Hi Synapticians!

NO, AI models aren’t born biased; they simply inherit the biases lurking in our data, an imperfect mirror of society. And society can sometimes, with all its probabilistic quirks 😉, generate every kind of bias (sexist, racist, age-related, you name it).

Before we can fix those biases, we have to spot them. Good news: the Luxembourg Institute of Science and Technology (LIST; proud 😊) has just updated its bias leaderboard for major open models (OpenAI, DeepSeek, Mistral, etc.). For each model you’ll find the bias category, a detailed score, and most importantly a crystal-clear visualization to guide your choices.

And correction? That’s where the philosophical debates kick in. My PhD supervisor used to say that what we now call “AI” was once simply “discriminant analysis” ; its job: sift data to describe or predict better. Deciding which biases to offset has become more a matter of ethics than of tech (step aside, geeks), and the LIST scoring is the best fuel for that conversation.

Congrats to LIST for this initiative!

Top AI news

1. LIST Leaderboard Names the Most Toxic Chatbots
The Luxembourg Institute of Science & Technology (LIST) just unveiled an interactive LLM Leaderboard, stress-testing popular chatbots—from ChatGPT and Bard to open-source newcomers—against hundreds of prompts that probe seven social-bias categories. Powered by the open-source LangBiTe framework, each model earns a score that reflects how often it dodges ageism, sexism, racism, xenophobia, religious discrimination, political bias and LGBTIQ+-phobia. A simple color-code lets you spot toxic blind spots at a glance: deep red means trouble, vibrant green means fairer outputs. The leaderboard is refreshed with every model update, providing a real-time ethics barometer for developers, regulators and curious users alike. Full downloadable reports are available—once you drop your contact details—so anyone can audit the data behind the rankings.

2. OpenAI's AI Models Outperform Doctors in HealthBench
OpenAI has introduced HealthBench, a new benchmark for evaluating AI in healthcare, developed with insights from 262 doctors. The benchmark tests AI models' performance in realistic medical conversations across multiple specialties and languages. The latest models, GPT-4.1 and o3, outperform doctors in these scenarios, although the comparison doesn't fully reflect clinical practices. Despite notable progress, especially in efficiency and worst-case performance, challenges remain. OpenAI's initiative highlights AI's growing role in healthcare, encouraging further research and development.

3. Nvidia Ships 18,000 AI Chips to Saudi Arabia for Data Center
Nvidia is set to deliver 18,000 of its advanced AI chips to Saudi Arabia in collaboration with the AI startup Humain, which is backed by the country's sovereign wealth fund. This initiative aims to support a 500 megawatt data center project, enhancing Saudi Arabia's AI and cloud computing capabilities. The announcement was made during a Saudi-U.S. investment forum, highlighting the strategic importance of foreign investment in advancing national technological infrastructure.

Bonus. xAI Misses AI Safety Framework Deadline Amid Criticism
Elon Musk's xAI has failed to release its promised AI safety framework by the self-imposed deadline. This delay, noted by The Midas Project, adds to concerns about xAI's risk management, especially given its track record with the Grok chatbot. The draft framework, presented at the AI Seoul Summit, lacked details on risk mitigation strategies. This issue reflects a broader industry trend, with other AI labs also criticized for deprioritizing safety amid growing AI capabilities. The need for robust safety protocols is increasingly crucial as AI technologies become more advanced and integrated into daily life.

Meme of the Day

XXXXX

Theme of the Week

Small language model - Fierce Wireless “Tiny AI”

Fierce Wireless frames “tiny AI” as the sleeper trend behind next-gen mobile services and IoT. Quotes from IBM and Qualcomm execs reveal why bandwidth-savvy telcos are betting on edge models. Market analysts peg the ROI at 5× thanks to lower cloud egress and subscription churn. A lively, business-focused narrative you can share with non-tech readers.

Stay Connected

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