Hi Synapticians!

Hope you had a great weekend!

On the AI front, two big topics today: Grok 3’s DeeperSearch mode, and Tencent’s new reasoning model, based on a Mamba architecture (not the number 5 ;)) — an architecture competing with the famous Transformers — built on the Stated Space Model mechanism ; a alternative of the attention mechanism.

By the way, did you know that this mechanism was invented by the world’s top AI researcher? Uh… no, actually it was an intern 🤣

This week’s main theme is especially close to our hearts… we’re going to talk about AI in the service of mental health — a vast topic 🙂

Top AI news

1. Grok 3 introduces DeeperSearch and AI image editing
Grok 3 introduces DeeperSearch, a slower but more selective research tool, and a new AI image editing feature. DeeperSearch prioritizes reputable sources like The Guardian and BBC but may miss key information, while DeepSearch is faster but includes questionable sources. The image editing tool allows users to modify images via text input, similar to Google’s Gemini models. xAI has yet to provide full details on its capabilities. These updates highlight the challenge of balancing speed, accuracy, and usability in AI tools.

2. Tencent’s Hunyuan-T1 rivals OpenAI in AI reasoning
Tencent has introduced Hunyuan-T1, an AI model that competes with OpenAI’s reasoning systems. It excels in logic and math, scoring 87.2 on MMLU-PRO and 96.2 on MATH-500. The model leverages reinforcement learning and the Transformer Mamba architecture for faster processing. However, benchmark results don’t always translate to real-world performance. Tencent offers Hunyuan-T1 via its cloud and Hugging Face, signaling growing competition in the AI space.

3. Anthropic’s 'Think Tool' enhances Claude’s problem-solving abilities
Anthropic has introduced a 'Think Tool' for its AI assistant Claude, allowing it to take notes and structure its reasoning in real time. This improves performance in complex, multi-step tasks, with a 54% boost in customer service simulations. The tool helps Claude analyze data, follow rules, and make step-by-step decisions, enhancing reliability in AI-driven systems.

Bonus. AI-powered app helps speech-impaired users communicate naturally
Researchers at Northeastern University are developing Speak Ease, an AI-powered app designed to help speech-impaired users communicate more naturally. The app integrates speech recognition, text input, whole-word selection, emojis, and personalized text-to-speech synthesis. Unlike traditional tools, Speak Ease allows users to express emotions and moods, making conversations feel more authentic. It leverages large language models to predict phrases and improve real-time communication. The app is particularly useful in medical settings and for individuals with degenerative conditions, helping them preserve their voice. This innovation represents a significant step forward in AI-assisted communication.

Meme of the Day

Theme of the Week

AI for mental health - The concept

The idea started gaining momentum around the 2010s, but it’s really in the past five to ten years that AI for mental health has stepped into the spotlight. There’s no single inventor or breakthrough moment. Instead, it’s the result of cross-pollination between psychologists, AI researchers, and startups like Woebot Health, Wysa, or Ginger that reimagined what mental health support could look like in a digital world.

Stay Connected

Feel free to contact us with any feedback or suggestions—we’d love to hear from you !