ONE BIG READ

xAI’s newest AI comes with better reasoning, image generation, and less guardrails

X, the platform that was once Twitter, launched Grok-2 and Grok-2 mini, the latest AI models from xAI. These models bring enhanced reasoning abilities and, for the first time, allow users to generate images directly within X.

According to xAI, Grok-2 marks a significant upgrade from the previous Grok-1.5 model. Both Grok-2 and Grok-2 mini have been tested on a wide range of academic benchmarks, covering areas like reasoning, reading comprehension, math, science, and coding. 

The results have shown Grok-2 to be competitive with other advanced models, excelling in fields such as graduate-level science knowledge (GPQA), general knowledge (MMLU), and math competitions (MATH). Grok-2 also stands out in vision-based tasks, delivering top-tier performance in visual math reasoning (MathVista) and document-based question answering (DocVQA).

The new image generation feature is already sparking interest, but mainly for its lack of guardrails. Browse the site for a few minutes, and you’ll find any number of wacky generations, usually involving politicians or beloved animated characters. According to xAI, Grok-2 uses FLUX 1, an AI model built by German-based Black Forest Labs, a company that recently raised $31 million from the likes of AI6Z.

Despite the model’s advancements, many early testers noted that Grok-2 still has room for improvement, particularly in areas like code generation and news summaries, which had caused issues with the first version of Grok.

Developers will soon have access to Grok-2 and Grok-2 mini via an enterprise API, which is expected to roll out later this month. For those eager to try it out, Grok-2 is available in beta for Premium users on X.

Check it out

PRODUCT HIGHLIGHT

This AI tool handles your HR so you can focus on shipping

Embarking on a startup is probably one of the most exciting experiences in life. It’s not all fun and games, though. If you want to build a real, true-to-life startup, you’re going to have to think of some of the more tedious back-office tasks like company compliance, taxes, and payroll. These are things that, if not done correctly, could land you and your company in some very hot water. 

That’s why Central exists. After founder Nilay Modi had to pay over $20,000 in government fines, he set out on a mission to ensure this didn’t happen to other startups. Central is a YC-backed platform built to handle a lot of government-orientated startup work. 

Say you’re hiring a new developer. Rather than pulling your hair out from frustration, you can hand off the process to Central. From there, it will handle things like offer letters, payroll registration, compensation, payroll taxes, benefits, and more. It also handles more general things like city and state registrations and business insurance, including liability insurance.

The goal is to allow you and your team to focus on building without losing time working through legal jargon or, worse, potentially facing a fine that could very well set your company back. 

Check it out

AI IN THE HEADLINES

🖼️ Google has expanded access to Imagen 3 to all US-based users. 

🎨 Procreate vowed not to introduce generative AI features in its app.

🤖 Anthropic launched a new feature that makes prompting AI easier.

📱 Here’s every AI reveal announced at Made by Google 2024.