- Bot Eat Brain
- Posts
- OpenAI and Scale's new partnership
OpenAI and Scale's new partnership
PLUS: NVIDIA's $6 billion AI profit
Michael Parrish & Dan Bogachek
August 25, 2023
Good morning, human brains. Welcome back to your daily munch of AI news.
Here’s what’s on the menu today:
Stephen King’s AI statement 📚👀
The horror legend acknowledges AI trained on his massive catalog.
OpenAI and Scale’s partnership 👫
Time to fine-tune some AI models.
NVIDIA profits over $6 billion from AI 🤑
H100 GPUs are the biggest cash cow by far.
APPETIZER
Stephen King’s AI statement 📚 👀
TL;DR: he doesn’t forbid AI to train from his books.
🤡:
1/ Stephen King says that stopping AI from his works mirrors King Canute’s effort to halt the tide.
2/ He states that one must read extensively to write well, and AI developers use this principle effectively.
3/ He states he is not currently threatened by AI, but the topic makes him uneasy.
4/ He references “Colossus,“ hinting at a future where humanity becomes overly dependent on sentient machines.
Our take: the list of authors suing AI companies is now over 8,000 long and grows every day. Stephen King’s perspective is a refreshing take amidst a reactionary anti-AI swell in the creative community.
BUZZWORD OF THE DAY
Fine-tuning
In the context of AI, fine-tuning refers to the process of taking a pre-trained model and adding more training with smaller, specific datasets to adapt it to a particular task or domain. This allows the model to leverage prior knowledge from the general training while specializing in the nuances of the new data.
MAIN COURSE
OpenAI and Scale’s partnership 👫
Open AI and Scale have partnered up as of yesterday. The goal: support more companies with fine-tuned OpenAI models.
They hope to squeeze more performance out of AI models with greater steerability and customization.
A fine-tuned GPT-4 is in the works as well.
Honeymoon:
1/ Open AI claims Scale is a preferred partner due to its effective, secure data use.
2/ Scale already fine-tunes many commercial and open-source models and will now leverage GPT-3.5’s APIs.
3/ Scale successfully fine-tuned GPT-3.5 for Brex, a startup bank. By transitioning from GPT-4 to a fine-tuned GPT-3.5, Brex was able to improve usage costs, latency, and quality.
Our take: the partnership seems more about commercial interest than true AI advancement. And while fine-tuned models offer a boost in performance, we’re wondering if it’s a bit of a detour from things like GPT-5. Are they buying time, or does this exercise set the foundation for newer GPT’s?
A LITTLE SOMETHING EXTRA
NVIDIA profits over $6 billion from AI 🤑
NVIDIA reported a 101% revenue growth in the second quarter. Its total revenue for Q2?
$13.5 billion.
Whaaa?
Profit surged by over 800% to $6.2 billion.
Offshore accounts:
1/ The primary driver: huge demand for its H100 GPU for large-scale AI model training.
2/ Its data center business grew by 171% year-over-year, which has contributed over $10 billion in revenue.
3/ Next quarter is projected to hit $16 billion in revenue.
Our take: Until now, NVIDIA had virtually no competition. Now it faces supply chain issues and tech giants (like Google, Amazon, Microsoft, etc.,) developing their own AI chips.
But the endless demand likely means NVIDIA will sell as much, and as quickly, as they can make.
MEMES FOR DESSERT
YOUR DAILY MUNCH
Think Pieces
How ChatGPT changed generative AI. How it evolved AI into an “anything tool.”
Can AI read minds? A virtual reality headset called “MinD-Vis“ allows users to control a metaverse with their minds.
How do AI agents act in nuclear code exchange? Do they cooperate or deceive each other? This simulation plays it out in real time.
Startup News
Naver, the South Korean internet search company, unveils AI features. The biggest one? An AI chatbot called HyperCLOVA X.
Tumblr will get AI enhancements. It will soon have algorithmic feeds and other AI features.
Ideogram AI bags $16.5 million in funding. The startup is founded by AI experts from Google Brain, UC Berkeley, CMU, and University of Toronto members.
Research
StableLLaVA — a method to enhance multimodal LLMS by synthesizing image-dialogue data.
ProAgent — a framework that uses LLMs to create a proactive AI agent that anticipates teammates’ actions and improves its own plan.
Tryage — a context-aware routing system that efficiently selects expert language models from a library based on user input prompts.
Tools
OpenArt [Sponsored] — create stunning AI images without writing any prompts.
Gentrace — a collaborative generative AI platform to evaluate different AI models simultaneously.
Profilio — an AI-powered recruitment tool to efficiently analyze CVs and find the best candidate.
Notably — AI-powered summaries, insights, and templates for research data.
TWEET OF THE DAY
The co-founder and CEO of Hugging Face announced a $235 million funding round, which puts the platform at a $4.5 billion valuation.
Tag us on Twitter @BotEatBrain for a chance to be featured here tomorrow.
RECOMMENDED READING
🏥 Healthcare AI News — 5 minutes or less is all it takes to elevate your Healthcare knowledge with this expert-curated weekly AI newsletter. Stay informed and stay ahead.
👨 The Average Joe — Market insights, trends and analysis to help you become a better investor. We like their easy-to-read articles that cut right to the meaty bits.
💌 Marketing Letter — The newsletter keeping 30k+ marketers in the loop. Read by marketers who work at LinkedIn, Techcrunch, and Disney.
AI ART-SHOW
Until next time 🤖😋🧠
What'd you think of today's newsletter? |