- Future Perfect
- #102. 🤕 LLMs' biggest weakness 🚗 Tesla's $300M AI cluster | 🙊 How AWS gave MS an opening in AI
#102. 🤕 LLMs' biggest weakness 🚗 Tesla's $300M AI cluster | 🙊 How AWS gave MS an opening in AI
PLUS: Balancing daily challenges ✨ AI-powered ideas ✨ viral posts
❓Is this new format of Future Perfect too short now?
Instead of 9 article summaries per issue it's just 3 now. Is that ok for you?
📰 AI NEWS
RAG (Retrieval Augmented Generation) is a new AI method that helps models answer questions with information not in their original training set by accessing a vector database.
Traditional model-building methods like fine-tuning or building from scratch are costly and have drawbacks, such as a risk of the model "forgetting" original training data.
Companies are increasingly adopting RAG, storing data as "embeddings" in vector databases from providers like Pinecone, Weaviate, and Chroma.
The RAG approach still faces challenges in determining the most relevant information to answer complex queries; however, investor enthusiasm for this method is growing.
Tesla is launching a new supercomputer with 10,000 Nvidia H100 GPUs, offering peak performance surpassing the world's fourth most powerful supercomputer, Leonardo.
The supercomputer aims to accelerate the training of Tesla's Full Self-Driving (FSD) technology, potentially making Tesla more competitive in the automotive industry.
Due to high demand and supply constraints for Nvidia GPUs, Tesla is also developing its own custom supercomputer called Dojo, which will handle data processing for Tesla's vehicle fleet.
Elon Musk revealed plans to spend $4 billion on AI and FSD training over 2023 and 2024, highlighting Tesla's commitment to resolving computational bottlenecks.
AWS planned to release its own large language model (LLM), Bedrock, but postponed it due to technical issues. Upon seeing ChatGPT's capabilities, AWS acknowledged that Bedrock was not at the same level.
AWS shifted its strategy, repurposing the name Bedrock for a new service that allows integration with various large-language models. A limited release occurred in April, with plans for wider availability before the end of September.
AWS missed opportunities for partnerships with AI startups like OpenAI, Cohere, and Anthropic. Intuit, a significant AWS customer, substantially increased its spending on Microsoft’s Azure mainly due to its use of OpenAI software.
AWS contends that the market for AI cloud services is still nascent, with AWS itself generating $80 billion in revenue last year. Despite customer interest in new offerings like Bedrock and Titan, these services are not yet at par with competitors like ChatGPT.
⚖️ AI LAW FROM THE MARC HOAG LAW BLOG
Sorry, nothing new for today; please check back again tomorrow!
⚡️ CHATGPT POWER PROMPTING!
Today’s a 2fer! Here’s a couple short-and-sweet but super practical and helpful prompts across two different domains of life.
(1) Make two columns. On one side, list the tasks that bring you joy at work; on the other, tasks that drain you. For the draining tasks, outline a plan to delegate or eliminate them within a month.
(2) I had a disagreement with my boss at work. Here's what happened: [insert key facts]
I'd like to write him an email but I'm not sure how to write it. Please make sure to write it on friendly/familiar terms but with a tone of formality since he is, after all, my boss.
Whatever I ask you, make sure that you ask me clarifying questions in return if you need anything more from me. If you do not need anything further, please say something like "I don't need anything else" or similar.
💡 AI RESOURCES
🎉 THAT’S ALL FOR TODAY!
What'd you think of today's email?
This helps me improve it for you!
🙏 PLEASE SHARE this post if you liked it!
Thanks for reading!
Looking for past newsletters? You can find them all here.