EU AI Act checker flags compliance issues in big tech’s AI models

Key Points 

• A new report shows that several popular AI models are not fully compliant with EU regulations, particularly regarding cybersecurity and bias issues.

• The European Commission endorses the LLM Checker as a first step toward enforcing AI regulations. 

📔News - A new report has found that some of the most widely used AI models are not fully compliant with key European regulations, particularly around cybersecurity and discriminatory output. The evaluation comes as the European Union's AI Act begins to roll out, aiming to regulate how AI technologies are used across various sectors. 

Swiss startup LatticeFlow, with backing from EU officials, developed a tool to test AI models from tech giants like Meta and OpenAI.

👨🏻‍💻What did they find? LatticeFlow's "Large Language Model (LLM) Checker" scored models from companies like Alibaba, OpenAI, and Meta on a scale from 0 to 1, with most models achieving scores of 0.75 or higher. However, the tool also exposed significant weaknesses. For instance, OpenAI’s GPT-3.5 Turbo received a low score of 0.46 for discriminatory output, highlighting ongoing issues of bias around race, gender, and other factors. Alibaba Cloud’s model performed even worse, scoring only 0.37.

The tool also evaluated cybersecurity threats, like "prompt hijacking," a cyberattack where malicious prompts trick the model into leaking sensitive information. Meta’s Llama 2 scored 0.42 in this category, while Mistral’s model scored 0.38.

👨🏻‍🏫Why this matters - The European Commission acknowledged the LLM Checker as an early step toward enforcing the new AI regulations. Although the Commission cannot officially verify external tools, they welcomed this initiative, saying it aligns with the technical goals of the AI Act. 

Developers can access the LLM Checker online to test their models for free. 

AI and cloud funding in U.S., Europe, and Israel set to reach $79 billion in 2024, Accel reports

Key Points 

• Funding for AI and cloud companies is projected to reach $79.2 billion in 2024, marking a 27% increase from 2023.

• Approximately 80% of the $56 billion in generative AI investments went to U.S. companies over the last two years.

• While AI funding is rising, the broader tech landscape is shifting focus from growth to profitability.

📈News - Funding for artificial intelligence and cloud companies in the U.S., Europe, and Israel is on the rise after three years of decline, with estimates projecting it to reach $79.2 billion by year-end, according to a report by venture capital firm Accel. This marks a significant 27% increase from the $62.5 billion recorded in 2023. 

Notably, investments in generative AI companies make up about 40% of the 2024 total. Philippe Botteri, a partner at Accel, emphasized the magnitude of this shift, stating, "The tectonic shift we are seeing with AI right now is bigger than anything that we have seen in the past, be it broadband or mobile or cloud."

🕵🏻‍♂️A closer look -  Of the combined $56 billion invested in generative AI for 2023 and 2024, approximately 80% has gone to U.S. companies, leaving only 20% for Europe and Israel. Two-thirds of this funding, totaling $37 billion, has been directed toward companies developing foundation models. 

Major funding rounds include OpenAI, which raised $6.6 billion earlier this month, and Elon Musk’s xAI, which secured $6 billion in May. Meanwhile, in Europe, Mistral, Aleph Alpha, and DeepL led funding efforts.

🫠Beyond AI - While the outlook for AI funding appears robust, the broader technology landscape is facing challenges. Accel noted that the era of high software growth is being replaced by a greater emphasis on profitability. This shift may alter investment strategies as companies navigate the changing market dynamics.