Elon's Staggering Prediction and the AI Landscape Shift

Regulators are waking up, ChatGPT features are rolling out wide, and Elon is predicting massive demand for humanoid robots.

by @GregMushen and @KyleShannon

The Big Stuff

AI on Capitol Hill & EU Regulation

US Regulation

AI specialists, including Sam Altman, OpenAI's CEO and co-founder, presented their testimonies before Congress this week, discussing AI regulation. Altman acknowledged the potential benefits of AI, like enhancing various aspects of life and job opportunities, but also emphasized its risks. He suggested that before launching AI models, companies should scrutinize test results, and independent audits should be executed to guarantee the model's precision and dependability.

Altman also expressed apprehension about the detrimental impact of AI on elections, particularly through social media manipulation. He proposed establishing a new regulatory body to license AI models, develop safety standards to detect and control hazardous capabilities, and mandate expert audits to ascertain compliance with these standards. He underscored the significance of language inclusivity in AI, mentioning OpenAI's collaboration with Iceland to optimize models for the Icelandic language. See the full video here.

EU Regulation

The European Parliament is progressing with a potential first AI law in the West, known as the European AI Act, which includes regulations specifically targeting "foundation models" such as OpenAI's ChatGPT. The Act adopts a risk-based approach, with obligations determined by the level of risk associated with each AI system, and could enforce safety checks, data governance measures, and risk mitigations on developers. However, the proposed legislation, viewed by some as a necessary step towards ethical AI and by others as potentially inhibiting beneficial AI applications, is still under review and not yet law. (link)

ChatGPT - New Model Update, Plugin release, new UI, and iPhone App w/lock screen widget!

OpenAI had a massive release this week, with a new model update, release of Plugins and web browsing to all ChatGPT Plus subscribers, and an iOS app with a lock screen widget.

ChatGPT iOS AppThe ChatGPT iOS app is an exemplary piece of technology, offering a seamless and efficient ChatGPT experience on the move. It features a lock screen widget for immediate app access right from the lock screen. The app boasts of high-speed performance, allowing swift model-switching between versions 3.5 and 4. A right swipe on the app offers rapid access to searchable chat history, while a left swipe initiates a fresh conversation. Here's the link to the ChatGPT iOS app (link)

Plugins Examples:

• Chat/analyze/question with any PDF directly from ChatGPT (link)

• Do trip planning with Kayak (link)

• Order groceries with Instacart (link)

• Write emails in your writing style (link)

• Create tweets (link)

• Create Google Sheets (link)

• Create diagrams for you (link)

• Make restaurant reservations for you via OpenTable (link)

• Do undergrad quantum physics with the Wolfram plugin (link)

Web BrowsingWeb browsing allows you to use web searches as input to ChatGPT. In this example, I used ChatGPT web browsing to find all of the top restaurants in Seattle with dishes under $20, and write a quick summary of them in the style of Anthony Bourdain (link).

Elon Musk: Optimus will be the majority value of Tesla

Unleashing a new era in AI, humanoid robots like Tesla's Optimus are already making strides, with computer vision — the cornerstone technology that empowers them to navigate real-world scenarios. Elon Musk, in a recent shareholder meeting, forecasted a future where Optimus outpaces Tesla's auto division in revenue generation. Although it might seem a leap of faith, Musk's bold predictions have often materialized — a case in point being his successful foresight of electric cars dominating auto sales by 2030.

The race to market humanoid robots is heating up, but Tesla holds the upper hand with its Optimus already ambulatory. Here's why Tesla has the edge:

1. It can capitalize on its sophisticated computer vision system initially developed for self-driving cars, sidestepping reliance on LiDAR. This essentially means a smooth transition of AI models from self-driving to humanoid tasks, which don't demand as rigorous visual processing.

2. Tesla's vast experience in manufacturing hardware at scale gives it a lead many competitors lack, as they must build these capabilities from scratch.

Tesla's potential to be the first in the humanoid robot market could set off a data-feedback loop akin to their self-driving success story. The unveiling of autopilot drove up car demand, offering more data for Tesla to fine-tune its self-driving models, making them better and triggering even more demand. We could see a similar flywheel effect with humanoid robots, setting the stage for another success story for Tesla.

Med-PaLM 2 Outperforming Doctors?

Med-PaLM 2, a large language model from Google, achieved an 86.5 rating on the MedQA test, which is higher than 70% of physicians. While scoring better on a question set doesn't make Med-PaLM 2 better than doctors, it is a new milestone for medical AI performance. In addition, in 1066 consumer medical questions, physicians preferred Med-PaLM 2 answers to those produced by physicians on eight of nine axes pertaining to clinical utility. Link to paper (link).

More Big Stuff

• Unreal Engine 5.2 is out, and it. is. impressive. (link)

• Amazon's home robot, Astro flopped last year, but they are allegedly re-releasing it with an LLM. Just what we need. (link)

• Apple is releasing voice cloning software. They have designed this so people can still continue to use their voice after they've lost it (link)

• Meta announced a new chip, shared details on their 16k GPU Research SuperCluster (RSC) and announced their plan to build the next generation infrastructure backbone for AI.

• Poe makes Claude-instant-100k, an LLM with a 100k context window, available to subscribers (link). If you want to see what a 100k context window is capable of, watch this demo of someone summarizing a 3 hour long Lex Fridman podcast almost instantly (link)

• Claude-v1 is now the number two ranking model according to lmsys.org (link)

• Google and Adobe team up once again with Adobe Aero, a geospatial creator (link)

Smaller But Still Cool Things:

• DragGAN is released and lets you modify images. You must watch this demo (link)

• Machine learning model able to detect signs of Alzheimer’s across languages (link)

• Rahul, of Ligma Johnson fame, releases HoopsGPT.ai, which allows users to analyze NBA stats using natural language. (link).

• BT to cut up to 55,000 jobs by 2030 as fibre and AI arrive (link)

• OpenAI CEO Predicts AI Will Someday Give Birth To Twins, Their Names Will Be God And Satan (link)

Going Deeper

• The future of chemistry is language - GPT-4 can accomplish (link)

• How to use ChatGPT + Zapier to organize tax receipts (link)

• On-boarding your AI Intern - Ethan Mollick explains "There's a somewhat weird alien who wants to work for free for you. You should probably get started." (link)

• How to recreate hand signs in Midjourney, a tutorial (link)

• Decoding OpenAI Evals - Learn how to use the eval framework to evaluate models & prompts to optimise LLM systems for the best outputs. (link)

• MEGABYTE: Predicting Million-byte Sequences with Multiscale Transformers (link)

Tweets of the Week

Logan, doing the lord's work

Eye Candy

Ear Candy

AI is making big strides with generative music. One of the leading companies in this space is Cassette.ai. Their platform is able to generate music and vocals from text prompts. Check out the five examples in this thread. Next week, we will go deeper in the future of content.

Fly me to the walls let the sweat (NSFW: Explicit Lyrics)

Do you have 30 seconds for a quick survey to help us improve Everday AI?

We'd love your feedback! Click here.

Do you like what you're reading? Share it with a friend.