Welcome to Get The Gist, where every weekday we share an easy-to-read summary of the latest and greatest developments in AI—news, innovations, and trends—all delivered in under 5 minutes! ⏱

In today’s edition:

• Grok Blocks Results Labeling Musk as Misinformation Spreader

• Apple Moves Closer to Integrating Google Gemini Into Siri

• Claude 3.5 Sonnet Outperforms GPT-4o and o1 in Coding

• Gemini Expands Document Upload Feature to All Free Users

• And more AI news….

Top Developments

Grok

Grok Blocks Results Labeling Musk as Misinformation Spreader

The Gist: xAI’s chatbot, Grok, was briefly modified to block results stating that Elon Musk and Donald Trump spread misinformation. The change was made by an employee without approval and has since been reversed.

Key Details:

• xAI’s head of engineering, Igor Babuschkin, blamed an ex-OpenAI employee for the unauthorized change.

• Grok’s system prompt is publicly visible, allowing users to see how it is programmed to respond.

• The chatbot previously claimed Musk, Trump, and VP JD Vance were harming America, leading engineers to intervene.

• Musk promotes Grok as a “maximally truth-seeking” AI designed to understand the universe.

Apple

Apple Moves Closer to Integrating Google Gemini Into Siri

The Gist: Apple is reportedly exploring the integration of Google’s Gemini AI into Apple Intelligence, potentially offering an alternative to OpenAI’s ChatGPT in Siri and Writing Tools.

Key Details:

• iOS 18.4 beta code references both OpenAI and Google as third-party AI options.

• Apple has already integrated Google’s visual search into the iPhone 16’s Visual Intelligence tool.

• Gemini could enhance Siri’s ability to generate, summarize, and edit text.

• Apple has previously used Google’s cloud-based TPUs to train its own AI models.

Anthropic

Claude 3.5 Sonnet Outperforms GPT-4o and o1 in Coding

The Gist: A new OpenAI study found that Claude 3.5 Sonnet outperformed GPT-4o and o1 in software engineering tasks based on real-world freelance coding jobs. Despite its lead, AI models still struggle with complex debugging and problem-solving.

Key Details:

• The study used SWE-Lancer, a benchmark based on 1,488 freelance coding tasks worth $1M.

• Claude 3.5 Sonnet earned $400K in simulated payouts, ahead of GPT-4o ($304K) and o1 ($380K).

• AI models were better at reviewing proposals than solving coding problems from scratch.

• Performance improved when models were allowed to generate multiple answers and refine their responses.

Quick Gist

Google has revealed that its Veo 2 video-generating AI model will be priced at 50 cents per second, totaling $1,800 per hour. Read More

xAI's co-founder is blaming a former OpenAI employee after Grok briefly censored responses about Elon Musk and Donald Trump. Read More

Amazon's investment in Anthropic has surged in value from $8 billion to $14 billion, marking a 75% gain since 2023 as the AI startup's valuation reaches $60 billion. Read More

Google Gemini has expanded its document upload feature to all free users, allowing them to analyze various file types and receive AI-generated summaries. Read More

Google has started allowing free Gemini users to generate images of people using its Imagen 3 model, a feature previously exclusive to Gemini Advanced subscribers. Read More

THAT’S IT FOR TODAY!