ChatGPT's image generation is free now:

If you remember last week's Newsdrop talking about ChatGPT's new "native" image generation, then you should be happy to know that it's now free for all users! The tool has been extremely popular ever since its launch last week, being used to generate anything from paintings to... Studio Ghibli recreations? Free users can't use it limitlessly, though - the free tier apparently only offers three image generations a day, though all users now have the chance to use this popular tool.

PaperBench:

In other OpenAI news, they just recently released Paperbench, a benchmark to measure how well AI agents can perform complex tasks, namely in replicating AI research. The AI agents have to accurately replicate 20 major papers from a machine learning conference, being tested on comprehending the research, coding, and conducting experiments. One example of an AI on this benchmark is Claude 3.5 Sonnet, with an average replication score of 21%. Compared to top human machine learning experts though, current AI models still have a ways to go.

MCP by Anthropic - OpenAI edition:

OpenAI becomes the newest AI giant to acknowledge and plan to integrate MCP into its products! MCP, or Model Context Protocol, is a way made by Anthropic to provide new tools to an AI's LLM. It helps models connect to outside data sources without needing to convert the sources to an integrate-able format. Think of it as like a USB-C of sorts, but between specialized databases and AI models!

News drop sponsored by OpenAI apparently, anyways

Runway:

Another huge step forward for image - or rather - video generation! Runway, a company focused on AI generative media, has released a new model called Gen 4 that is VERY GOOD at maintaining scene and character composition across multiple shots! If you've ever checked out an AI video but found that the people and scenes look uncannily inconsistent, well, Runway seems to be on its way to solving that problem. The video synthesis is only available to paying users for now, but it looks very promising in terms of video generation in the field.

Amazon:

Amazon has introduced the Nova Act, an AI model that supports advanced "agents" executing complex tasks within areas as small as web browsers. Agents have historically been handling smaller tasks like answering search queries, but Nova aims to redefine them as being capable of completing bigger, multi-step tasks in flexible environments.
Nova has also scored pretty well on usability benchmarks:

• Near perfect score on measuring language instructions for text-based interactions, like fonts.
• High score on interactions with visual icons, like rating stars.
• Low in terms of navigating user interfaces

We can see that Nova is a very ambitious and all-encompassing project so far, and curious as to where Amazon takes this in the future!

Google:

Isomorphic Labs, an establishment under Alphabet, has raised more than $600 million in its first ever external funding round. From who, you ask? Well, only from Thrive Capital (the firm that supported big AI minds like OpenAI), Google, and of course Alphabet itself. Isomorphic is technically a spinoff company from Google Deepmind, focusing on using AI to advance drug discovery. It and Deepmind has already developed AlphaFold, a technology that can impressively predict the structure of millions of proteins. We're excited to see what other plans Isomorphic has in store for us.

That's all for this week! Have a great rest of the week everyone!