Good morning. It’s Friday, January 10th.

Did you know: On this day in 2007, Apple unveiled the original iPhone?

In today’s email:

• Grok “Unhinged” and iOS App

• How To Write Better Code with AI

• China’s New AR Glasses

• 3 New AI Tools

• Latest AI Research Papers

You read. We listen. Let us know what you think by replying to this email.

Grok may soon get an 'Unhinged Mode'

Elon Musk’s xAI is on the brink of launching an “Unhinged Mode” for its Grok chatbot, a feature designed to push boundaries with responses deemed “objectionable, inappropriate, and offensive.” This mode, outlined in a recent FAQ update, reflects Musk’s intention to create an AI that strays from conventional filters, like a chatbot moonlighting as an edgy stand-up comedian.

While Grok already veers into unfiltered language when prompted, this mode aims to take things a step further. Despite current attempts to curb its left-leaning biases, the move signals Musk’s ambition to shape an AI capable of answering controversial queries without the usual constraints. Read more.

Related Story: 𝕏 launches Grok’s iPhone app in the US

Repeated "write better code" prompts can make AI-generated code 100x faster

Max Woolf’s experiment with Claude 3.5 demonstrated that iterative prompts to “write better code” resulted in an astonishing 100x speed improvement. Starting with a Python task, the code's runtime was reduced from 657 milliseconds to a mere 6 milliseconds. However, the most striking development was the code’s unforeseen evolution; by the fourth iteration, Claude autonomously incorporated enterprise-level features, suggesting an internal association between “better code” and corporate software architecture.

While precise prompt engineering accelerated performance, it also introduced minor errors, underscoring the importance of meticulous guidance. Woolf’s findings highlight AI’s potential to generate innovative solutions, but also stress that human intervention is indispensable for refinement and error correction. Read more.

China’s XReal One Blazes Past Apple’s Vision Pro with Cinematic AR Glasses

XReal’s XReal One AR glasses challenge established players like Apple’s Vision Pro by offering unprecedented performance and a cinematic AR experience. Powered by the proprietary X1 chip, these optical see-through glasses deliver 120Hz refresh rates and an industry-leading 3ms motion-to-photon latency, outpacing competitors like the Vision Pro, which has a 12ms latency.

The XReal One has a 57-degree field of view, a significant improvement over previous models, and a lightweight magnesium alloy frame for comfort. Paired with any device featuring USB-C video-out, it supports a range of content platforms. The glasses’ audio experience, tuned in collaboration with Bose, further enhances immersion. With a detachable 12MP RGB camera and plans for AI capabilities, the XReal One promises a versatile AR solution at a competitive price point, starting at $499. Read more.

• Microsoft rolls back its Bing Image Creator model after users complain of degraded quality

• Microsoft makes powerful Phi-4 model fully open-source on Hugging Face

• Omi is another AI companion wearable — but this one’s trying to read your mind

• Apple's inaccurate AI news alerts shows the tech has a growing misinformation problem

• Salesforce Will Hire No More Software Engineers in 2025, Says Marc Benioff

• Companies prioritize AI training over job cuts, WEF study finds 

• Microsoft's rStar-Math Takes on OpenAI's o1-preview in math reasoning

• Midjourney's v7 Overhaul: Speed, Quality, and a Leap into 3D

• Sundar Pichai teases new Google AI products, features in 'next few months'

• Adobe’s TransPixar Brings Transparency to AI-Generated Effects

• Diffbot Replaces Guesswork with Verified Data in AI

• ChatGPT's newest feature lets users assign it traits like 'chatty' and 'Gen Z'

• Alibaba’s Qwen Chat Streamlines AI Interaction with Powerful New Web UI

• Meta knew it used pirated books to train AI, authors say

• NVIDIA's Jensen Huang says that IT will 'become the HR of AI agents'

• OpenAI blocks developer who linked ChatGPT to weapons system 

• VLC player demos real-time AI subtitling for videos

• The coolest gadgets at CES 2025

📄 Agent Laboratory: Using LLM Agents as Research Assistants

📄 REINFORCE++: A Simple and Efficient Approach for Aligning Large Language Models ​​

📄 Towards System 2 Reasoning in LLMs: Learning How to Think With Meta Chain-of-Though

📄 Efficient Generative Modeling with Residual Vector Quantization-Based Tokens

📄 Fourier Position Embedding: Enhancing Attention's Periodic Extension for Length Generalization

Thank you for reading today’s edition.

Your feedback is valuable. Respond to this email and tell us how you think we could add more value to this newsletter.

Interested in reaching smart readers like you? To become an AI Breakfast sponsor, reply to this email or DM us on X!