Bot Eat Brain
Posts
Meta's Code Llama is here

Meta's Code Llama is here

PLUS: Alibaba's new AI models

Dan Bogachek & Michael Parrish
August 28, 2023

TOGETHER WITH

Good morning, human brains. Welcome back to your daily munch of AI news.

Here’s what’s on the menu today:

AMD & Mipsology team up 💽 🇫🇷
AMD acquired the team behind AI software Zebra.
Meta released Code Llama 🖥️ 🦙
The hotly anticipated coding LLM is out.
AIibaba’s AI models 🤖 🇨🇳
Two new multimodal AI models: Qwen-VL and Qwen-VL-Chat.

Sponsor Bot Eat Brain | New here? Subscribe!

APPETIZER

AMD acquires Mipsology 💽 🇫🇷

AMD acquired Mipsology. They’re the team behind Zebra.

Exactly like this one, except it’s different: 🦓.

Zebra is an AI software used in autonomous vehicles, robotic manufacturing, and smart retail.

Source: AMD

Cheese and wine:

1/ The strategy behind the acquisition: use Zebra to accelerate AI computing tasks on AMD chips.

2/ Mipsology has extensive experience tailoring Zebra for AMD hardware.

3/ AMD’s Vice president, Scott Herkelman, tweeted that it’s planning for “major product announcements“ at Gamescom 2023.

Source: @sherkelman

Our take: AMD launched the M1300x, its newest AI chip, in June. This new partnership is another maneuver to challenge NVIDIA.

But AMD isn’t the only challenger. They’re also contending against Google, Amazon, Microsoft, and virtually every tech giant — who are all developing their own AI chips to get a slice of the pie.

BUZZWORD OF THE DAY

Coding LLM

Coding LLMs are large language models that generate, understand, and optimize computer code. They’re designed to assist in code completion, bug detection, and code translation amoung other coding-related tasks.

FROM OUR PARTNERS

Do you want to master ChatGPT & AI — for free?

Join this 3-hour power-packed workshop by GrowthSchool to become the top 1% in ChatGPT & AI.

This workshop, worth $49, will be free for the first 100 Bot Eat Brain readers to register.

Click here to register. (Limited seats: free for the first 100 people only)

Attended by 250,000+ individuals & rated 9.5/10.

✅ In this workshop, you will learn how to:

Unleash ChatGPT and become a Master Prompter.
Discover the power of image, video, and audio manipulation.
Master Code Interpreter and Advanced Plugins Use Cases.
Build AI-powered automation and products in no time.

✅ Who is this workshop for?

Entrepreneurs and freelancers.
Creators, designers, coders, and students.
Working professionals in any field.

This workshop offers cutting-edge insights on AI, and practical techniques to make the best use of ChatGPT and will help you stand out from the crowd.

MAIN COURSE

Meta’s Code Llama is here 🖥️ 🦙

Meta released Code Llama. It’s a new fine-tuned coding LLM.

And it’s available in three sizes — 7 billion, 13 billion, and 34 billion parameters.

Source: Meta

Snippets:

1/ Code Llama outperformed Codex and AlphaCode. It scored similarly to ChatGPT in Benchmark tests.

2/ It’s trained on 500 billion tokens of code data.

3/ It’s built on Llama 2.

4/ It’s available on Perplexity.

Click here to play with it.

Source: Meta

Code Llama seems promising, but Meta’s unreleased “Unnatural“ model is also making waves. It’s trained on synthetic data, and developers have created fine-tuned versions that outperform GPT-4.

A LITTLE SOMETHING EXTRA

Alibaba’s new AI models 🤖 🇨🇳

Alibaba launched two new AI models — Qwen-VL (Qwen Large Vision Language Model) and Qwen-VL-Chat.

“New is always better.”

Both are multimodal, Vision-Language models.

Source: GitHub repo for Qwen-VL

Brass tacks:

1/ The models understand input images, texts, and bounding boxes, and generate captions based on open-ended queries.

2/ “Multi-image interleaved conversations,“ enable users to engage using multiple images and questions.

3/ Both models are open-source.

Our take: Qwen-VL and Qwen-VL-Chat were released on Chinese Valentine’s Day. Is AI love in the air? 😍

The “Multi-image interleaved conversations,” is a game-changer. It’s one of the first models that detects bounding boxes through open-domain language expression in both Chinese and English.

Other LLMs will undoubtedly incorporate this feature, and fast.

MEMES FOR DESSERT

YOUR DAILY MUNCH

Think Pieces

Nat Friedman and Daniel Gross discuss AI hype. They believe we’re at the height of a “AI hype cycle.“

Is it possible to build a trustworthy AI chatbot for news? A new chatbot called “Smart Answers” appeared on Macworld, PCWorld, Tech Advisor, and TechHive. More on how it’s performed so far.

Many companies block OpenAI’s GPTBot. Here’s why.

Startup News

NVIDIA plans to buy $25 billion of its shares back. Does NVIDIA believe it’s undervalued?

Qualcomm plans to release “Holy Grail“ AI on mobile devices. A new line of Snapdragon chips promises embedded AI, and not just for pretty photos.

LinkedIn recently upgraded its feed with AI. It now recommends more personalized content.

Research

Meta’s research paper on Code Llama.

Motion-to-Matching — a method that improves LIDAR detection for motion tracking.

Prompt2Model — a framework that supports model distillation, dataset generation, evaluation, dataset retrieval, and model retrieval research.

Tools

Recast [Sponsored] — turn any article into a concise audio podcast. Your AI co-hosts save you time by explaining all the important details with engaging dialogue.

Team Town [Sponsored] — your own in-house design team without those pesky design bottlenecks that slow you down. Trusted by Staples, Built, Jeep, and many others.

AI Dad Jokes — AI creates a “dad joke“ based on uploaded pictures.

WITAS (Wait is that a sticker?) — AI-powered sticker generation tool for mobile devices.

TWEET OF THE DAY

If you’re eager to work for Elon, now could be your chance. xAI is hiring.

Source: @xai

Tag us on Twitter @BotEatBrain for a chance to be featured here tomorrow.

AI ART-SHOW

“Light Line Round“ @sara21222122

Until next time 🤖😋🧠