AI: Weekly Summary
...week ending August 4, 2023
Third edition of the Saturday weekly AI summary here on ‘AI: Reset to Zero’. We’re in the last stretch of a scorching summer everywhere with August. On the AI front, it was anything but a slow week.
OpenAI makes ChatGPT better: The company announced a slew of improvements for both subscribers to ChatGPT users. Important one is defaulting to GPT4 (earlier used to be a choice), which indicates perhaps that they’ve freed up additional GPU Compute capacity. Also, they now allow multiple file uploads to their very popular Code Interpreter Plugin, which is a big feature for power users, and a variety of other improvements. This in addition to their global release of a ChatGPT app for Android recently, makes sure that ChatGPT remains front and center for millions of users worldwide. All welcome improvements. I for one am looking forward to not having to pick GPT4 at every App open.
Voice to get LLM AI makeovers: Kicks off with Google prepping a big LLM AI makeover of their #2 market leading Google Assistant and Nest products and services. Seeing similar signs from Amazon as well, with a re-org of their AI and market leading Alexa units. Apple under AI chief ‘JG’ is likely focused on similar ‘ML’ efforts to improve Siri, which used to be the market pioneer in the currently $9 billion Voice assistant market. OpenAI’s latest iOS ChatGPT app supports voice input, and presumably more capabilities on the voice front to come via Partners. Get ready for talking to AI Chat clients shortly in addition to text Prompts. Expected development when LLM AI has the ability to make us all ‘software programmers’, as Nvidia founder/CEO Jensen Huang reminds us all. Maybe we’ll see “Voice Prompt Engineers” as the new, hot hiring category in addition to ‘Prompt Engineers’.
Meta readying AI-powered chatbots: Just like we saw ahead of Meta’s Llama 2 release, the drumbeat gets louder around the Zuckerberg led AI ‘smart agents’, coming soon via Instagram, Facebook and possibly other Meta applications. As Techcrunch highlights an FT report: “Meta is preparing to launch a range of artificial intelligence-powered chatbots that exhibit different personalities as soon as next month, in an attempt to boost engagement with its social media platforms. The tech giant led by chief executive Mark Zuckerberg has been designing prototypes for chatbots that can have humanlike discussions with its nearly 4bn users, according to three people with knowledge of the plans. These people said some of the chatbots, which staffers have dubbed “personas”, take the form of different characters.” As expected, Meta sources emphasize safety related precautions to mitigate misuse of these AI ‘personas’. To be seen.
Google adds LLM AI to Robots: Google is leveraging their deep experience at DeepMind with robots, adding LLM AI capabilities to robots, specifically one named RT-2 (Robotic Transformer-2). Like the Voice category, robots are also overdue for LLM AI upgrades, be they from Elon’s ‘Optimus” robots, Boston Dynamic (Hyundai) with their social media-savvy robots, Softbank’s robot aspirations, and of course Amazon, with their budding ‘Astro’ robot line, especially after their purchase of Roomba. Google’s RT-2 development is pretty impressive from the NYTimes account. There’s also a deeper AI Paper for those interested. This whole area is so cool from my perspective, I had to do a longer write-up on why AI and robots are like peanut butter and jelly.
Strong Results & AI Focus by Amazon & Apple: We now have all five of the big tech quarterly results, following Meta, Microsoft and Google last week, collectively approaching a $10 trillion market capitalization. First, Amazon reported strong AWS results, and made notable progress on improving their Retail Flywheel with substantially improved Same Day delivery in major markets. Notably, Amazon reported strong customer interest and engagement around deploying custom Generative AI solutions using a range of open and closed LLM AI models. Apple reported strong high margin Services revenues (App store & Subscriptions), which although only a quarter of their total revenue, reports over 70% of the gross margins for the quarter. Shows that their multi-year strategy of focus on this revenue stream is working. Have links to their transcripts above for further reading. Also, here’s a summary write-up I did on the quarter-end reports of the Big Tech companies, Meta, Microsoft, Google, Apple and Amazon, and the AI tail-winds.
In AI Papers of note,
As highlighted above, the Google DeepMind RT-2 “Vision-Language-Action Model” AI Paper is notable. Lots of cool details for those more technically inclined.
LoraHub: “Boosting Generalization across Tasks with Efficient LoRA Composition”, focuses on LLM computational efficiency and memory usage, with an innovative method employing Low-rank Adaptation (LoRA) technology to optimize LLMs.
Open Problems and Fundamental Limitations of Reinfrocement Learning from Human Feedback (RLHF). This paper surveys a range of open problems and issues with RLHF, which is critical to solve for LLM AIs.
More on the ‘Model Collapse’ failure issues from increasing Synthetic data ‘polluting’ LLM AI models, a new paper from researchers at Stanford and others titled “Self-Consuming Generative Models Go MAD”. This will be an ongoing issue for LLM AIs and an urgent area for solutions to mitigate the issues.
Additionally, also from this week,
A piece I did on why Google likely more than holds their own against OpenAI and Microsoft over the next year and beyond. Much like how Microsoft did against Netscape in the 1990’s. History likely rhymes.
And finally, how Amazon may be relatively under-appreciated amongst the Big Tech companies for their longer term AI opportunity. And not as well understood given the LLM AI framing around the ChatGPT phenomenon.
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).