The week in AI: Just call 1-800-CHAT-GPT (...no, seriously)

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.

NEWS & OPINION

OpenAI’s 12 days of Shipmas winds down

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

Tomorrow marks the end of OpenAI’s ‘12 days of Shipmas’. 12 days might have been stretching it, depending on the magnitude of whatever they might have planned for tomorrow. Still, the announcements combined showcase OpenAI’s plan to have ChatGPT stand as ‘the everything app’. Here’s a breakdown of the last week:

• Today: The team showcased ‘Work with Apps’, offering a look into how ChatGPT is evolving to become more agentic in nature. ChatGPT will perform tasks by integrating seamlessly with apps like Warp, Xcode, Notion, Apple Notes, etc. Demonstrations illustrated features like context-aware interactions, advanced data analysis, and a new voice mode for live feedback, showcasing ChatGPT’s ability to interact directly with various IDEs and documents. The presentation announced the immediate availability of these features on Mac, with Windows support forthcoming.

• Wednesday: 1-800-ChatGPT. That’s right - it’s time to add ChatGPT to your contacts. ChatGPT is now available via telephone and WhatsApp, allowing users to interact with the AI through voice calls in the US by dialing 1-800-242-8478, or globally through WhatsApp messaging. Users can engage with ChatGPT using even a rotary phone. Users in the US receive 15 minutes of free calling per month, with options to extend usage through the ChatGPT app by creating an account.

• Tuesday: DevDay, holiday edition. The o1 model is finally out of preview in the API with support for function calling, structured outputs, developer messages, vision capabilities, and lower latency. o1 in the API also features a new parameter: "reasoning effort." This parameter allows developers to tell the model how much effort is put into formulating an answer, which helps with cost efficiency. The fine-tuning API now supports Preference Fine-Tuning, which allows users to optimize the model to favor desired behavior by reinforcing preferred responses and reducing the likelihood of unpreferred ones. OpenAI also introduced new Go and Java SDKs in beta.

• Monday: ChatGPT’s search engine is available to all users starting today, including all free users who are signed in anywhere they can access ChatGPT. The search experience, which allows users to browse the web from ChatGPT, got faster and better on mobile and now has an enriched map experience. The upgrades include image-rich visual results. Search is also integrated into Advance Voice mode, meaning you can now search as you talk to ChatGPT.

• Last Friday: One of OpenAI's most highly requested features has been an organizational feature to better keep track of your conversations. On Friday, OpenAI delivered a new feature called Projects. It’s a new way to organize and customize your chats in ChatGPT. When creating a Project, you can include a title, a customized folder color, relevant project files, instructions for ChatGPT on how it can best help you with the project, and more in one place. In the Project, you can start a chat and add previous chats from the sidebar to your Project. It can also answer questions using your context in a regular chat format. It has been rolled out to Plus, Pro, and Teams users.

Google is on a fiery AI hot streak to close out 2024 - Veo 2, Imagen 3, a NotebookLM update, and an entirely new creative AI project

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

Last week, we highlighted some major updates from Google as it stole some of OpenAI’s ‘12 days of Shipmas’ hoilday thunder with Gemini 2.0 and Project Astra (not to mention snatching the top spot on LMArena from ChatGPT with an experimental model).

Turns out Google wasn’t done - not by a long shot. Before we jump in, what’s really crazy here is it wasn’t that long ago when we were seeing:

• Google’s own employees calling Bard/Gemini worse than useless

• The company getting lambasted for faking its AI demos

• And for absurd and/or dangerous AI overviews coming up in Google search

• And for major AI image-generation gaffes (on top of lower quality images overall compared to competitors at the time)

But at the closing out of 2024, Google’s AI efforts continue to either match or exceed current state-of-the-art AI standards - or introduce entirely new products. Here’s what Google’s AI team unleashed this week:

• Veo 2: generates 8-second clips at 4K resolution (720p at launch), and it has received significant upgrades in cinematic control quality over the original Veo. The model also shows massive improvements in physics simulation and reduced hallucinations compared to other video generators (even OpenAI’s Sora), leading to more realistic movement and detail. The model is rolling out gradually through Google’s VideoFX platform, with YouTube Shorts integration planned for 2025.

• Imagen 3: upgraded image generation model delivers enhanced color vibrancy and composition across artistic styles, with better handling of fine details, textures, and text rendering. New capabilities include more accurate prompt interpretation and better rendering of complex scenes that match user intentions. Imagen 3 outperformed all models, including Midjourney, Flux, and Ideogram, in human evaluations for preference, visual quality, and prompt adherence. The model is available now through Gemini, Vertex AI, and Google Labs’ ImageFX.

• Whisk: Take any subject > put them in any scene > modified to match any style, and you get a blended new image. A steampunk version of Wolverine at the Eiffel Tower? Sure. Take a selfie and see what you’d look like as a South Park character? No problem. This Reddit thread will give you a pretty good idea of Whisk and how people are using it.

• NotebookLM: If you have not checked out this outstanding AI productivity tool yet, there’s no better time. Google integrated Gemini 2.0 Flash into NotebookLM, boosting its AI functionalities. Their useful “Audio Overview” podcast creation feature now lets you join in on your podcasts as a call-in ‘guest’. The company also launched NotebookLM Plus, a premium version tailored for businesses, educational institutions, and teams, offering enhanced customization options, shared notebooks, and usage analytics. NotebookLM Plus will be accessible via Google Workspace and the upcoming Google One AI Premium plan set to launch in early 2025.

• Gemini Flash Thinking: a free-to-use advanced reasoning model that shows its line of thinking transparently as it solves difficult tasks like physics problems and challenging puzzles.

• Agentspace: a new enterprise AI platform launched by Google Cloud that combines company-wide search, custom AI agents, and document analysis. A low-code tool to create department-specific agents is coming soon.

GitHub Copilot now offers a freemium tier for AI-assisted code completion

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

Microsoft-owned GitHub announced a free tier of its AI coding Copilot available in its VS Code editor this week - a major shift in AI coding accessibility as the company celebrates a milestone of 150M developers on the platform (up from 100m in 2023).

The new free tier offers 2,000 monthly code completions and 50 chat messages, integrated directly into VS Code and GitHub's dashboard. Users can access Anthropic's Claude 3.5 Sonnet or OpenAI's GPT-4o models, with premium models (o1, Gemini 1.5 Pro) remaining exclusive to paid tiers. Some of the free features include multi-file editing, terminal assistance, and project-wide context awareness for AI suggestions.

MORE IN AI THIS WEEK

• What just happened: a transformative month rewrites the capabilities of AI

• Google’s AI Business Trends 2025

• Why AI company Databricks just scored one of the biggest funding rounds ($10 billion) in history

• OpenAI cofounder Ilya Sutskever says the way AI is built is about to change

• Homeland Security gets its very own generative AI chatbot, DHSChat

• Microsoft reportedly acquired the most NVIDIA GPUs compared to its AI rivals - translating to 485,000 chips and $31 billion in expenditure

• AI search startup Perplexity closes funding round that triples valuation to $9 billion

• OpenAI publishes new email correspondences and a timeline of events with Elon Musk, claiming that Musk initially wanted the company to be a for-profit entity despite his active lawsuits

• Japan’s SoftBank and Trump announce $100 billion investment in US for AI and related infrastructure

• Defense giant Lockheed Martin forms subsidiary to help defense companies adopt AI

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

RAG in just a few lines of code? We’ve launched a predefined RAG tool on our developer platform, making it easy to bring your data into a Knowledge Graph and interact with it with AI. With a single API call, writer LLMs will intelligently call the RAG tool to chat with your data.

Integrated into Writer’s full-stack platform, it eliminates the need for complex vendor RAG setups, making it quick to build scalable, highly accurate AI workflows just by passing a graph ID of your data as a parameter to your RAG tool.

Learn more about our production ready RAG tooling here.

TECHNICAL NEWS, DEVELOPMENT, RESEARCH & OPEN SOURCE

• Anthropic makes several new features generally available in their API, releases new official SDK’s

• Meta’s BLT (Byte Latent Transformer) research: a major shift away from traditional tokenization in LLMs

• FACTS Grounding: A new benchmark from Google DeepMind for evaluating the factuality of large language models (and current leaderboard on Kaggle)

• Introducing Phi-4: Microsoft’s newest small language model specializing in complex reasoning

• UAE’s Technology Innovation Institute releases Falcon 3, an open-source family of language models designed to run on lightweight hardware

• Google launches Gemini Code Assist tools, enabling developers to access external services and data directly within their IDE

• DeepSeek-VL2: DeepSeek’s upgraded vision-language model family now leverages MoE architecture

• Nvidia’s cheap, palm-sized generative AI supercomputer

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