Apple Joins Open-Source, OpenAI’s Chip Challenge, Tiny Titans: Mistral and Nvidia Reveal Compact AI Powerhouse

Welcome to another DSA edition! From the lab to the marketplace, AI is reshaping our world at lightning speed. This week’s top stories showcase the breadth and depth of AI advancements:

• Hardware Revolution: Tech giants are racing to build custom AI chips, with OpenAI and the Mistral-Nvidia duo leading the charge.
• Humanoid Horizons: AI is stepping out of the digital world and into ours, with new strides in robotics and humanoids.
• Model Mania: OpenAI’s latest GPT-4o mini and Apple’s open-source model are set to democratize AI access.
• AI for Everyone: From building your own web app to discovering new AI tools, this week’s news empowers users and businesses alike.

Want to dive deeper into the world of AI? Join our community and gain access to exclusive content, insights, and tools. Sign up for our newsletter and collaborate with fellow AI enthusiasts.

Let’s shape the future of AI together!

OpenAI wants its own AI?chip

The Brief: OpenAI is reportedly in talks with chip designers like Broadcom to develop its own AI chip, aiming to reduce dependence on scarce and expensive GPUs from Nvidia.

The details:

• OpenAI has already hired former Google employees who worked on Google’s tensor processing unit (Google’s AI chip).
• The ChatGPT maker has been talking to chip designers including Broadcom, however production of the new chip isn’t expected until 2026 at the earliest.
• The company is exploring various chip packaging and memory components to optimize performance.
• OpenAI is also considering creating new companies with outside investors to finance infrastructure like data centers.

Why it matters: This move isn’t just about OpenAI playing chip designer?—?it’s a power play. By developing its own chips, OpenAI could break free from the GPU shortage bottleneck, potentially supercharging its mission towards AGI.

Mistral and Nvidia drop small AI powerhouse

The Brief: Mistral AI and Nvidia just unveiled Mistral NeMo, a new open-source, 12B parameter small language model that surpasses competitors like Gemma 2 9B and Llama 3 8B on key benchmarks alongside a massive context window increase.

The details:

• NeMo features a 128k token context window, and offers SOTA performance in reasoning, world knowledge, and coding accuracy for its size category.
• The model also excels in multi-turn conversations, math, and common sense reasoning, making it versatile for various enterprise applications.
• Mistral also introduced ‘Tekken’, a tokenizer that represents text more efficiently across 100+ languages, allowing for 30% more content within the context window.
• NeMo is designed to run on a single NVIDIA L40S, GeForce RTX 4090, or RTX 4500 GPU, bringing powerful AI capabilities to standard business hardware.

Why it matters: Small language models are having a moment?—?and we’re quickly entering a new shift toward AI releases that don’t sacrifice power for size and speed. Mistral also continues its impressive week of releases, continuing to flex the open-source muscle and compete with the industry’s giants.

AI isn’t a magic wand for customer?service

The Brief: While every company is jumping on the AI bandwagon, 64% of customers prefer companies that don’t use it for service. Bottom line?—?half-baked AI tools can compromise your customer experience.

Help Scout’s support platform is made for customer-first teams, featuring:

• AI generated drafts you can edit before sending
• AI summaries that get the team up to speed quickly
• An AI assistant that makes it easy to escalate unresolved issues to your team

Try Help Scout today and experience the perfect blend of human touch and AI efficiency.

The state of AI humanoids and?robotics

The Brief: Coatue Management, an investment firm with over $70B assets under management, has released a comprehensive report on the current state and future prospects of AI humanoids and robotics.

The details:

• Coatue argues that robotics won’t have a sudden “ChatGPT-like” moment due to restricting factors such as physical limitations to adoption, high upfront costs, and a nascent ecosystem.
• 2023 saw a surge in new robotics companies, with over 20 launching humanoid initiatives since 2021?—?which is expected to continue in 2024.
• The report predicts that AI and quality training data, rather than hardware, will be the key differentiator for humanoids.
• Coatue envisions robots going through a progression from warehouse and store automation to more complex tasks like firefighting, surgery, and in-home assistance.

Why it matters: While we might not see a robot butler in every home overnight, Coatue’s analysis suggests we’re on the cusp of a gradual yet profound transformation. As AI and robotics capabilities increase, many industries, from manufacturing to healthcare, will be completely reshaped.

OpenAI debuts new GPT-4o mini?model

The Brief: OpenAI just announced the launch of GPT-4o mini, a cost-efficient and compact version of its flagship GPT-4o model?—?aimed at expanding AI accessibility for developers and businesses.

The details:

• GPT-4o mini is priced at 15 cents per million input tokens and 60 cents per million output tokens, over 60% cheaper than GPT-3.5 Turbo.
• The model scores 82% on the MMLU benchmark, outperforming Google’s Gemini Flash (77.9%) and Anthropic’s Claude Haiku (73.8%).
• GPT-4o mini is replacing GPT-3.5 Turbo in ChatGPT for Free, Plus, and Team users starting today.
• The model supports a 128K token context window and handles text and vision inputs, with audio and video capabilities planned for future updates.

Why it matters: While it’s not GPT-5, the price and capabilities of this mini-release significantly lower the barrier to entry for AI integrations?—?and marks a massive leap over GPT 3.5 Turbo. With models getting cheaper, faster, and more intelligent with each release, the perfect storm for AI acceleration is forming.

Showcase Your AI-generated Artistic Creativity

The Brief: The Stanford Technology Training Program is sponsoring an AI-Generated Visual Art competition. Your entry should explore this year’s Techie Festival’s theme of: AI New Frontiers.

Create stunning images or videos showcasing your best AI work to win amazing prizes, including:

• First Place: Apple iPad Pro
• Second Place: Apple Watch
• Third Prize: AirPods

Submit your entry today! For more details on how to enter, click here.

Apple’s new 7B open-source AI?model

The Brief: Apple just unveiled a new 7 billion parameter open-source AI model, dubbed DCLM-7B, that outperforms Mistral 7B and closes in on other leading open models like Llama 3 and Gemma across key benchmarks.

The details:

• The DCLM-7B model achieves 63.7% 5-shot accuracy on MMLU, surpassing Mistral 7B and nearing performance of Llama 3 and Gemma.
• Apple open-sourced not just the model weights and training code, but also the complete pretraining dataset.
• The model was trained on 2.5 trillion tokens using pretraining recipes based on the OpenLM framework.
• The model was released alongside the family of DCLM models, and are available on Hugging Face.

Why it matters: While Apple has been a bit slow to the open-source hype train, developers from around the world will still appreciate the unexpected open dataset. What this means is that researchers and developers now have open access to not just the model, but the entire process behind its creation.

? AI?RESEARCH

Groq’s new AI models surge up leaderboard

The Brief: AI startup Groq just released two new open-source AI models specializing in tool use, surpassing heavyweights like GPT-4 Turbo, Claude 3.5 Sonnet, and Gemini 1.5 Pro on key function calling benchmarks.

The details:

• Groq’s two models, Llama 3 Groq Tool Use 8B and 70B, are both fine-tuned versions of Meta’s Llama 3.
• The 70B achieved 90.76% accuracy on the BFCL Leaderboard, securing the top position for all proprietary and open-source models.
• The smaller 8B model was not far behind, coming in at №3 on the leaderboard with 89.06% accuracy.
• The models were trained exclusively on synthetic data, and are available through the Groq API and on Hugging Face.

Why it matters: Groq made waves earlier this year with its blazing-fast AI speeds?—?and now its pairing those capabilities with top-end specialized models. Near real-time speeds and highly-advanced tool use opens the door for a near endless supply of new innovations and user applications.

AI TRAINING

Build Your Own AI-Powered Web App for?Free

The Brief: Learn how to create a web application using AI without any coding experience.

The details:

• Utilize Claude 3.5 Sonett to generate code for your app.
• Store data locally on the user’s browser.
• Deploy the app for free using CloudFlare Pages.

Why it Matters: This method empowers individuals to build and deploy web applications without technical expertise, fostering innovation and creativity.

…

How to test and customize GPT-4o?mini

The Brief: Last Thursday, OpenAI revealed the GPT-4o mini, a compact version of its flagship GPT-4o model. While the model is slowly rolling out into ChatGPT accounts, the OpenAI Playground lets you access and customize the model instantly.

Step-by-step:

1. Go to OpenAI’s Playground and sign in with your credentials.
2. On the left side, click where it says “Chat”.
3. Click on the Models dropdown menu and select “gpt-4o-mini”.
4. Enter a prompt where it says “User,” play around with the different parameters (e.g., system prompt, temperature, or maximum length), and personalize your outputs.

Get more AI tutorials →

…

Automate your content creation?workflow

The Brief: Zapier Central can automatically generate content drafts based on your writing style and preferences, saving you time and effort with your personal or business content creation process.

Step-by-step:

1. Visit Zapier Central and sign up for free.
2. Click “New bot” on the dashboard to create your AI content creation bot.
3. Set up the Google Docs integration in the “Instant actions” section.
4. Upload a writing sample in the “Data sources” tab as a reference for the AI.
5. Instruct your bot to generate content and create a new Google Doc with the draft.

Get more AI tutorials →

…

Leverage AI to grow on LinkedIn with?Taplio

The Brief: Taplio is the all-in-one, AI-powered LinkedIn tool to grow your personal brand?—?keeping you steps ahead of the competition with stand-out posts and an engaged audience.

With Taplio, you get:

• AI-powered content creation to craft posts and carousels in seconds
• Access to a library of over 4M viral posts to inspire content and thread hooks
• Smart scheduling to find your post’s optimal timing
• Advanced analytics to track key metrics like followers, impressions, and engagement

Get started now. P.S.?—?be among the first 25 readers to use code RUN50 and enjoy 50% off a yearly plan (5 months free)!

NEW TOOLS & JOBS

Trending AI?Tools

• GPT-4o mini?—?OpenAI’s latest model that tops GPT-3.5 turbo in efficiency and pricing
• Cohesive AI?—?AI-powered web scraping and research in Google Sheets
• Immersity AI?—?Convert 2D images to 3D motions and videos
• Proton Scribe?—?Privacy-first email writing assistant with end-to-end encryption
• Shaped?—?Increase engagement and conversion with adaptive recommendations
• Kapkap?—?AI-powered dynamic video subtitle editor

Browse more AI tools →

New AI Job Opportunities

• Character AI?—?Brand Designer
• Waymo?—?Research Scientist, Foundation Models for Simulation and Planning
• Glean?—?Community Manager
• Fiddler AI?—?Machine Learning Engineer

Browse more AI jobs →

QUICK HITS

Anthropic is reportedly working on a new screenshot tool for Claude, potentially allowing users to seamlessly take screenshots from other screens or tabs.

Electronic Arts created digital clones of 11,000 college football players for their upcoming video game in 3 months?—?using AI to create the avatars using athletes’ headshot photographs.

California state legislators are pushing for a proposed bill that would require big tech companies to test AI for “catastrophic” risks before public release.

Ukraine is reportedly developing AI-enabled war drones, aiming to overcome Russian signal jamming and enable swarm operations on the battlefield.

AMD claims its new laptop chips can outperform Apple’s M3?—?boasting improved performance in multitasking, image processing, and gaming.

Google is reportedly developing a ‘Prompts Gallery‘ to allow users to curate a collection of favorite prompts and get inspiration from other users within it’s Gemini chatbots.

Perplexity rolled out Voice Mode to it’s AI assistant on its iOS app, allowing Pro users to chat and ask questions to the AI search engine through various voice modes.

Sam Altman revealed in a post on X that the alpha release of OpenAI’s Voice Mode is coming ‘later this month’, with general availability coming ‘a bit after’.

AI pioneer Fei-Fei Li launched World Labs in April, a ‘spatial intelligence’ startup already valued at over $1B?—?working on developing AI capable of understanding and navigating 3D environments.

OpenAI is reportedly in talks with Broadcom to develop a new AI chip, hoping to reduce reliance on Nvidia and boost global semiconductor infrastructure.

Tech giants including Google, OpenAI, Microsoft, and Amazon formed the Coalition for Secure AI (CoSAI), an open-source initiative aiming to develop best practices and address challenges in AI security.

Meta and Apple are planning to exclude EU customers from upcoming AI releases, citing regulatory uncertainty surrounding data usage and training under GDPR data protection laws.

OpenAI also introduced new compliance and administrative tools for ChatGPT Enterprise, aimed at supporting large-scale AI deployments in regulated industries.

SPONSOR US

Get your product in front of over 700k+ AI enthusiasts

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

How would you rate today’s newsletter?

Comment below to help us improve the newsletter for you.