The week in AI: Anthropic + Amazon = the new AI power couple?

In partnership with

Welcome to The Dispatch! We are the newsletter that keeps you informed about AI. Each Thursday, we aggregate the major developments in artificial intelligence - we pass along the news, useful resources, tools and services; we highlight the top research in the field as well as exciting developments in open source. Even if you aren’t a machine learning engineer, we’ll keep you in touch with the most important developments in AI.

Happy Thanksgiving! Hopefully we caught you before the tryptophan kicks in…

NEWS & OPINION

Amazon doubles down on Anthropic with another $4b investment

-------------------------

Amazon just announced a new $4b investment in AI startup Anthropic, bringing its total investment to $8b and deepening a strategic partnership focused on cloud computing and AI development. With the deal, AWS becomes Anthropic's primary cloud and training partner, with Claude models optimized for Amazon's Trainium and Inferentia chips (Anthropic had previously been training on Google hardware). Anthropic is also collaborating with Amazon's Annapurna Labs to develop and optimize future-gen AI processors.

This second investment comes as little surprise, with Amazon already announcing that the forthcoming AI-infused version of Alexa will be powered by Anthropic’s frontier LLM, Claude. Additionally, as Amazon’s AI division is focused more on developer productivity than traditional chatbots, it makes sense to cozy up to team Anthropic - Claude 3.5 Sonnet is often anecdotally viewed as the most impressive coding model currently available.

And - perhaps most appealing to a tech giant like Amazon as their competitors face delays to major AI products - Anthropic are fast movers. In just the last week, Anthropic announced a number of updates/new features, primarily for Claude:

• Custom styles so Claude can automatically generate outputs in the writing style you prefer. In addition to presets and custom user-instructed styles, you can upload sample content that reflects your preferred way of communicating, specify instructions to match your needs, and edit your preferences.

• Global profile preferences for Claude to consider in every chat, along with your styles and project instructions. Profile preferences are a great way to instruct Claude on things you always want it to keep in mind, like your preferred coding language, or context about your work function.

• Google Doc integration to ground Claude in broader company knowledge.

• A new Model Context Protocol (MCP): an open source project lead by Anthropic that allows AI models to access and use business tools, content repositories and development environments, enabling more relevant and context-aware responses. Major companies like Block and Apollo have already integrated MCP, and developers can start building with prebuilt servers for platforms like Google Drive and Slack. Anthropic Head of Claude Relations Alex Albert posted a demo showcasing the MCP, with Sonnet 3.5 connecting to GitHub to create a repo and pull request.

Those might not seem like major updates or upgrades at first glance, true. But Anthropic’s proven ability to rapidly design, build and ship useful new features is something Amazon’s AI division is currently lacking, and why many see Anthropic as OpenAI’s biggest rival.

The new funding for Anthropic comes just as the DOJ is attempting to roll back Google’s own $2b investment as part of its sweeping federal antitrust case.

Nvidia’s new AI audio generator, Fugatto, can generate and edit music, speech, or sounds you’ve never heard before

-------------------------

A meowing trumpet? A saxophone howling like a dog? There are a lot of AI audio tools out there, but Nvidia has announced the first one that can “invent” entirely new sounds. Fugatto promises some impressive capabilities in creating sounds, music, and speech. The model responds well to highly imaginative prompts, enabling it to generate totally unique sound effects from text.

It can also manipulate voice attributes, altering accents, tones, or emotional expressions, and even edit existing music by isolating vocals, adding instruments, or replacing melodies. Fugatto has multi-accent and multilingual capabilities, and Nvidia highlighted potential use cases from advertising to video game development.

A paper released with the announcement shows the long list of all the datasets Nvidia says Fugatto was trained on, one of which includes a library of sound effects from the BBC. The researchers expanded the model’s task range with innovative instructions, allowing it to achieve high accuracy and tackle entirely new tasks without additional training data. This approach enables Fugatto to generate and edit audio that it was not specifically trained on.

The announcement comes amid legal and ethical scrutiny over AI-generated music and audio, with some startups facing copyright lawsuits. Nvidia has not yet disclosed when Fugatto will become widely accessible.

Zoom 2.0: “Allow us to reintroduce ourselves” 

-------------------------

The company that defined remote work during the pandemic is now laser focused on AI to redefine its future. Zoom has announced a rebrand from 'Zoom Video Communications’ to ‘Zoom Communications’, aiming to expand on the company’s video conferencing roots and position itself as an AI-first workplace platform. Now, it no longer wants to be known for just video as workers have returned to offices - and bigger, better-resourced competition from Google, Microsoft, and Slack offer video conferencing as a feature of the office suites most companies already pay for.

Zoom said its AI Companion will be the “heartbeat” of the push, with expanded context, web access, and the ability to take agentic actions across the platform. The rebrand follows recent launches, including the AI Companion 2.0, Zoom Docs, and other AI workplace tools aimed at competing with other tech giants. CEO Eric Yuan reiterated a vision to create fully customizable AI digital avatars/twins, which he believes could shorten work schedules for many to just four days a week.

There has been a fair share of skepticism about Zoom’s post-pandemic trajectory - they were in a precarious situation since most messaging platforms already had some form of video conferencing (and then were updated to mirror Zoom's interface, which is really all the company had that was unique). But the company continues to show steady growth - and an all-in bet on AI avatars might be the best one available under the circumstances.

MORE IN AI THIS WEEK

• OpenAI is considering launching its own browser to challenge Google

• Former Android leaders are building an ‘operating system for AI agents’

• Will the bubble burst for AI in 2025, or will it start to deliver?

• Trump eyes AI czar to partner with Elon Musk

• (Prof. Ethan Mollick) Getting started with AI: Good enough prompting

• Amazon, Google and Meta are ‘pillaging culture, data and creativity’ to train AI, Australian Senate inquiry finds

• Nvidia’s new AI audio generator Fugatto can generate and edit music, speech, or sounds you’ve never heard before

• Artists leak OpenAI’s Sora video generator in protest of unfair practices and “art washing”

• LinkedIn co-founder Reid Hoffman warns against Elon Musk’s ‘conflict of interest’ in setting AI policies

• Inflection AI CEO says it’s done trying to make next-generation AI models

• Perplexity and Quartr announce partnership that will bring Perplexity AI-powered live earnings call analysis, summaries, and qualitative financial research

• DeepSeek: The quiet giant leading China’s AI race

Writer RAG tool: build production-ready RAG apps in minutes

• Writer RAG Tool: build production-ready RAG apps in minutes with simple API calls.

• Knowledge Graph integration for intelligent data retrieval and AI-powered interactions.

• Streamlined full-stack platform eliminates complex setups for scalable, accurate AI workflows.

Learn more about our production ready RAG tooling here.

TECHNICAL NEWS, DEVELOPMENT, RESEARCH & OPEN SOURCE

• The Allen Institute for AI (Ai2) introduces OLMo 2: The best fully open language model to date

• Alibaba challenges OpenAI’s o1 with open-source reasoning model

• Andrew Ng releases aisuite: an open-source Python framework that lets you swap between LLMs with one line of code

• Unsloth AI, an open-source platform for fine-tuning LLMs, now supports vision and multimodal models

• Hugging Face unveils SmolVLM: a fast, open-source visual language model with optimized GPU RAM usage

• You can now test and compare model performance directly from the OpenAI dashboard - use your custom data to iterate prompts and refine outputs easily

• Frontier AI now runs at instant speed - Cerebras achieves record-breaking 969 tokens/s with Llama 3.1 405B

• OuteAI introduces OuteTTS-0.2-500M: an open text-to-speech model with multilingual support and fluid speech synthesis

That’s all for this week! We’ll see you next Thursday.