🍓 OpenAI Unveils o1, Apple Intelligence Is Coming, and Sora's Competitor by Adobe

PLUS HOT AI Tools & Tutorials

Sep 13, 2024

Hello and welcome to our weekly roundup!

What do you think of Apple's latest event? Honestly, if it weren't for Apple Intelligence (which we already know almost everything about from WWDC24), I'd call it the most boring one yet. It wasn't bad; it was just too sterile.

But of course, today, we're interested in a very different release: the much-anticipated “Strawberry” by OpenAI. Let's not delay and get straight to that topic.

This week, we've rounded up some of the most notable AI tools for creators and freelancers that you haven't heard of. Follow this link:
AI Tools for Creatives & Freelancers
Stepan Ikaev
·
Sep 12
Read full story

This Creators’ AI Edition:

Featured Materials 🎟️
News of the week 🌍
Useful tools ⚒️
Weekly Guides 📕
AI Meme of the Week 🤡
AI Tweet of the Week 🐦
(Bonus) Materials 🎁

Featured Material 🎟️

OpenAI Unveils o1 Model

OpenAI's “Strawberry” AI model is finally here.

After many rumors about the mysterious “Strawberry,” OpenAI has revealed a new family of models. It's called o1 and now includes two models: o1-preview and o1-mini. Their main draw is their penchant for reasoning. OpenAI claims that o1 has been trained using a perfect new optimization algorithm and a new training dataset developed for this model.

We are introducing OpenAI o1, a new large language model trained with reinforcement learning to perform complex reasoning. o1 thinks before it answers-it can produce a long internal chain of thought before responding to the user.

OpenAI has taught previous GPT models to mimic patterns from its training data. With o1, the company trained the model to solve problems on its own using a technique known as reinforcement learning, which trains the system with rewards and penalties. It then uses a “chain of thought” (chain of thought) to process queries, similar to how humans process problems.

The new approach has resulted in o1 models being noticeably better at writing code and solving multi-step problems than their predecessors. Here's an example to help you understand the difference a little better. On the qualifying exam for the International Mathematical Olympiad (IMO), a math competition for high school students, o1 solved 83% of the problems correctly, while GPT-4o solved only 13%.

The model primarily focuses on data analysis, science, and coding tasks. GitHub has already tested o1 with the help of its assistant GitHub Copilot and reported that the new AI does an excellent job of optimizing application code. At the same time, developers had to sacrifice speed: the model sometimes takes 30 seconds to formulate an answer.

The o1 also has several other serious limitations. Unlike GPT-4o, the new models cannot browse the web or analyze files. The model has image analysis features, which are disabled pending additional testing. In addition, the o1 is limited in frequency; weekly limits are currently 30 messages and 50 messages.

Keep your mailbox updated with key knowledge & news from the AI industry

A fairly significant drawback of o1 is its cost. In the API, o1-preview now costs $15 for 1 million input tokens and $60 for 1 million output tokens. This is three times more expensive than GPT-4o for inputs and four times more expensive for outputs.

To access o1 in the ChatGPT client, subscribe to Plus and Team. Enterprise and educational users will get access early next week. OpenAI also said that free ChatGPT users can use o1-mini in the future, but the exact release date is unknown.

First Impressions

We are already testing the o1 and are preparing a dedicated post for the new release. Here are some first impressions that we got:

o1 is needed for severe tasks - marketing research, scientific reports, etc. To use it instead of GPT-4o for simple tasks is a bad idea. It is too slow.
Prompt engineering is not needed. Moreover, it may worsen the results.
The model is far from perfect. It still makes mistakes like any other SOTA-level model.
The o1 works best when researching or solving a “high-level problem.” (We'll show a detailed example in the next post)
The 30-message limit is a serious problem. Working with o1, you have to count the messages yourself and plan the session with LLM in advance 😖 (there are no limits in the API, but so far, only users with a Level 5 have been given access)

More details and a full technical breakdown next week!

In general, the release of o1 looks like another (albeit small) step towards human-level AI reasoning. Yes, it's slow, expensive, and not perfect right now, but with the company calling the new product a “critical breakthrough,” we'd like to believe that progress will soon move much faster.

News Of The Week 🌍

Apple Intelligence Comes to iPhone, iPad, and Mac

Apple finally held its biggest event of the year. As usual, the company unveiled new iPhone, Apple Watch, and AirPods models (but that's not really important to us), but this time, it also talked a lot about AI. Repeating many of their WWDC reveals, developers praised the capabilities of the Apple Intelligence platform and announced that new features will be available to iPhone, iPad, and Mac owners as early as next month.

To test Apple Intelligence, users must update to iOS 18.1, iPadOS 18.1, and macOS Sequoia 15.1. The AI features will only be available in the US.

First innovations include writing tools, mail and notifications summaries, a more natural and flexible Siri, and the clean up tool in Photos.

Adobe Previews its Text-to-Video Generative AI Tools

Adobe has unveiled some of its upcoming generative AI video tools, including a new feature that can create video clips from still images. This latest preview is based on the in-development Firefly video model the software giant demonstrated in April. The platform will power AI video and audio editing features in Adobe Creative Cloud apps.

The tool allows users to create video clips using text descriptions and adjust the results using various “camera controls” that mimic camera angles, motion, and shooting distance. The company also showed image-to-video conversion for Firefly, which can create clips using specific reference images.

Adobe plans to release a new AI product later this year.

So we have another full-fledged competitor for Sora. It's funny how all the companies seem to have time to release their video generators before OpenAI does.

Mistral Releases Pixtral 12B Multimodel Model

Mistral AI Unveils Pixtral 12B, a Multimodal AI Model for Text and Image Processing

Mistral AI is finally entering the multimodal arena. Today, the French AI startup, which competes with OpenAI and Anthropic, released Pixtral 12B, its first multimodal model with built-in language processing and vision capabilities. While the model is not currently available on the public Internet, its source code can be downloaded from Hugging Face or GitHub for testing on individual instances.

According to initial testers, the 24GB model’s architecture appears to have 40 layers, 14,336 hidden dimension sizes, and 32 attention heads for extensive computational processing. On the vision front, it has a dedicated vision encoder with 1024×1024 image resolution support and 24 hidden layers for advanced image processing. This, however, can change when the company makes it available via API.

Mistral again broke the typical trend of releasing AI models by first posting a torrent link to download the new model's files. I love these guys.

Sam Altman & Bill Gates on Oprah’s AI Special

Thursday evening, Oprah Winfrey aired a special on AI, appropriately titled “AI and the Future of Us.” As part of the show, the host was joined by Sam Altman, Bill Gates, Marques Brownlee, and other guests. Oprah noted in prepared remarks that the AI genie has already been let out of the bottle, for better or worse, and that humanity will have to learn to live with the consequences. As TechCrunch notes, the dominant tone was skepticism and wariness.

The issue was very intense. I recommend watching it in its entirety or reading a brief retelling with highlights.

Sharing is caring! Refer someone who recently started a learning journey in AI. Make them more productive and earn rewards!

Refer a friend

Bloomberg: OpenAI Discussing Raising $6.5B in Equity Financing

Bloomberg reported that OpenAI is seeking to raise $6.5 billion at a $150 billion pre-money valuation. The new report comes two weeks after rumors surfaced that the company was in talks for a new multibillion-dollar funding round. At the time, CNBC's sources said that OpenAI was looking at a more than $100 billion valuation.

It is assumed that Microsoft will also participate in the round. The company has provided OpenAI with about $13 billion in capital and cloud loans to support its AI research. Nvidia and Apple, which recently partnered with OpenAI, may also participate.

News that OpenAI is about to raise another round is not surprising. However, it's important to capture important milestones for the AI industry. Plus, with new stages of growth, OpenAI is attracting more and more attention from regulators. And their reaction will determine how the market develops in the future.

Useful Tools ⚒️

Your eCom Agent - AI Tools for Amazon Sellers

Ouro – Collaborative monetization platform for technical creators

Genkin – Track your cash flows with a chat-based system

Ask Fellow Copilot – ChatGPT for your meetings

Scripe – Create viral LinkedIn posts with AI

At first glance, it may seem that Scripe is just another post generator for social networks. But this is only partially true. In addition to text creation, this platform allows you to translate your voice memos into the basis for LinkedIn posts, provides analytical tools, and helps with strategy planning. You can track your LinkedIn growth, capture the timing of your posts, and quickly optimize your content.