AI: Weekly Summary
...week ending September 22, 2023
Welcome to the Saturday weekly AI summary here on ‘AI: Reset to Zero’. Here are this week’s AI highlights:
Google and Microsoft show off AI Productivity App and OS Plans: Week saw Google releasing their AI Bard connecting to Google productivity apps and services. Reviews have been mixed, deeming it a work in progress. Microsoft also announced an updated CoPilot that works across Windows 11, Microsoft 365 Office productivity suite, Bing, Edge browser, and a slew of new Surface devices. AI enhanced WIndows 11 on September 26 next week. This area remains a slow work in progress, as I explain here. Separately, both companies are getting ready multimodal versions of their next generation LLMs. My deeper dive on that here.
Google YouTube ramping AI tools: YouTube got more AI enhanced tools for creators to produce videos, incorporating AI music and other capabilities, along with tools to reach a wider audience and connect with advertisers. This is separate from the company eventually using the vast trove of YouTube data to train its next generation LLM AI models from Gemini and beyond. More on that opportunity here and here.
OpenAI unleashes Dall3-3 via ChatGPT: Their new version of Dall-E3 was released, featuring full integration with ChatGPT now. This compares with MidJourney, still accessible via third party Discord servers. Initial reactions have been highlighting the more robust feature sets and capabilities. Stability, Adobe and others remain other popular options in the AI image generation arena.
Big Tech customers working hard on vertical AI infrastructure: Even while Google, Microsoft, Amazon and other big AI cloud data center providers are working as closely as possible with chip and networking providers like Nvidia and others, they’re also concurrently investing to build their own ability to create their own hardware/software chip infrastructure down the road. Example of this is Google’s efforts to build its Tensorflow TPUs in-house, with lesser dependence on infrastructure vendors like Broadcom. Amazon AWS similarly continues to explore more third party chip sourcing and their own custom solutions down the road, with a focus on reduced dependence on Nvidia over time. Apple is a poster child for this in what they’ve done with Apple Silicon leveraging their Arm Holdings and TSMC relationships vs Intel and others to date. Nvidia remains in the lead position for now.
Amazon Alexa with LLM AI soon: At its annual release of updated Alexa and other hardware products at its Fall Devices Event this week, Amazon announced a new AI model to power an updated line of Alexa and Echo devices to come. This is concurrent with similar efforts by Google with its Nest and Google Assistant products, and the next versions of voice assistant Siri by Apple. Voice enabled LLM AI features are coming soon in the coming months from all the usual suspects.
Other AI Papers and Research:
Meta’s new approach to reduce LLM AI Hallucinations: Called ‘Chain-of-Verification (COVE), the new paper outlines new methods to combat hallucinations in LLMs, significantly reducing errors and misinformation. More detailed run-through here.
New VC AI market maps and projections: Sequoia had an updated report on Generative AI market, with market maps, and a post discussing current AI infrastructure spend on LLM AI compute in the hundreds of billions vs the eventual possible enterprise and consumer markets. A16z had a counter-view to that via Twitter/X, with a distinctly more optimistic outlook.
Thanks for joining this Saturday with your beverage of choice.
For your Sunday tomorrow, this week’s ‘AI In my View’ is coming up. Stay tuned.
(NOTE: The discussions here are for information purposes only, and not meant as investment advice at any time. Thanks for joining us here)