🧙🏼‍♂️ GPT-4o mini, the new king of small models

Howdy, wizards.

AI can help you create stuff with code, no matter what background you have. Watch how this non-technical guy built a note taking app with Claude Artifacts and deployed it with Replit in just a few hours.

Dario’s Picks

The most important news stories in AI this week

1. OpenAI launches a low cost model. GPT-4o mini is OpenAI's most cost-efficient small model, can do up to 16K token output lengths (longer than the regular GPT-4o), and is good at function calling. This combination makes it a good choice for things like coding and long form content, and for building apps that need fast, real-time text responses like customer support chatbots. It’s available in ChatGPT and on the API. See how it stacks up against other models below.

   ‎

   ‎ Why it matters‎ ‎ GPT-4o mini shows us once again how quickly the price of AI is dropping: the cost per token is 99% lower than OpenAI's text-davinci-003 model from 2022.

   ‎

2. OpenAI is training AI to communicate more clearly. OpenAI found that when training advanced language models to generate text that can be easily verified by weaker models, humans also understood the text better. The technique involves a prover-verifier game, a framework where the prover generates a solution and the verifier checks it for accuracy.

   ‎

   ‎ Why it matters‎ ‎ Communication is key to any relationship – if AI will soon surpass human intelligence, then we better make sure we understand what it's saying. The research might be a step forward to improving both accuracy and readability of AI generated text, especially useful in fields where precise, clear communication is important.

   ‎

3. Anthropic and Menlo ventures launch the Anthology fund. It's a $100 million fund to fuel the next generation of AI startups, and leverages Menlo's (venture capital firm) experience in building company's with Anthropic's expertise and AI tech. The type of projects the fund will be backing are AI infrastructure, novel applications of AI, consumer AI, trust & safety tooling, and AI that maximizes societal benefits. Checks start at $100k and include $25k in Anthropic credits.

   ‎

   ‎ Why it matters‎ ‎ Funds like these are often the places where big, disruptive companies are born. If you're an AI startup with an innovative idea (eg not yet another PDF reader...) you might have a chance to get a head start with this fund – not only in capital but also an incredible network and access to the latest AI tech.

   ‎

4. Claude is now available for Android. It can pick your conversations where you left off on desktop, it has real-time language translation, and vision.

   ‎

   ‎ Why it matters‎ ‎ Having AI in your pocket opens up a ton of possibilities. Translate any voice or text while travelling, brainstorm on any subject anywhere, or my favourite using vision: identifying weird animals.

   ‎

5. Mistral launches a new mini model. The French AI startup launched Mistral-NeMo, a 12B multilingual model built in collaboration with Nvidia. It's available under the Apache 2.0 license and surpasses other open, small models from Meta and Google on key benchmarks.

   ‎

   ‎ Why it matters‎ ‎ Seems like it's the week for small models! Despite the lack of consensus of what open-source actually means when it comes to LLMs – what's beyond doubt is that the license here is fully permissible, meaning anyone interested can use, modify and distribute it, including for commercial purposes.

   ‎

6. TLDR covers the most interesting tech, science and coding news in just 5 minutes. TLDR covers the best tech, startup, and coding stories in a quick email that takes 5 minutes to read.

   ‎

   No politics, sports, or weather – we promise. And it's read by over 1,250,000 people. Subscribe for free ‎ (sponsored)

This issue is brought to you by

Grow Your Business with HubSpot's Free CRM

HubSpot offers an intuitive customer relationship management platform tailored for small businesses. Manage leads, track sales performance, and understand your customers with ease. Best of all, it’s completely free, with no limits on users or data, allowing you to store and manage up to 1,000,000 contacts.

Behind the news

GPT-4o mini: OpenAI’s new, low cost model

GPT-4o mini is OpenAI’s latest, low cost model – and it’s preferred over GPT-4 on the LLMSYS leaderboard. The model currently supports text and vision, with full multimodal support coming in the future.

It has a score of 82% MMLU, which is the best score on this benchmark among comparable models (Gemini Flash: 77.9%, Claude Haiku: 73.8%).

The biggest unique selling point for GPT-4o mini though, is its low cost – it can generate a 2,500 page book for 60 cents. It’s way more affordable than existing models, even the fast cheap ones:

The second big advantage of GPT-4o mini is its long output token length, which is double of the standard GPT-4o. If you’ve tried generating long outputs with ChatGPT, such as when coding, and reached the limit where you have to prompt again or click continue – that’s less likely to happen with this model.

‎ Why it matters‎ ‎ For ChatGPT users, we now have a go-to model when we expect really long outputs and don’t want to get cut off. For developers, it allows building and scaling AI apps and features at a much lower price.

The wizard’s favourite AI newsletters

what i’m reading right now

TLDR💨 - essential tech news in 5 mins. Read by 1.2 million people

simple.ai - 🕵🏻‍♂️ - deep-dives on AI by Hubspot’s co-founder

Bagel Bots🥯 - best hands-on tips & tricks

The Neuron😸 - easy weekday read on AI’s latest developments

Affiliate disclosure: To cover the cost of my email software and the time I spend writing this newsletter, I sometimes link to products and other newsletters. Please assume these are affiliate links. If you choose to subscribe to a newsletter or buy a product through any of my links then THANK YOU – it will make it possible for me to continue to do this.