Introduction
This past week has been nothing short of a whirlwind for Google. From groundbreaking AI announcements to public apologies and bizarre rumors, it's been a rollercoaster ride. Let's dive into the key events that defined Google's week in the world of tech.
Gemini 1.5: A Quantum Leap in Language Models
Google started the week strong with the unveiling of Gemini 1.5, a large language model (LLM) that is already causing serious ripples. The model boasts superior performance to GPT-4 on most benchmarks and features an incredible 10 million token context window. This surpasses competitors like Claude and GPT Turbo significantly. The author states they had early access and was able to upload an entire codebase from a side project, and have Gemini build features on top of it. Gemini performed better than Copilot or any other tool they've used.
The massive context window allows for more effective understanding of custom data, potentially diminishing the need for complex Retrieval Augmented Generation (RAG) stacks. Imagine uploading an entire codebase and having the AI understand its structure and dependencies, enabling it to build new features with ease. The author used Fireship Pro course videos and Gemini automatically extracted code and wrote tutorials about the videos. According to the author, Gemini 1.5 makes GPT-4 look like an antique from 2023.
Gemma: Google's Open-Source AI Offering
In addition to Gemini 1.5, Google introduced Gemma, a family of open-source models designed to compete with Meta's Llama 7b and Mistral. Google's benchmarks (taken with a grain of salt, as the author mentioned) indicate that Gemma models excel, particularly in math and coding. These models are free to use and can be incorporated into commercial applications. However, users must adhere to a "prohibited use policy," which restricts the types of applications they can be used for.
Image Generation Controversy and the Anti-Racism Paradox
The week took a turn when users discovered some problematic behavior with Gemini's image generator. The model's attempt to be inclusive and anti-racist seemingly backfired, generating images that were perceived as historically inaccurate and insensitive. Multi-racial Nazis and founding fathers stirred significant controversy. Google issued an apology and temporarily suspended Gemini's ability to generate images of people. This highlights the complex challenges of creating AI systems that are both inclusive and historically accurate.
The Groundbreaking Sign-In Page Update
Amidst the AI advancements and controversies, Google also rolled out a highly anticipated update to its sign-in page. The change involved shifting from a vertical to a horizontal layout. The author jokingly suggests this may have involved hundreds of product managers making $500k a year, and multiple vice presidents making over a million a year, all just to have an intern modify some HTML.
Gmail Apocalypse Averted
The week culminated with a prank email that sent shockwaves across the internet. The email, purportedly from the Gmail team, announced the sunsetting of Gmail in August 2024. The news spread rapidly, causing widespread panic among Gmail's 1.5 billion users. Fortunately, Google quickly clarified that the email was a hoax and that Gmail is not shutting down.
Conclusion
Google's week was a mixture of groundbreaking AI innovation, public relations challenges, and internet-wide panic. Gemini 1.5 represents a significant advancement in language models, while Gemma expands access to AI technology. The image generation controversy underscores the difficulty of creating unbiased and inclusive AI systems. And finally, the Gmail hoax serves as a reminder of the power of misinformation in the digital age. It’s certainly been a wild ride, hinting at the rapid pace of change as we move "towards the singularity".
0 Comments