THE SUPER BOWL

🏈 AI takes spotlight in Super Bowl commercials

Image source: BodyArmor

The AI Tech studio: AI made a splash in this year's Super Bowl commercials, with companies showcasing both its benefits and its potential for absurdity during the first big game in the artificial intelligence era.

The details:

• Microsoft’s Copilot spot (which we wrote about last week) showcased its AI assistant to the masses.

• A Google Pixel spot highlighted a new AI ‘Guided Frame’ feature helping blind or low vision users to more easily take photos.

• Illumination took a shot at AI video while promoting the upcoming ‘Despicable Me 4’ film, featuring Minions prompting comically bad generations.

• A HeGetsUs ad of people washing feet (🤔) faced heavy criticism, including speculation that the images were AI-generated — though Ad Age reported they were real photographs.

• BodyArmor’s ‘Field of Fake’ took a jab at AI generations, showing low-quality AI videos while pushing its ‘real sweeteners’ over ‘artificial’ alternatives.

• Etsy’s commercial showed off its Gift Mode feature, an AI recommendation tool to help users curate gift ideas.

Why it matters: While AI-related ads have been rampant over the last year, its Super Bowl presence is a big mainstream moment. The wide range of commercials is also representative of the mixed sentiment across the country — with many brands and people still conflicted on how to feel about the tech.

TOGETHER WITH OCTOAI

💰 Cut LLM costs by 50% for text summarization using Mixtral

The AI Tech Studio: If you are building products in AI and scaling text summarization features around GPT, you’re leaving money on the table by not exploring cheaper open-source options such as Mixtral.

Enter OctoAI:

• The fastest place to run open-source models

• Find the most cost-effective pricing and swap in Mixtral in 3 lines of code

• Benchmarks show 50%+ cost savings over GPT 3.5 with comparable quality

Check out all the data here, then sign up for OctoAI to get $10 in free tokens to put Mixtral to the test.

OPENAI

💰 Altman seeks TRILLIONS for global AI chip initiative

Image source: Bloomberg

The AI Tech Studio: OpenAI CEO Sam Altman is reportedly angling to raise TRILLIONS of dollars to reshape the semiconductor industry and massively boost global chip-building capacity to support AI development.

The details:

• Altman is said to be looking for between $5-7 trillion to ramp up global AI chip production, hoping to address both current and future projected shortages.

• Discussions are reportedly underway with investors like the UAE government, with plans also involving partnerships with chipmakers like TMSC.

• Altman wants OpenAI to be the primary customer of the new initiative, though US concerns over foreign tech influence could complicate plans.

Why it matters: Altman once tweeted that ‘nobody can out-accelerate him’, and aiming to raise nearly 10% of the world’s GDP would seem to back that up! Altman’s ambitions also show he sees AGI creation as a supply problem, not a tech/innovation one.

AI TRAINING

🌅 How to generate and edit Dall-E 3 images in Copilot

Image source: DALLE—3

The AI Tech Studio: Microsoft Copilot AI now supports inline image editing and styling and has improved prompt suggestions for DALL-E 3 images. This means you can now edit images directly within Copolit!

Step-by-step:

1. Head to Microsoft Copilot.

2. Enter a prompt to create your desired image.

3. Select your preferred image to change the style or elements using a menu of options.

4. Edit the background of the image, such as adding “Blur“ or “Color pop“ by moving your mouse over the image and selecting the main subject.

AI RESEARCH

🎥 Enabling AI to understand long videos

Image source: Midjourney

The AI Tech Studio: Researchers from Google DeepMind and the University of Cornell just developed a method for AI systems to comprehend events across longer videos better.

The details:

• Currently, most video AI models are limited to short clip lengths due to both the complexity and computing needed.

• The new method, called MC-ViT, stores a compressed "memory” of past video segments, allowing the model to efficiently reference past events.

• The methodology is inspired by human memory consolidation theories from psychology and neuroscience.

• The system achieves state-of-the-art accuracy in tasks like action recognition and question answering, despite using far less resources than comparable and much-larger models.

Why it matters: Enabling AI to understand longer videos is a necessary step forward for expanding its reasoning and real-world application use cases. Even more impressive is accomplishing it with lower computing and off-the-shelf transformers — continuing to push boundaries in AI with fewer resources.

AI tools powered by Supertools

• 📚 Augie Storyteller- Discover personalized bedtime stories brought to life with AI (link)

• 👨‍💻 CodeSignal Cosmo- Hone in-demand tech skills with your AI instructor, Cosmo (link)

• 🖼️ Midjourney GPT- Crafts detailed Midjourney image prompts (link)

• 🏝️ Jan- Experience the power of AI offline (link)

• ✍️ Citation Copilot GPT- Academic citation assistant (link)

• 🎙️ Rizzle AI- Convert text and podcasts into captivating videos (link)

• 📚 Book Summarizer GPT- Summarizes books and provides key takeaways (link)

AI jobs powered by The AI Tech Jobs

• 🎯 Hive- Account Executive - Sports, Media, & Marketing at Hive (link)

• 🌟 Anthropic- Policy Outreach Lead (link)

• 🔒Open AI- Enterprise Security Engineer (link)

• 🇫🇷 Synthesia- Customer Onboarding Manager - French Speaking (link)

*Sponsored listing

Researchers introduced AiSee, a a wearable device which utilizes a camera, AI algorithms, and bone conduction headphones to help visually impaired individuals "see" and understand their surroundings.

Pika Labs released a Valentine’s Day experience, allowing users to create a ‘Pika Boo’ AI video by answering a selection of questions.

Artists achieved a partial victory in a copyright battle against AI art generators, with a judge refusing requests form Stable Diffusion, Midjourney and DeviantArt to dismiss a lawsuit.

Nvidia's Grace Hopper GH200 superchip performed on par with or surpassed high-end AMD and Intel CPUs in initial benchmarks, validating initial performance claims.

A reddit user discovered that both Gemini and Copilot believed the Super Bowl had already happened prior to the game, hallucinating stats and results.

Imran Khan, Pakistan’s jailed former prime minister, utilized an AI video to declare victory from prison after his P.T.I. party scored a surprising win in the chaotic and contested election.

An AI deepfake video featuring deceased Indonesian dictator Suharto went viral ahead of the country’s elections, urging citizens to support his former party and raising concerns of AI voter manipulation.