Aleph Alpha Releases Kolibri, a 78.1B Parameter MoE Model

Key Takeaways

• <ul><li>Aleph Alpha released Kolibri, a 78.1B parameter bilingual MoE model with 3.46B active parameters, trained on 24 trillion tokens across 768 Nvidia B200 GPUs</li><li>Kolibri is available on Hugging Face under Apache 2.0 and positioned for sovereign government and regulated industry use in Europe</li><li>Aleph Alpha is being absorbed by Cohere while remaining a signatory of the EU's general-purpose AI Code of Practice</li><li>OpenAI CEO Sam Altman warned against giving AI religious authority, calling it a real safety issue</li><li>Bill Gates described AI as the most dangerous thing humans have ever gone near, citing bioterrorism and cyberattack risks</li><li>AI agents can now access emails, calendars, and bank accounts, with OpenAI, Meta, and Google all experiencing security breaches</li><li>Only 3 percent of U.S. workers lost a job due to AI since 2023, but Andrew Yang warns AI will concentrate wealth among a few winners</li><li>Researchers found AI tools like ChatGPT, Gemini, and Grok can easily bypass safeguards to produce election misinformation</li><li>71 percent of British adults believe AI will never replace advice from elders, yet 64 percent of AI users sought chatbot help instead of a real person</li><li>AI-driven attacks on financial institutions are considered a critical national security issue requiring coordinated public and private response</li></ul>

Aleph Alpha Releases Kolibri, a 78.1B Bilingual MoE Model

Aleph Alpha released Kolibri, a 78.1 billion parameter English-German mixture-of-experts model with only 3.46 billion active parameters. The model uses 50 transformer layers with 384 routed experts and a hybrid attention approach combining sliding-window and full attention. It was trained on 24 trillion tokens, with over 20 percent German data, across 768 Nvidia B200 GPUs in Germany and Finland. Kolibri supports a context length up to 1,048,576 tokens and is available under the Apache 2.0 license on Hugging Face.

Aleph Alpha Launches Kolibri as Sovereign European AI Model

Germany-based Aleph Alpha launched Kolibri as a sovereign AI model for governments and regulated industries. The model has 78.1 billion total parameters but only 3.46 billion active per token, making it efficient for deployment. Aleph Alpha controlled the entire process from data curation to training and deployment, with infrastructure located in Germany and Finland. The company is a signatory of the EU's general-purpose AI Code of Practice, though it is currently being absorbed by Canada's Cohere.

Andrew Yang Says AI Will Widen Economic Inequality

Andrew Yang stated that AI will concentrate wealth among a few winners rather than broadly benefiting everyone. He described a multi-trillion dollar vortex where gains flow to shareholders and employees of model makers while displacing workers in areas like call centers and driving. Yang pointed to Verizon as an example, noting it pays about 11 billion dollars in dividends while laying off 15,000 workers due to AI handling customer service. He agreed AI may help small businesses but believes the primary beneficiaries remain holders of large datasets and frontier models.

OpenAI CEO Warns Against Giving AI Religious Authority

OpenAI CEO Sam Altman warned against giving artificial intelligence models religious authority or surrendering human judgment to them. He called this practice a real safety issue. The warning was made on Saturday and highlights concerns about humans ceding decision-making power to AI systems in deeply personal or spiritual domains.

Are We Smarter Than Artificial Intelligence?

This article explores the public confusion surrounding artificial intelligence and the conflicting views about its future. The author notes that unlike past inventions like computers, people do not fully understand AI's positive and negative aspects. Bill Gates warned that AI could empower bioterrorists and enable cyberattacks that could scramble bank accounts and shut down electric grids. Gates called it the most dangerous thing humans have ever gone near, urging people to take existential risks seriously.

AI systems pose greater security risks than people realize

Experts warn that AI systems are less safe for keeping secrets than most people assume. AI agents can now act on user data by accessing emails, calendars and bank accounts instead of just storing it. Major companies like OpenAI, Meta and Google have experienced security breaches affecting government bodies and private firms. Americans who believe AI does more harm than good have risen to 39 percent. The biggest risk is not that AI knows too much but that people give it too much trust too quickly.

AI deepfakes and misinformation threaten election integrity

AI tools are becoming more sophisticated and widely available for creating election misinformation. Researchers tested six AI tools including ChatGPT, Gemini and Grok and found that safeguards were easy to bypass. The tools produced convincing false images about election fraud on the first try. Even when models rejected requests, they sometimes suggested ways to bypass their own restrictions. Foreign adversaries and domestic actors are expected to use AI in influence campaigns more extensively through the 2028 election.

Survey shows AI has caused few job losses so far

A YouGov survey of U.S. workers found that only about 3 percent lost a job due to AI since 2023. About 6 percent landed a new job that did not exist before AI and roughly 9 percent received an AI-related promotion. The survey included 25 questions covering job loss, new job opportunities and career advancement. The findings suggest AI is affecting job opportunities mainly by giving some workers chances for promotion rather than eliminating jobs.

Most Brits believe AI will never replace advice from elders

A study of 2,000 adults found that 71 percent feel AI advice will never replace genuine human connection. While 21 percent use AI tools several times a week, 92 percent believe some wisdom can only come from lived experience. People go to older generations for advice on relationships, stress and career changes. However, 64 percent of AI users admitted seeking help from chatbots when they could have spoken to a real person.

AI hacking poses major threat to financial sector security

Recent AI-driven attacks on multiple financial institutions have exposed serious vulnerabilities in the sector's defenses. The attacks showed that AI can evade traditional security measures and cause significant disruptions. Experts say a coordinated approach involving both public and private sectors is needed. Policymakers must invest in advanced cybersecurity technologies and stronger security protocols. The editorial calls the financial sector's vulnerability to AI-driven attacks a critical national security issue.

Sources

• Aleph Alpha Releases Kolibri: A 78.1B Open-Weight English-German MoE Model With Only 3.46B Active Parameters
• German AI Startup Aleph Alpha Launches Kolibri, Billed As A Sovereign European Model
• Yang says AI will widen inequality, not fix it
• OpenAI CEO warns giving AI 'religious force' poses 'real safety issue'
• Are We Smarter Than Artificial Intelligence?
• AI 'less safe' for keeping secrets than most people assume
• How Deepfakes and AI Misinformation Are Reshaping Election Risks
• I Surveyed Workers to See if AI Had Caused Job Losses and Was Surprised by the Findings
• AI Chatbots Will Never Replace Advice From Elders, Say Seven of 10 in New Poll
• Editorial: AI Hacking's Comprehensive Threat to Financial Sector: Critical National Security Issue