The week in AI: (Re?)Introducing ChatGPT Search

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

ChatGPT search, history search, and other OpenAI updates

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

It was a busy week in the news for ChatGPT-maker OpenAI. Here’s the scoop:

• ChatGPT search is here! Or, err, back. ChatGPT will now choose automatically to search the web, based on what you ask - or you can manually choose to search by clicking the web search icon. It’s available now for Plus and Enterprise users, and in the coming months will be available for free users as well.

• You can also now search your chat history on ChatGPT web to find a buried conversation or pick up a previous one where it left off.

• Advanced Voice Mode is now available for macOS and Windows ChatGPT desktop apps

• The company disbanded yet another safety team after another high profile resignation. Team lead Miles Brundage cited increasing constraints on his research and publication freedom at OpenAI as major reasons for his resignation. The AGI Readiness team previously advised the company on handling AI that could match/exceed human intellect.

• Despite OpenAI’s warnings to not use the software in high-risk domains, hospitals have rushed to use OpenAI’s transcription tool, Whisper. Unsurprisingly, troubling research has now emerged about hallucination rates.

• CFO Sarah Friar said that 75% of OpenAI’s revenue comes from ChatGPT subscriptions. The company is expected to lose about $5 billion this year.

Additionally, Sam Altman shot down rumors that the company’s next flagship model, codenamed Orion, would be arriving in December, around the two-year anniversary of ChatGPT.

The first Apple Intelligence AI features are now generally available

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

The first batch of Apple Intelligence features are now out of developer beta and available on select Apple devices. The earliest offerings include writing tools, image cleanup/editing, and an enhanced Siri with better context understanding. While more substantial features will arrive before the end of the year (ChatGPT+Siri integration, Image Playground, Genmoji), this initial Apple Intelligence offering feels like a definite soft launch into the AI era. Earlier this month, it was reported that some Apple employees believe the company is at least two years behind in AI development. Some of the early AI summary results are not encouraging.

Apple Intelligence is currently available only on iPhone 15 Pro models, iPads with A17 Pro or M1 chips, and Macs with M1 or later. You can enable Apple Intelligence by going to Settings > Apple Intelligence & Siri and toggling it on. If you want to learn more about Apple Intelligence, your local Apple store most likely started putting on drop-in lessons yesterday.

Apple also refreshed its Mac lineup with new M4 chips, focusing on AI performance. The M4 Pro and M4 Max, built on second-gen 3nm tech, feature improved Neural Engines that double the speed of previous models. All new Macs, including Mac Mini, MacBook Air and MacBook Pro, now come with 16GB of RAM by default to meet the demands of on-device AI processing. Despite the slow roll of AI into its software ecosystem, Apple’s latest hardware lineup is already fully optimized for on-device AI processing.

Mysterious new state-of-the-art AI image generator revealed

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

Earlier this week, an unknown text-to-image generator codenamed “red_panda” made a dramatic entrance at the top of Artificial Analysis’ text-to-image model leaderboard, with an ELO rating well ahead of industry leaders like Midjourney, FLUX, and Ideogram.

Shortly after declaring that they were the red_panda model in question, startup Recraft announced Recraft V3 on their blog. While image generators are getting better all the time, this was a pretty stunning entry into the top of a loaded field. The image quality overall is among the best, and Recraft has some additional distinctive/noteworthy features:

• Advanced control options and long-text capability: Recraft lets users control text size and precise placement (and you can put in a lot more text than most image generators), giving designers more flexibility to integrate/edit text seamlessly.

• Realistic anatomy and prompt precision: Recraft has very realistic anatomy and body proportions. Outputs also align closely with complex prompts, down to exact colors and object positions.

• Vector graphics support: Unlike most image generators, Recraft supports vector output, from simple pictograms to intricate vector art.

• Style consistency and brand alignment: Users can define specific brand styles without retraining the model. Recraft accepts style inputs, making it possible to maintain consistent brand aesthetics across multiple outputs.

• End-to-end AI editing suite: Recraft’s suite includes advanced tools like AI Eraser, Modify Area, Inpainting, Outpainting, Clarity and Creative Upscalers, Background Remover, even an “AI Mockuper”.

• API access: The new API allows integration of Recraft’s features into business workflows, including style consistency and vector art generation.

MORE IN AI THIS WEEK

• President Biden issues first-ever National Security Memorandum on AI

• Meta is pushing for the government to use its AI

• Alphabet Q3 results beat analyst estimates as cloud revenue soars on AI demand

• Google DeepMind is building an AI to keep us from hating each other

• How a Mumbai drugmaker is helping Putin get Nvidia AI chips

• Why building in AI is nothing like making conventional software

• AI models fall for the same scams that we do

• UK man who used AI to create child abuse images jailed for 18 years

• Google CEO says over 25% of new Google code is generated by AI

• OpenAI disbands another safety team, as head advisor for ‘AGI Readiness’ resigns

• Meta signs deal with Reuters to provide real-time news responses through its AI chatbot across Facebook, Instagram, WhatsApp, etc.

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

• GitHub Universe AI news: Copilot to go multi-model and come to Apple’s Xcode, and Spark - an AI-powered tool for creating micro apps with natural language

• Introducing the next-level of AI-powered workflows with Amazon Q Developer inline chat

• LLMs reflect the ideology of their creators

• OmniParser: an open-source vision tool from Microsoft that converts UI screenshots to structured formats

• Integuru: an AI agent that generates integration code by reverse-engineering platforms' internal APIs

• NotebookLlama: Meta’s open source version of Google’s NotebookLM

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