Inflection releases massive new AI model

Read time: 2.5 minutes

Welcome back, Superhuman

While many of us are winding down for Thanksgiving, AI startups are trying to make the most of the crisis at OpenAI by launching one major AI product on top of another.

After Anthropic and Grok made major announcements earlier this week, Stability AI just dropped a new AI model that it claims can beat Google and Meta’s models. Big words.

TODAY’S MENU

• Data will decide who wins the AI race

• Inflection releases new AI model, claims it can beat Meta and Google

• Tutorial: An Introduction to LLMs and how they work

• Friday Laughs: The real reason Microsoft wants Sam Altman

• AI Generated Image: Futuristic cars

TODAY IN AI & TECH

• Chips Ahoy: Chipmaker Broadcom completes $69 Billion acquisition of VMware.

• Virtual Connection: A new Church of AI wants to build a connection between humans and AI.

• Hammered: US court throws out most of Sarah Silverman’s case against Meta’s use of AI.

• Missed Call: Nvidia’s chips for the Chinese market suffer delay.

• Bargain Hunter: Check out these Black Friday deals for tech products.

INSIGHT

Ownership and access to unique data will decide which companies win the AI race

The race to build the biggest, baddest, meanest AI models is heating up as startups and tech giants alike swallow tens of billions of dollars to buy GPUs and train AI models in search of use cases that unlock enormous new revenue streams.

And while GPT-4 is currently leading the pack for general purpose models, competitors like China’s Baidu claim that their models are just as good, and many industry-specific AI models are also carving out valuable niches.

But what determines how well a model performs? And who has the ability to build the best models? The answer: companies that have access to the biggest and most unique datasets.

An OpenAI employee recently claimed that all AI models trend towards the same endpoint regardless of training tactics or strategy, and that the main differentiator is the dataset. To which Elon Musk replied: “data is the spice.“

This could partially explain why Musk purchased Twitter (Now X). On top of the tens of billions that Tesla is investing into AI, Musk now has access to one of the biggest text datasets in the world with X’s purchase. No wonder Musk highlighted Grok’s access to X’s posts as a unique selling point when announcing the chatbot.

Venture capitalist Chamath Palihapitiya highlighted some more companies that own datasets with “huge value“:

• “Tesla’s real world environment data”

• “Meta’s real people preference and behavioral data”

• “Google’s search intent, YT behavior and Gmail”

• “Amazon’s purchasing behavior”

As the race to build the most powerful AI models ramps up, expect those with the ability to produce and own unique datasets to capture most of the value.

IN PARTNERSHIP WITH GROWTHSCHOOL

AI & ChatGPT Workshop (Black Friday Offer)

Join this 3-hour power-packed workshop (worth $99) for free and learn hacks to 10X work output and grow your business. Click here to register 

With AI & ChatGPT, you will be able to:

✅ Make smarter decisions based on data in seconds using AI

✅ Automate daily tasks and increase productivity & creativity

✅ Skyrocket your business growth by leveraging the power of AI

✅ Save 1000s of dollars by using ChatGPT to simplify complex problems

Click here to register (Limited seats: free for First 100 people only)

NEXT IN AI

Inflection AI releases new AI model, claims it can beat Meta and Google’s models

While the big boys fight it out to see who can build the best AI models, startups aren’t planning on watching from the sidelines. Founded by LinkedIn cofounder Reid Hoffman and DeepMind cofounder Mustafa Suleyman, Inflection AI just announced Inflection-2, a new AI model that it claims is “the best model in the world for its compute class and the second most capable LLM in the world today.“

Big claims. But Inflection has some big numbers to back it up. Inflection-2 outscored Meta’s LLaMA 2 and Google’s PaLM 2 on various benchmarks, according to the company. It also did nearly as well as OpenAI’s GPT-4 on some tasks.

CEO Mustafa Suleyman said that the model will soon be integrated into the company’s Pi chatbot.

BUZZWORD BUSTER

RLHF

Reinforcement Learning from Human Feedback (RLHF) is a technique used for developing AI models like the one behind ChatGPT. RLHF combines traditional datasets with human feedback. It’s a powerful training method that lets models learn from real world interactions and adapt to human preferences.

5 AI Tools to Supercharge Your Productivity

Postnitro: Create carousels for social media in seconds.

Embra: AI assistant for professionals. Connect your data, workflows, and team to save hours of work.

Taplio: Use AI to get more reach, engagement and followers on LinkedIn.

Submagic PRO: Generate video captions effortlessly using AI.

Dear AI: Generate intimate, thoughtful, and beautiful letters instantly with AI.

TUTORIAL

An introduction to Large Language Models (LLMs) by OpenAI scientist Andrej Karpathy

FRIDAY LAUGHS

A lighthearted moment from AI to kickstart your weekend

AI GENERATED IMAGES

Futuristic cars

Prompt used: A futuristic version of a [insert brand name] car

📈 Feature your product in the world’s biggest AI newsletter

Superhuman is the world’s biggest and fastest-growing AI newsletter with 450,000+ readers working at the world’s leading startups and enterprises. Companies like Masterworks, Brave, SEMRush, and 1Password feature their products in Superhuman. Main ads are sold out for the next 4 weeks. You can book future ad spots here.  

🧞 Your wish is my command

Reviews of the day

Thanks for reading.

Until next time!

Zain

p.s. if you want to sign up for this newsletter or share it with a friend or colleague, you can find us here