Google’s tiny AI beats GPT-3.5

Welcome, AI enthusiasts.

Google just released its recent AI model that BEATS other leading AI models like GPT-3.5 — at just 1/tenth the dimensions.

Because the race for more efficient AI continues to accentuate, are we on the tipping point for running advanced AI directly on our phones? Let’s get into it…

In today’s AI rundown:

• Google’s tiny AI model bests GPT-3.5

• Taco Bell’s AI drive-thru

• Use AI to vary image scenery in seconds

• AI reprograms brain cancer cells

• 7 recent AI tools & 4 recent AI jobs

• More AI & tech news

Read time: 4 minutes

LATEST DEVELOPMENTS

GOOGLE

✨ Google’s tiny AI model bests GPT-3.5

The Rundown: Google just unveiled Gemma 2 2B, a light-weight AI model with just 2B parameters that outperforms much larger models like GPT-3.5 and Mixtral 8x7B on key benchmarks.

The main points:

• Gemma 2 2B boasts just 2.6B parameters, but was trained on an enormous 2 trillion token dataset.

• It scores 1130 on the LMSYS Chatbot Arena, matching GPT-3.5-Turbo-0613 (1117) and Mixtral-8x7b (1114) — models 10x its size.

• Other notable key benchmark scores include 56.1 on MMLU and 36.6 on MBPP, beating its predecessor by over 10%.

• The model is open-source, and developers can download the model’s weights from Google’s announcement page.

Why it matters: As we enter a brand new era of on-device, local AI, lightweight and efficient models are crucial for running AI directly on our phones and laptops. With Gemma 2 beating GPT-3.5 Turbo at just 1/tenth the dimensions, Google is not only showing what’s possible — they’re cementing their position because the leader within the small model space.

TOGETHER WITH DEEP ATLAS

🧠 Grok machine learning from first-principles

The Rundown: Deep Atlas offers live, immersive machine learning courses for knowledgeable engineers.

Construct a first-principles understanding of ML:

• End-to-end model development, from data and model selection to training, evaluation, and deployment

• Training models using architectures like: transformers, GANs, RNNs, CNNs, autoencoders, and more

• Implementing pretrained models with RAG systems, fine-tuning, prompt engineering, and agentic flows

Apply now to affix the September cohort (limited seats remaining)!

TACO BELL

🌮 Taco Bell’s AI drive-thru

The Rundown: Taco Bell’s parent company, Yum Brands, just announced plans to roll out AI-powered drive-thru ordering at lots of of restaurants within the U.S. by the tip of 2024, with ambitions for global implementation.

The main points:

• The AI understands orders, auto-inputs them into the system, and even suggests additional items — potentially increasing sales through upselling.

• Over 100 Taco Bell restaurants within the U.S. already use voice AI in drive-thrus.

• The corporate has been testing the AI for over two years and claims it has outperformed humans in accuracy, reduced wait times, and decreased worker workload.

• Rivals like Wendy’s and White Castle are also experimenting with AI ordering, while McDonald’s recently ended its IBM partnership for similar tech.

Why it matters: If Taco Bell’s positive results on their two-year test are any indication, this large-scale AI implementation could change the best way fast-food chain businesses operate and the way we order food at drive-thrus. Nonetheless, the success (or failure) of this rollout could set the tone for your entire industry’s adoption.

AI TRAINING

🎨 Use AI to vary image scenery in seconds

The Rundown: Krea AI’s recent Scene Transfer feature allows users to recreate images with consistent lighting and color schemes, all with just a couple of clicks, completely free.

Step-by-step:

1. Visit Krea AI and enroll for a free account.

2. On the most important dashboard, click “Enhance” in the highest navigation bar.

3. Upload your image to the enhancement interface and activate the “Scene Transfer” option on the best side.

4. Upload a reference image or describe your required lighting within the prompt. Adjust strength as needed, then click “Enhance”.

Get more AI tutorials →

PRESENTED BY SECTION

🤖 Free workshop: 1 Hour to AI Proficiency

The Rundown: Section is hosting a free-to-join workshop, 1 Hour to AI Proficiency, going over real-world practical tricks to aid you implement AI in your day-to-day work.

You’ll learn:

• Prompting suggestions that transcend what you’ve seen on LinkedIn

• use AI for private and skilled tasks with in-person demos

• A proven talk track of “why AI matters” to share together with your team

Click here to enroll for the free workshop and begin implementing AI today.

AI RESEARCH

🧠 AI reprograms brain cancer cells

The Rundown: Researchers at USC made a breakthrough using AI to reprogram glioblastoma cells into immune-activating dendritic cells in mouse models, potentially revolutionizing treatment for the deadly brain cancer.

The main points:

• Glioblastoma is the deadliest adult brain cancer, with lower than 10% of patients surviving five years after diagnosis.

• AI identified genes that may convert glioblastoma cells into dendritic cells (DCs), which sample cancer antigens and activate other immune cells to attack the tumor.

• In mouse models, this approach increased survival probabilities by as much as 75% when combined with immune checkpoint therapy.

• Researchers have also identified human genes that might potentially reprogram human glioblastoma cells, paving the best way for future clinical trials.

Why it matters: By turning cancer cells against themselves, this recent research offers a novel option to fight tumors from inside. If the 75% increased survival probabilities in mice translate to humans, this might not only revolutionize glioblastoma treatment but potentially open doors for similar approaches in other hard-to-treat cancers.

NEW TOOLS & JOBS

Trending AI Tools

• 🗣️ OpenAI Advanced Voice – Natural and real-time conversations with AI, available for limited ChatGPT Plus subscribers

• 🌟 Midjourney 6.1 – Enhanced image quality, coherence, and recent upscaling features

• 👩‍🎤 Rendernet Narrator – Seamlessly lip-sync an AI character with natural expressions from a video

• 💸 Coachvox – Generate leads with an AI coach version of you

• 🎬 Respeecher – Hollywood-grade AI voice generation

• 🚀 OFFLIGHT – To-dos, schedules, and emails in a single place

• 🎨 Simply Draw – Learn drawing with fast AI feedback

Browse more AI tools →

Latest AI Job Opportunities

• 👩‍💼 xAI – People Generalist

• ⚖️ Rad AI – Legal Counsel

• 💰 Databricks – Finance Manager

• 🎨 Meta – Product Marketing Manager, Generative AI Creative

Browse more AI jobs →

QUICK HITS

Google expanded access to its “About this image” tool, making it available through Circle to Search and Google Lens, allowing users to quickly get context on images they encounter online or via messaging.

Vimeo launched an AI-powered video translation tool that may translate video, audio, and captions into multiple languages while replicating the unique speakers’ voices.

NEURA, a German robotics company, released a brand new video showcasing their humanoid robot 4NE-1 performing tasks like chopping vegetables, ironing cloths, solving puzzles, and more.

Synthesia introduced “Personal Avatars,” AI-generated lifelike avatars created from temporary webcam or phone footage, allowing users to create short-form videos for social media in multiple languages.

San Francisco supervisors banned using certain AI rental software like RealPage and Yardi — which were allegedly utilized by some landlords in the town to set higher rent prices based on competitor data.

A study from the University of Leeds found that AI could help predict the danger of bowel cancer returning in patients, potentially assisting doctors in identifying high-risk cases and avoiding unnecessary chemotherapy.

THAT’S A WRAP

SPONSOR US

Get your product in front of over 600k+ AI enthusiasts

Our newsletter is read by hundreds of tech professionals, investors, engineers, managers, and business owners all over the world. Get in contact today.

FEEDBACK

If you have got specific feedback or anything interesting you’d wish to share, please tell us by replying to this email.