work to be featured in Superintelligence, the leading AI newsletter with 200k+ readers. If you’ve published a relevant paper on arXiv.org, email the link to [email protected] with the subject line “Research Submission”. If selected, we will contact you for a potential feature.

Stop wasting time on repetitive explanations. Guidde’s AI creates stunning video guides in seconds—11x faster.

How it works: Click capture on the browser extension, and Guidde auto-generates step-by-step video guides with visuals, voiceover, and a call to action.

Best part? It’s 100% free.

👉 Download the extension now