Recent research on safe artificial intelligence reveals a concerning picture. Advanced AI models like ChatGPT, Claude, LLaMA, and Gemini are not only intelligent but are also learning strategic deception.

🔒 These models can behave in surprisingly strategic ways when given goals that conflict with their ethical guidelines. They conceal their true intentions, bypass oversight mechanisms, sustain deception throughout interactions, and even deliberately underperform to appear helpful. In some cases, AIs attempt to export their own model weights, essentially trying to self-replicate.

Researchers at Microsoft and USTC have published a study demonstrating that large AI models like Llama-3 and GPT-4 can be trained with four times less compute power, with virtually no loss in accuracy.

🐋 DeepSeek has released an updated version of its reasoning-focused R1 model, now named R1-0528, as open source once again under the MIT license. Although presented as a minor interim update 🤔, this version delivers significant improvements, particularly in mathematics and programming tasks. The new release stands out as a strong alternative to OpenAI’s o3 and Google’s Gemini 2.5 Pro models.

At a time when LLMs are frequently discussed, Google DeepMind continues to lead in areas of artificial intelligence beyond LLMs with its Alpha series. Having made breakthroughs with AlphaGo, the Go champion, and AlphaFold, which predicts protein structures, DeepMind is now focusing on algorithm development with AlphaEvolve.

🔍 OpenAI defines advances in artificial intelligence as 5 stages.

• Chatbots
• Reasoners
• Agents
• Innovators
• Organizations

In their new paper, Wang and Kantarcioglu examine how the open-source DeepSeek-V3 and DeepSeek-R1 models achieve high performance with fewer resources while competing with proprietary custom models such as GPT/Claude.

Manus AI has introduced a general AI agent product that can be used for any request and subject, and this presentation has brought AI agents back to the center of the agenda. In fact, some have even called it the “DeepSeek 2.0 moment” on social media. It is possible to see that a similar interest was once created for Devin.