- AI Weekly Wrap-Up
- Posts
- 8-30-2023
8-30-2023
Keith Collins MD & Addie Politi
August 30, 2023
Top Story
OpenAI Releases ChatGPT for Enterprise
On Monday, OpenAI announced the launch of ChatGPT for Enterprise, the business-friendly, more secure version of the world’s most famous chatbot. The new service has a lot to offer:
Stronger Security - user data will not be used to train the model, eliminating the chance of leaks of proprietary information
Based on GPT-4, OpenAI’s most advanced model - GPT-4 is OpenAI’s flagship, and still generally considered the most capable commercially available LLM, despite recent stiff competition.
Unlimited access to Advanced Data Analysis - (the app formerly known as Code Interpreter) - an extremely capable data analysis tool
Much larger context window - up to 32,000 tokens, or about 25,000 words, so that prompts can be guided by voluminous relevant information, to put the query “in context”
Administrative tools - including an administrative console, Single Sign-On, domain verification, and a dashboard for usage statistics
OpenAI is clearly stepping on the accelerator in its efforts to penetrate the enterprise market, one reason it now has a revenue run rate of $1 billion annually - 30+ times last year, and 5x projections of just a few months ago.
Clash of the Titans
Google Deepmind CEO says Gemini will eclipse ChatGPT
Deepmind CEO Demis Hassabis boasts that his team at Google are hatching a next-gen LLM code-named Genesis, that is multi-modal (manipulates text, audio, and graphics) and uses the advanced techniques that allowed its AlphaGo model to beat the best Go players in the world. The result, he promises, will blow ChatGPT out of the water. Big talk. Hype? Truth? Time will tell.
Overseas AI workers toil in “digital sweatshops”
Large Language Models like ChatGPT have benefited enormously from fine tuning by feedback from thousands of human raters, who do the tedious work of ranking the helpfulness, accuracy, and safety of countless chatbot responses. These ratings are fed back into the model, to improve future responses. Most of these workers are recruited from overseas, and they sometimes work in spartan environments for shockingly low rates of pay.
One of the major players in the recruitment and management of raters is ScaleAI, a multibillion dollar company that is a preferred partner of OpenAI. The Washington Post has printed an expose’ of some of ScaleAI's seamier practices, including poor working conditions and just not paying the agreed-on rate.
Amazon will bring AI video tools to Thursday Night Football
For its second season as the official broadcaster of Thursday Night Football, Amazon is going all-in on AI-powered graphics and video. Amazon has trained AI video models to understand the game of football, and to automatically generate explanatory graphics as the game unfolds. The goal is to get viewers more engaged in the strategy and tactics of the game by analyzing critical moments in real time. Now even your idiot brother-in-law can follow the game instead of asking annoying questions. At least, that’s the hope.
Fun News
AI gives a paralyzed woman her voice back
A woman who suffered a brainstem stroke at age 30 which left her severely paralyzed, now can speak through a digital avatar controlled by her thoughts. Her brain activity is captured by a digital brain implant which is connected to an AI that translates the brain signals into fluent, natural-sounding speech. Watch the videos - they are amazing.
Using AI to talk to whales
Scientists using hydrophones dropped into the water have deciphered the chirps and clicks of humpbacked whales. Now they are using AI to synthesize artificial chirps and talk back to the whales.
Elon demos a self-driving car guided by pure AI
Progress toward true autonomous vehicles has been frustratingly slow, since the real world has countless edge cases and exceptions to the rule. Tesla CEO Elon Musk used his X social media platform last week to demonstrate what he claims is a car whose guidance system has not a single line of human-generated code. An AI model purportedly learned to drive by watching countless hours of video captured by the 8 cameras that come standard on each of the 4 million existing Tesla vehicles. This end-to-end AI system apparently outperforms the best of the current guidance systems. Watch Elon ride in his AI-only autonomous car.
Video special effects on your iPhone, made easy
The link below takes you to what is reported to be a demonstration of a new video effects app that performs CGI in real time, on your iPhone. Huge, if true.
Regulation and Politics
China’s People’s Liberation Army is gunning to have AI in the military
China’s PLA wants to integrate AI into a range of military equipment, but political, economic and scientific challenges are standing in their way. Josh Baughman, an analyst at China Aerospace Studies Institute, said “China understands the need to be a first mover in generative AI on the battlefield to firmly grasp the strategic initiative of intelligent warfare.” This urgency to integrate is likely spurred in part by the current tensions between the US and China. US policymakers are understandably wary of China’s moves into AI on the battlefield, and this is just one of the drivers of the US government’s ban on exports of high tech AI chips to China.
Economics
UAE joins with Mastercard to accelerate adoption of AI
The UAE’s AI Digital Economy and Remote Work Applications Office and Mastercard jointly signed a memorandum of “Understanding to increase artificial intelligence capabilities and readiness in the region.” H.E. Omar Sultan AL Olama, Minister of State for AI, highlighted the importance of this “bilateral cooperation” to continue to innovate (aka no hostilities please US aka we are not a threat!). This agreement is a step forward for the UAE’s AI goals for 2031, which include becoming a worldwide hub for AI technology.
Goldman Sachs picks 50 companies ripe for long-term AI gains
After the recent AI market boom which has resulted in Nvidia joining the elite group of trillion-dollar companies, Goldman Sachs has released a report on the 50 companies that they feel are best poised to profit from the burgeoning capabilities of AI. They are not all whom you’d think. Ryan Hammond and David Kostin, the bank’s analysts, selected a group of firms with the “largest potential long-term EPS boost via AI’s ability to create a more productive workforce, low labor costs, or both.” However, the projections come with “plenty of caveats.”
Uber Eats is becoming your personal food advisor.
Uber Eats’ new AI chatbot is in its early development phase, but it promises to offer food recommendations while keeping in mind users’ budgets and food preferences. Trying to push DoorDash and Instacart out of the AI race as they also start to incorporate AI into their platforms, Uber hopes that an AI-food-recommender might help them take the lead in the food-delivery market.
Hawaiian Fires sparked controversy over the electric grid infrastructure. Now they are turning to AI for solutions while insurance companies are drowning in red ink.
After the disastrous fires spread through Maui because Hawaiian Electric failed to de-energize the power lines, major electric companies are rethinking their grid infrastructure. Something that has come to light in the recent lawsuit is that electric utilities in the western United States have ignited “hundreds of fires.” In response, a number of utility companies have begun to modernize their grids. Emerging forms of AI are being incorporated into these upgrades, to more efficiently transfer energy. This new tech might prevent another $4 billion to $6 billion of property damage losses that insurance companies are currently struggling to cover.
AI in Medicine
ChatGPT shows “impressive” accuracy in medical decisions
Researchers from the prestigious Massachusetts General Brigham health system in Boston, Massachusetts queried ChatGPT with a series of standardized complex clinical scenarios, to judge its clinical decision-making skills. Overall, the chatbot was deemed to perform at the level of a medical intern or resident, showing promise for this new technology to become an important aid in patient care.
AI analyzes eye scans to detect Parkinson’s Disease early
UK researchers have trained an AI system to detect signs of incipient Parkison’s Disease in eye scans, allowing diagnosis as much as 7 years earlier than prior methods. This enables early intervention to mitigate and delay progression of symptoms.
Microsoft partners with Epic to bring AI to EMRs
Microsoft has joined with Epic, one of the leading vendors of electronic medical records (EMR) software to integrate AI into the system. Early goals for the partnership include generating patient visit notes automatically from an audio record of the interaction, as well as optimizing medical coding and billing for the encounter.
That's a wrap! More news next week.