Microsoft brings AI to the news...

PLUS: Roblox AI breaks language barriers

Welcome, AI enthusiasts.

In today’s surprising twist of events, Microsoft is teaming up with news outlets to bring AI-assisted stories, amid a lawsuit with the NYT.

The new partnerships are set to integrate AI into the news research process — though journalists will still write the actual articles. Let’s get into it.

In today’s AI rundown:

  • Microsoft partners with news orgs for AI research

  • Roblox bridges language barriers with AI chat translator

  • How to create coloring book images with AI

  • Apple boosts AI speech recognition with new method

  • 8 new AI tools & 4 new AI jobs

  • More AI & tech news

Read time: 4 minutes

LATEST DEVELOPMENTS

MICROSOFT

Image source: Midjourney

The Rundown: Microsoft just announced a series of new partnerships with media organizations to provide AI tools and guidance on using generative models for news gathering and operations.

The details:

  • Semafor’s 'Signals’ feed will be sponsored by Microsoft and leverage AI to research and curate insights, with the stories still written entirely by journalists.

  • Semafor’s writers will also use AI to sift through info across global news sources and various languages to enhance research capabilities.

  • Other Microsoft partnerships include CUNY’s journalism school, The Online News Association, The GroundTruth Project, and Nota.

  • The announcement comes on the heels of the NYT’s lawsuit against OpenAI, alleging the company trained ChatGPT on its news articles.

Why it matters: AI still produces occasional hallucinations in text (AKA it makes things up) — which is why we, ironically as an AI newsletter, still write every newsletter by hand. Media organizations are making the right move here by leveraging AI to streamline work across everything other than the writing.

TOGETHER WITH OCTOAI

The Rundown: Open source LLMs such as Mixtral and Meta’s Code Llama 70B are rapidly catching up to GPT-4. With OctoAI, you can swap the latest open-source models into your app in just three lines of code.

OctoAI lets you:

  • Test Mixtral-8-7B or Code Llama 70B in minutes

  • Find a faster and cheaper LLM

  • Stay ahead as open-source progresses

Don’t get locked into closed AI if there’s a better option.

ROBLOX

Image source: Roblox

The Rundown: Roblox just unveiled a real-time AI chat translator that enables players speaking different languages to seamlessly communicate in-game, supporting 16 languages and achieving near-instantaneous translation.

The details:

  • The system translates text between any combination of supported languages, bridging communication barriers between its diverse 70M daily active users.

  • Roblox trained an LLM on in-game terminology and public and internal datasets, helping it recognize internet slang and other platform-specific language.

  • The company also plans to expand translations across the platform, including text on images, 3D models, and voice chat.

Why it matters: We’ve posted often about AI’s ability to break down language barriers, and Roblox just provided a perfect real-world implementation. Pretty soon, every platform will likely have an AI system of its own — ending communication barriers as we know them.

AI TRAINING

Image source: Chase Lean (@chaseleantj on X)

The Rundown: With a good prompt, you can generate a coloring book version of any images for free using Microsoft Image Creator.

Thank you to Chase Lean for sharing this creative idea and prompt with us.

Step-by-step:

  1. Head to Microsoft Image Creator or Midjourney.

  2. Describe the image you want to generate using the following prompt:

    • “On the left, a black and white simple children's coloring book illustration featuring a [insert description of image] with no shading, with a white background. On the right, the same illustration with color. --ar 2:1 --style raw --sref [see image below] --stylize 250 --v 6

  3. After a few minutes you will receive a set of images you can choose from.

PRESENTED BY SECTION SCHOOL

The Rundown: The “AI class” is forming – don’t be left behind with the luddites. In 10 days, Section’s AI Crash Course will teach you to:

  • Build effective prompts, so you don’t get garbage back from ChatGPT

  • Use AI to take rote tasks off your to-do list

  • Harness AI’s three personas: assistant, strategist, and creator

Plus dedicated office hours with an AI expert and a library of AI resources. Enroll here.

AI RESEARCH

Image source: Apple

The Rundown: Apple researchers just introduced a new technique called "acoustic model fusion" that integrates an additional model into speech recognition systems, helping better recognize unfamiliar words and names.

The details:

  • End-to-end models (which complete speech-to-text processes in a single step) can often struggle with rare words not seen much during training.

  • The researchers found that fusing a secondary acoustic model to the end-to-end model during recognition provided useful additional pronunciation and audio data.

  • For some test sets, the approach cut word error rates by over 14%, with higher gains for less common words.

Why it matters: Apple providing new research on enhancing speech recognition is notable given Siri’s… comically lackluster capabilities in the area. But given the paper’s focus (and other rumored Apple AI developments), everyone’s least favorite voice assistant may get a serious upgrade soon.

NEW TOOLS & JOBS

  • 🎥 Typeframes- Create videos for YouTube, Instagram, and TikTok with simple text prompts (link)

  • 🤖 AI Form Roast- Grade your online forms with AI (link)

  • 👩🏼‍🦰 User Persona Generator GPT- Simulate an ideal customer without extensive interviews (link)

  • 📝 Flipner- Create masterful content faster than ever with this AI assist (link)

  • 📸 Product Descriptions GPT- Craft engaging e-commerce product listings (link)

  • 👥 Hearth- An AI agent that manages your contacts and relationships (link)

  • ✈️ Trip Planner GPT- Plan your trips effortlessly with a custom itinerary and expert advice (link)

  • 🛠️ Cohere- Engineering Program Manager, Generative ML (link)

  • 🧠 Robust Intelligence- Head of Operations (link)

  •  OpenAI- User Operations Generalist (link)

  • 🔍 Anthropic- Research Engineer, Model Evaluations (link)

QUICK HITS

Bumble launched a new AI-powered tool called Deception Detector, which automatically identifies and blocks 95% of spam and fake profiles on its dating platform.

Sam Altman posted that GPT-4’s performance should now be improved, saying the model had ‘a slow start on its new year’s resolutions’ to be less lazy.

Bayreuth University scientists are using AI to analyze galaxy structures and behaviors, a new method that efficiently tests astrophysical theories using Einstein’s principles of relativity.

Alibaba researchers released Qwen1.5, open-sourcing six model sizes with performance surpassing GPT 3.5 and Mistral-Medium, and competitive with GPT-4 on certain benchmarks.

A Hong Kong company lost $25.6M after scammers created a meeting with deepfake audio and video for multiple recreated coworkers, convincing a finance employee to transfer funds.

Chinese tech giant Huawei has reportedly had to slow its phone production despite leading the country in smartphone sales, prioritizing AI chips amid surging global demand.

THAT’S A WRAP

SPONSOR US

Get your product in front of over 450k+ AI enthusiasts

Our newsletter is read by thousands of tech professionals, investors, engineers, managers, and business owners around the world. Get in touch today.

FEEDBACK

How would you rate today's newsletter?

Vote below to help us improve the newsletter for you.

Login or Subscribe to participate in polls.

If you have specific feedback or anything interesting you’d like to share, please let us know by replying to this email.

Reply

or to participate.