Welcome back to Daily Zaps, your regularly-scheduled dose of AI news ⚡️ 

Here’s what we got for ya today:

• 📦 Google Jarvis AI leaked

• 🎖️ Defense Llama

• 🗳️ AI polling company defends wrong predictions

• 👩🏻‍💻 25% of Google’s new code generated by AI

Let’s get right into it!

Sign Up | Advertise | Tools

BIG TECH

Google Jarvis AI leaked

Google briefly revealed its upcoming "Jarvis AI" tool on the Chrome extension store, but quickly removed it. Jarvis, scheduled for release in December 2024, aims to automate everyday web tasks like research, shopping, and travel bookings, powered by an advanced Gemini AI model. Though some users downloaded the extension, they couldn't use it due to restricted permissions.

Jarvis reflects a growing trend among tech companies to develop AI agents that handle web-based tasks autonomously. Similar tools include Anthropic’s Claude AI, Apple’s “onscreen awareness,” and Microsoft’s Copilot+ Recall, though the latter faced backlash for privacy concerns around screenshot storage and was delayed for further refinement.

STARTUPS

Defense Llama

Scale AI and Meta have launched Defense Llama, the first large language model (LLM) specifically designed for American national security. Built on Meta's Llama architecture and Scale AI's infrastructure, Defense Llama integrates defense-specific data, allowing it to perform strategic intelligence, threat detection, and secure communications with enhanced precision.

This domain-specific model outperforms general-purpose LLMs in handling security-related tasks, aiding defense agencies in faster, more accurate decision-making and bolstering autonomous defense systems. Defense Llama marks a significant advancement in using specialized AI for national defense, demonstrating the potential of LLMs to move beyond general applications and address complex sector-specific needs.

POLITICS

AI polling company defends wrong predictions

An AI polling startup, Aaru, tested its 2024 U.S. presidential election model, predicting a narrow win for Kamala Harris over Donald Trump, though the actual outcome differed. Founded by young entrepreneurs, Aaru creates AI characters representing different demographics, feeds them news, and asks how they'd vote.

Despite inaccuracies, co-founder Cameron Fink defended AI polling as faster, cheaper, and potentially more accurate than traditional polling, noting that their predictions were within the margin of error. Aaru's approach reflects a shift toward AI-driven polling, with the potential for refinement over time as traditional polling faces declining accuracy.

BIG TECH

25% of Google’s new code generated by AI

Google is now using AI to generate over 25% of its new code, which is then reviewed by engineers to enhance productivity and speed up development, as shared by CEO Sundar Pichai. This shift could impact entry-level engineering roles as AI takes on more of the coding workload. However, concerns remain about potential copyright issues and security risks if AI models are trained on licensed or outdated code.

Beyond coding, Google is expanding its AI efforts across products like Search, YouTube, and developing new models, including a potential Gemini 3.0 release. Future models may have autonomous capabilities to complete tasks independently, raising security and oversight concerns. Google’s increasing reliance on AI has also led to higher energy demands, pushing the company to explore nuclear energy as a more sustainable solution.