November 2023 AI news roundup: What’s new in the world of AI?

November was a big month for AI: The release of Amazon Q and the new version of ChatGPT, OpenAI’s first DevDay, U.S. presidential announcements, a new version of Claude, and of course, the leadership tussle with Sam Altman. It was honestly a squeeze to get it all in this article! Here are the biggest headlines.

• AWS released Amazon Q, the first chatbot built specifically for companies
• AWS also upgrades Bedrock, offers new AI processors
• Anthropic upsized Claude 2.1 to 200k tokens, nearly doubling GPT-4
• Microsoft offers new Copilots, including Microsoft 365 Copilot AI & Studio
• Microsoft rebrands Bing Chat to Copilot
• ChatGPT can work with images and voice
• OpenAI released custom GPT building feature for ChatGPT
• OpenAI makes GPT-4 Turbo model available
• OpenAI’s new Assistants API was released
• Google DeepMind AI predicts structure of thousands of new materials
• Biden issued an exec order on AI, setting tone for future U.S. AI policy
• OpenAI loses their CEO, gets them back, and celebrates a birthday

This was hands-down the most exciting announcement at this year’s re:Invent. Amazon Q is the first AI chatbot made with business in mind from the ground up. You can plug it into data sources like Amazon S3, Microsoft 365, GitHub, Dropbox, Salesforce, Slack, and more, so you don’t have to upload company documents for it to understand your business — it already knows. You can set it to use the rule of least privilege and existing user permissions.

For a full rundown of Amazon Q, including how to use it, we’ve written a comprehensive article: “What is Amazon Q? The AI chatbot for companies.”

AWS re:Invent 2023 was full of AI announcements, including new expansions for Amazon Bedrock such as Guardrails for Amazon Bedrock, Agents for Amazon Bedrock, and enhanced Amazon Bedrock customization features. To power high-performance AI and ML training, AWS announced the next generation of two AWS-designed chip families: Graviton4 and Trainium2.

Claude, Anthropic’s flagship model, now has a 200k token window and a 2x decrease in hallucination rates. Don’t know why that matters? Let’s put it this way: a token is a basic unit of text an AI like Claude uses to process or create language, and a token is roughly 0.75 words. Harry Potter and the Sorcerer’s stone is 77k words, so it would be about 102k tokens — that means Claude could easily read or write something that long.

In comparison, OpenAI’s latest GPT-4 Turbo model only has a context window of 128k tokens — still long enough to write a HP novel, but not enough to produce Tolkien’s The Fellowship of the Ring at 187k (That said, we’re only talking about the quantity of output, not quality.) All of this is good news for Google and Amazon, who have both heavily invested into Anthropic and Claude in a bid to prop up a competitor to OpenAI, who are joined at the hip with Microsoft. 

Don’t know much about Claude? Check out our explainer article: “What is Claude AI, and how does it compare to ChatGPT?”

Microsoft 365 Copilot AI assistant was launched for certain enterprise customers, offering them an LLM AI assistant that changes the way Office documents are created and edited. They also launched Microsoft Copilot studio, a low-code tool to customize Microsoft Copilot or build standalone copilots.

If you’ve jumped onto Bing this year, you might have noticed the ChatGPT-like interface integrated into the search results. Despite being partnered with OpenAI, Microsoft has rebranded Bing to Copilot better position itself as the first option for consumers and businesses, and align with all the Copilot-branded products its offering.

As of November, you can now upload images and have ChatGPT interpret them, or have it create images for you. That means you can get help identifying things, get instructions, or use pictures to solve problems. If you’re using the mobile appl, you can also interact using your voice for input.

This was just one of the many announcements at OpenAI’s first ever developer conference, DevDay, in San Francisco. Expect DevDay to be a part of the tech circuit in 2024 onward.

You can now create tailored versions of ChatGPT to better suit your day-to-day needs. you’re one of those people who has a long list of useful prompts that you just copy-and-paste over and over again into ChatGPT (or even if you aren’t), then GPTs are for you. Not only can you create your very own, but you can also share them with others.

There was also meant to be a GPT Store opening where you could earn revenue off the GPTs you build, but the store launch has been pushed back to early 2024. If you're looking to make cash off this, I'd take the time to practice and test your GPTs in this window, and get in with your ideas early.

To learn more about GPTs and how to make them, check out our guide: “How to create custom GPTs in ChatGPT.”

While completely overshadowed now by Anthropic’s Claude announcement, OpenAI released GPT-4 Turbo, which has a 128k token content window and is more affordable than GPT-4. Turbo has a 3x lower price for input tokens and 2x lower price for output tokens compared to GPT-4, and has knowledge of world events up until April 2023.

OpenAI also plans to roll out new features for GPT-4 Turbo, including function calling improvements and a new JSON mode for developers. This means developers can now build applications that maintain continuity over lengthy interactions, ideal for detailed customer service, long educational dialogues, or in-depth content creation.

Capabilities around function calling have improved, giving you the ability to call multiple functions in a single message.  For example, if you want to get the weather for Miami, Madrid and Marrakech, you can do that with just a single trip to the model.

Google DeepMind used AI to predict the structure of two million new materials, of which 400k could soon be produced in lab conditions. These could lead to advancements in better performing batteries, solar panels, and computer chips.

This is a big deal, because discovering and synthesizing new materials takes a lot of money and time. For example, it took two decades until lithium-ion batteries became commercially available, which we use for everything from phones to laptops.

“We're hoping that big improvements in experimentation, autonomous synthesis, and machine learning models will significantly shorten that 10 to 20-year timeline to something that's much more manageable,” said Ekin Dogus Cubuk, a research scientist at DeepMind.

The Assistants API was also announced, a new feature that allows developers to build more complex, agent-like AI applications. This API equips developers with the tools to build AI assistants that not only understand complex commands but can also leverage an array of models and tools to carry out tasks.

To learn more about using Assistants API, check out our writeup here: “OpenAI DevDay: Diving into ChatGPT's new features, APIs.”

President Joe Biden issued an executive order on AI, setting the tone for how the technology will be governed in America moving forward. The order seeks to address nearly every AI issue that’s cropped up in the last twelve months. It includes testing mandates for advanced AI models to make sure they’re not weaponized, mandatory watermarking of AI-generated content, and minimizing the risk of workers losing their jobs to AI.

The order is incredibly long, so we unpacked it for you in this article here: “Biden's big AI executive order says a lot about the future of tech.”

It’s highly unlikely you missed this one, but we’d be remiss not to include it in this month’s updates. OpenAI CEO Sam Altman was fired by the board in a shocking ouster, then re-hired after most of the company’s staff threatened to quit. Now the old board has been replaced (mostly), and everything seems to be more or less back to normal. The new board is putting in rules to prevent the same thing happening again.

Also, ChatGPT turned one. Can you believe it’s been a year since the AI chatbot completely changed the world?

November was perhaps the biggest month for AI in the year, with AWS scrambling to challenge Azure’s apparent AI dominance, Microsoft trying to both work with and position themselves ahead of OpenAI, and the U.S. Government trying to keep pace. It will be interesting to see if this red-hot pace continues unabated, or if things will lull over the holiday break.