Holiday News Round-Up 📝

Your weekly dose of AI clarity, minus the complexity

Author

Jamie Oarton
December 28, 2024

Your weekly does of AI clarity, minus the complexity

Welcome, AI Freedom Finders

This week’s AB is a pithy holiday summary - we’ll be back to the usual deep dives in 2025.

Wishing you all a very Happy New Year and every success in 2025!

In today’s Brief:

  • Google uses Claude to evaluate Gemini's AI performance 🤖

  • Cybersecurity threats escalate through data breaches 🕵️

  • OpenAI's o3 model hints at potential AGI breakthrough 🧠

Source: OpenAI | The AI Brief

The Brief: Google is systematically comparing its Gemini AI against Anthropic's Claude model, potentially breaching commercial terms while exploring competitive AI development strategies.

Details:

  • Google contractors are comparing Gemini's AI outputs against Anthropic's Claude model, using detailed evaluation criteria for responses.

  • Internal correspondence revealed contractors noting Claude's stricter safety settings compared to Gemini, particularly in handling sensitive prompts.

  • Anthropic's commercial terms forbid using Claude to build or train competing AI models without explicit approval from the company.

  • Google DeepMind denies training Gemini on Anthropic models, stating they only compare model outputs as part of standard industry evaluation.

Why It Matters:

The AI rivalry intensifies with Google leveraging Anthropic's Claude to benchmark Gemini, potentially breaching commercial agreements and highlighting the cutthroat tactics emerging in the race to develop superior artificial intelligence.

Source: OpenAI | The AI Brief

The Brief: Major cybersecurity stories in 2024 exposed widespread data vulnerabilities, including massive hacks at AT&T, Cencora, and emerging AI-driven identity fraud risks across multiple industries.

Details:

  • Snowflake suffered a massive hack, resulting in AT&T losing call and text records for nearly 110 million customers, with a $370,000 ransom payment.

  • Automakers are sharing drivers' behaviour data with third parties, leading to potential rate increases and prompting a congressional inquiry into data sales.

  • Bloomberg revealed drug distributor Cencora paid a record $75 million ransom to prevent the release of sensitive personal and medical data for 18 million people.

  • 404 Media exposed an underground site using AI to generate fake IDs, highlighting the vulnerability of Know Your Customer (KYC) verification processes.

Why It Matters:

Cybersecurity in 2024 reveals a stark reality: no data is truly safe. From massive corporate breaches to AI-generated fake identities, our digital lives are increasingly vulnerable, demanding urgent, comprehensive protection strategies.

Know someone who might like this newsletter? Press the forward button and share. If they sign up on the signup page you will earn credits to special VIP content.

Source: dev.to

The Brief: OpenAI introduces o3, an advanced reasoning model with remarkable benchmark performance, showcasing potential AGI abitities through adaptive computational reasoning and self-fact-checking capabilities.

AGI - Artificial General Intelligence - is essentially when an AI model can recognise patterns in novel situations and adapt knowledge to unfamiliar challenges. This is something that, despite their impressive abilities, no AI is fully able to do as yet. 

Details:

  • OpenAI unveiled o3, a new reasoning model family with o3 and o3-mini, claiming potential approaches to AGI capabilities.

  • The model can adjust its reasoning time and performs exceptionally well on benchmarks like ARC-AGI, SWE-Bench, and mathematics exams.

  • o3 uses a "deliberative alignment" technique and employs a "private chain of thought" to fact-check itself before responding to prompts.

  • Safety researchers can preview o3-mini now, with OpenAI planning to launch the model publicly towards the end of January.

Why It Matters:

OpenAI continues pushing AI reasoning boundaries with o3, suggesting potential AGI capabilities that spark both excitement and caution. The model's self-fact-checking and adaptable compute approach hints at more nuanced, intelligent AI systems emerging.

When you’re ready, here’s how I can help you: