OpenAI -> Closed AI? Claude3 Breaks Records, Chatbots Ignore Copyright, and MORE!
PLUS HOT AI Tools & Tutorials
👋 Hey, I’m Stepan and welcome to a ✨ news edition ✨ of Creators’ AI. By subscribing, you directly support Creators' AI's mission to deliver top AI insights & practical knowledge without ads or clutter. Your subscription allows us to grow our dedicated team and curate the most important AI Tools, Stories, and Tutorials in one place. - Stepan
Welcome to our weekly newsletter!
This week was rich in a variety of events. Anthropic introduced the Claude 3 family of models, OpenAI responded to Elon Musk's lawsuit, and the most popular chatbots underwent serious copyright testing. And that, of course, is not all!
So, without further ado, let's get started!
Oh, wait a minute! Do you want to know how creators are making $32K/mo on AI newsletters? Then this is the post for you:
🔗 Are you already subscribed and want to ensure you don’t miss our newsletters? Move our newsletter from “Promotions” to “Primary” in your inbox! (video instructions below)
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 🎟️
Anthropic has unveiled the next generation of Claude
A startup founded by former OpenAI developers has announced a significant update. The company has released the Claude 3 model family, which promises to become the new industry benchmark across various cognitive tasks. The family includes three state models: Claude 3 Haiku, Claude 3 Sonnet, and Claude 3 Opus.
All Claude 3 models demonstrate advanced analytics and predictive capabilities, detailed content creation, code generation, and communication in non-English languages, including Spanish, Japanese, and French.
According to the company, Opus (Anthropic's most intelligent model) outperforms the most common criteria for evaluating AI systems, including undergraduate-level expertise (MMLU), graduate-level expert thinking (GPQA), basic math (GSM8K), and more. The developers stated that It exhibits near-human levels of comprehension.
Here are the test results cited by Anthropic itself:
Haiku, meanwhile, is the fastest and most cost-effective. It can read a research paper with much information and data on arXiv (~10k tokens) with charts and graphs in less than three seconds. Anthropic will increase their performance in the near future.
Sonnet provides twice the speed of Claude 2 and Claude 2.1, with a higher level of data processing. It handles quick response tasks, such as knowledge discovery or sales automation.
Essential aspects of the update:
Fewer refusals: Previous Claude models often allowed unnecessary refusals, suggesting a lack of contextual understanding. Now, the company has made progress in this area: Opus, Sonnet, and Haiku are significantly less likely to refuse to respond to requests outside the system than previous generations.
Improved accuracy: Chatbots now recognize that users may not be confident in their answers. The Claude 3 will soon get a citation feature to point to exact sentences in reference materials to verify their answers.
Long context and near-perfect recall: The Claude 3 family of models will initially offer a 200,000-sized context window at launch. At the same time, all three models can accept inputs larger than 1 million tokens. The latter feature will be available for a limited number of clients.
And much more. Traditionally, chatbot developers have noted that they care about privacy and copyright and that Claude 3 is well-suited for applications. All three models are available in two price categories.
Prices:
Claude 3 Haiku - $0.25 | $1.25
Claude 3 Sonnet - $5 | $15
Claude 3 Opus - $15 | $75.
Changes to Chatbot Arena
The release of Claude 3 (and the introduction of the new Mistral AI models we wrote about last week) has changed the balance of power among chatbots.
Chatbot Arena is a platform where models are blindly matched up against real people. A user goes to the site, writes a query, sees two answers and chooses the best one. The more often one model's results are chosen, the higher it ranks. Chatbot Arena now has tens of thousands of votes.
For a long time, GPT-4-Turbo was the absolute leader of this rating, but now the situation is changing: Claude 3 Opus has almost caught up with GPT-4-Turbo, and Claude 3 Sonet is overtaking the May release of GPT-4 as well as Mistral Large.
Want to know which chatbot to use every day? We'll help you with your choice:
News Of The Week 🌍
OpenAI strikes back: a response to Musk (And Elon offered a new name)
We continue to follow the major confrontation in the AI industry. A few days after Elon Musk's lawsuit, OpenAI released a response statement. The company said it categorically disagrees with the Tesla founder's accusations. According to OpenAI, Musk had only ever contributed $45M to OpenAI and wanted to merge with Tesla.
In a blog post authored by the entire OpenAI band - Greg Brockman, Ilya Sutskever, John Schulman, Sam Altman, and Wojciech Zaremba - the Microsoft-backed startup revealed that since its inception in 2015, it had raised less than $45 million from Musk, despite his initial commitment to provide as much as $1 billion in funding.
In a separate memo obtained by Bloomberg, Altman called Musk a hero and said he misses the person he knew who competed with others by building better technology. Meanwhile, Musk suggested that OpenAI should be renamed ClosedAI because its work doesn't meet its stated goals.
All in all, it sounds good, and the domain is free for now. Although considering how Elon "killed" our favorite bird, I wouldn't rely on him to rebrand the company.
The DeepMind alums have revealed AI for video generation called Haiper
Two DeepMind alums, Yishu Miao and Ziyu Wang, have publicly released their video-generation tool Haiper with its own AI model underneath. Users can go to Haiper's website and create free videos by entering text prompts. Right now, you can only generate two-second HD and slightly lower-quality videos of up to four seconds.
Haiper has raised $13.8 million in a seed round led by Octopus Ventures with participation from 5Y Capital. Before that, angels like Phil Blunsom and Nando de Freitas helped the company raise a $5.4 million pre-seed round in April 2022.
Researchers tested leading AI models for copyright infringement. GPT-4 is the worst
Patronus AI, an AI model evaluation company founded by ex-Meta researchers, released research showcasing how often leading AI models produce copyrighted content. The company tested the most popular chatbots: GPT-4, Claude 2, Llama 2, and Mixtral, prompting them to generate text from popular books protected by copyright laws in the U.S.
OpenAI’s GPT-4 produced the highest amount of copyrighted content, responding to an average of 44% of prompts asking for text from books with the copyrighted text. That said, the other models did not perform well either. All chatbots violated copyright during the tests.
Klarna CEO says AI can do the job of 700 workers
The Swedish fintech said its AI assistant is already doing the equivalent of 700 full-time agents and handling two-thirds of Klarna's customer service chats in a month. AI has led to fewer errors, a 25% reduction in repeat requests, and an average chat time from 11 minutes to 2 minutes.
Klarna relies on a partnership with OpenAI to integrate AI into its tasks. The ChatGPT-based bot took over most of the company's work in just one month, including 2.3 million conversations across 23 markets and 35 languages.
Ema, a ‘Universal AI employee,’ emerges from stealth and raises $25M
A new startup called Ema from San Francisco has come out of hiding with a product that it believes will open a new chapter in how generative AI will change how we work. Ema's goal is to create a Universal AI employee. The platform will automate routine tasks and free up time for people to do more valuable work.
Ema's products — Generative Workflow Engine (GWE) and EmaFusion — are designed to "emulate human responses." They are suitable for customer service, including providing technical support to users tracking and other functions, as well as internal productivity applications for employees.
Political deepfakes are spreading like wildfire thanks to GenAI
In a study from the Center for Countering Digital Hate (CCDH), a British nonprofit dedicated to fighting hate speech and extremism online, the co-authors find that the volume of AI-generated disinformation — specifically deepfakes about elections — has been rising by an average of 130% per month on X over the past year.
Callum Hood, head of research at the CCDH, said the results indicate that the availability of free, easily jailbroken AI tools — along with inadequate social media moderation — is contributing to a deepfakes crisis. According to the co-authors, most of the deepfakes on X were created using one of four AI image generators: Midjourney, DALL-E 3, DreamStudio, or Microsoft’s Image Creator.
So be careful, it's getting harder and harder to trust information on the internet! (Or rather, you just shouldn't do it anymore)
Share this post with friends, especially those interested in AI stories!
Useful Tools ⚒️
Vmaker AI Video Editor – Turn your raw videos into wow videos in minutes show
Reporfy – Insightful performance reports at the speed of AI show
ThirdAI PocketLLM – Personalized search and Q&A with AI on your documents show
AI Assist by Dopt – Build remarkably relevant in-product assistance show
Structify AI - Get human-quality data from anywhere for any use case. Structify allows you to be your data provider. This AI tool structures sources (news articles, SEC documents, and social media postings) into custom datasets on demand.
For this purpose, the developer uses AI agents and knowledge graphs to index info from anywhere and generate datasets automatically. The creators say it's as easy as describing the data you want (or defining a custom schema) in plain text.
Structify is not yet available to a mass audience, but you can already book a demo.
Speaking of which, Structify's announcement on Twitter proved so popular that it generated funder over 100 demo calls in 48 hours!
Subscribe to stay in tune with AI breakthroughs!
Weekly Guides 📕
6 Ways to Make Money With OpenAI SORA (when it comes out)
How to Make ChatGPT AI Content Sound Human
How AI Art Works & What Tools You Can Use To Generate It
AI Meme Of The Week 🤡
AI Tweet Of The Week
The trajectory of AI awakens the mastodons! (You'll find a link to the full recording of Sergey Brin's talk about AI at Bonus Materials).
(Bonus) Material 🏆
Lecture from Google founder Sergey Brin about GenAI
Five Best: Books on Artificial Intelligence - by WSJ
Protecting Your Custom GPT Against Prompt Injection Attacks
⚡️ How was this week in AI? Share your content and ideas in the comments to this post so we can discuss or include them in the next edition!