- Synthetic
- Posts
- Exciting New Rumors on OpenAI Strawberry and Orion (GPT-5?)
Exciting New Rumors on OpenAI Strawberry and Orion (GPT-5?)
Plus, How to Scale AI From Pilots to Production
Steve Brown
August 29, 2024
Subscribe to Synthetic
This week’s most interesting and relevant AI news and analysis
This Week in AI
Hoping to succeed where previous AI-first devices have failed (we’re talking about you, Humane, and Rabbit), Plaud has launched a new AI wearable, described as a ‘wearable AI memory capsule,’ to help you record, transcribe, summarize, and action things in your life. You click to record, and their software takes care of the rest. The NotePin has a 20-hour battery life, supports 59 languages, and costs $169. Advanced features cost $79 per year. Suggested applications include summarizing client conversations, capturing lecture notes, and creating mindmaps. 📝
Many people are worried about AI’s potential to destroy humanity. Their fears can often be unfounded (perhaps fuelled by watching a few too many sci-fi movies), though there are reasons to be cautious as we develop ever more powerful AI models. AI industry researchers talk about p(doom), the probability of AI causing a global apocalypse. Most insiders argue that AI’s benefits far outweigh its dangers. 🤖💥
Researchers are using AI to read ancient scrolls recovered from the ancient cities of Pompeii and Herculaneum, both destroyed by the eruption of Mt. Vesuvius in AD 79. They use machine vision, convolutional neural networks similar to the first ones invented by Yann LeCun to read handwritten postcodes in the 1990s, and more modern transformers (think ChatGPT) to restore the fragmented texts. The result is the Ithaca model, developed in partnership with Google DeepMind. 📜
Intel’s AI chips are competitive (not leading-edge, but really excellent performance per dollar), though the chip giant had yet to secure a cloud customer….Until now. IBM Cloud signed up to use Intel’s Gaudi 3 chip in its cloud platform and will support Gaudi 3 systems with its new Watsonx AI and data platform. 🧠
Quick Hits
Alexa Gets an Upgrade…For a Price - Fresh from the rumor mill: Amazon will upgrade Alexa to ‘Alexa Plus’ in October with new generative AI features costing up to $10 per month. The current (dumb) Alexa will be rebranded as ‘Classic Alexa.’
Nvidia Beats Q2 Estimates - Strong demand for AI computing and networking products drove Nvidia’s stellar results with over $30B in revenue (up 122% YoY). Their new Blackwell chips will ramp production in Q4, and demand for these leading-edge platforms far outstrips supply. Some watchers were disappointed, expecting even more, and the stock fell. 💰
Actors Union Gets AI Protections - SAG-AFTRA has persuaded the California senate to pass a bill that requires explicit consent from a performer to use their ‘digital replica.’
AI is Becoming Commoditized - With no clear leader in the GPT race, Google, OpenAI, and other vendors compete by slashing input and output token pricing. 📉
Google Rolls Out Gems and Imagen 3 - Google rolls out ‘Gems’ their answer to OpenAI’s GPTs, customized versions of Gemini focused on specific tasks or goals. They also updated their image generator to fix embarrassing shortcomings.
Midjourney Getting into Hardware - The AI image generator company is said to be making $200m/year. No details on the hardware yet, though their founder was behind Leap Motion (now UltraLeap), and Midjourney recently hired an Apple engineer who worked on the Vision Pro. Midjourney is currently working on video and 3D model generation. 👀
Video: The Man Taking on Tesla in the Race for Humanoid Robots
We don’t often share Peter Diamandis’ interviews as they include ads for his various projects, but we are making an exception here. If you’re fascinated to learn more about Figure, the leading humanoid robotics company (suck it, Elon! 😀), check out this interview with their founder and CEO, Brett Adcock. Adcock demonstrates Figure 02, their latest robot, and discusses plans to cost-reduce the product by an order of magnitude and go into mass production with Figure 03, which could cost around $20,000 and be on trial inside homes within three years. (1h 25m) 🦾
“Everybody will own a humanoid and labor will be optional.”
AI Tech and Innovation
Synthetic previously reported that OpenAI has been working on a new multi-step reasoning model codenamed ‘Strawberry,’ and formerly known as Q-Star. A new report in The Information (not linked because it’s behind a paywall) claims the new reasoning model not only solves math problems it hasn’t encountered before but also performs ‘deep research,’ solves word puzzles, and performs high-level tasks such as developing market strategies. Strawberry’s capabilities seem consistent with Level Two of OpenAI’s five-level framework for AI systems. Today’s leading models work by accurately ‘guessing’ the next token in a series, akin to ‘fast thinking,’ also known as System 1 thinking. Strawberry and other technologies like it are closer to slower System 2 thinking, which uses multiple steps to reason through problems. Rumors abound that Strawberry will be a component of OpenAI’s next frontier model, codenamed ‘Orion’ (GPT-5?), and that this model has already been previewed with national security officials in the federal government who are in charge of frontier AI oversight. Buckle up; the next step function leap in AI capability may not be too far away.
Google and Tel Aviv University researchers revealed a new model, GameNGen, that interactively simulates (read: generates) the classic 1993 first-person shooter, Doom. The model can crank out 20 frames per second using a single tensor processing unit (TPU). A researcher said, "GameNGen is a proof-of-concept for one part of a new paradigm where games are weights of a neural model, not lines of code." In the future, they wonder if games can be developed from still images and text descriptions rather than through traditional programming. 🕹️
The Potential here is absurd. Why write rules for software by hand when AI can just think every pixel for you?”
Doom showcased leading-edge 3D graphics back in 1993
Databricks CIO Naveen Zutshi recently shared his thoughts on how organizations can determine deployment readiness, scale models, and transition them from pilots into production. It starts with clean data, then involves A/B testing, performance metrics, continuous refinement, data governance, bias mitigation, clear AI policies, change management, and keeping a close eye on what’s coming next.
AI Insights
By 2030, U.S. data centers (more accurately, compute centers) could suck 400 terawatt hours of electricity, more than the total electricity production of the United Kingdom in 2022, according to the International Energy Agency. Installing a new transmission line can take a decade to plan, permit, and build. Upgrading the U.S. grid infrastructure is expected to cost hundreds of billions of dollars. ⚡️
Synthetic’s Take: New mega data centers, which may consume up to ten gigawatts of power each, will be colocated with nuclear power plants and connected directly to them. The Rural Electrification Act of 1936 brought electricity to isolated rural communities. The byproduct of that investment was increased readiness of the industrial base to fight the Nazis in the 1940s. The U.S. may need another electrification act to assure national competitiveness in the 2030s and beyond.
The UK government is appealing to tech companies to develop new AI tools to aid student assessment, mark exams, and develop curriculum, lesson plans, and teaching materials. The program aims to reduce teacher workload and free educators to focus more on face-to-face teaching.
It’s important to recognize that AI is a tool and not something that can replace the human expertise and interaction with students that can only be provided by our highly skilled and professional education staff.
Buy now, pay later firm Klarna plans to use artificial intelligence to eliminate half its employees in the coming years. The Swedish firm has already shrunk its workforce from 5000 to 3800 employees and plans to shed another 2000 by using AI to power its marketing and customer service departments. Klarna says it will take advantage of natural turnover and employee attrition rather than firing anyone.
Synthetic’s Take: Expect many more articles like this in the next two years. The Philippines hosts a gigantic outsourcing industry worth over $38 billion annually. Employees in the world’s call center capital are already working alongside AI copilots, unwittingly training the AI and putting themselves out of work a few years from now.
Toolkit for the Future
Create and sell online courses from your website with this top-rated platform. It’s easy to use with hundreds of beautiful templates and a powerful generative AI assistant that makes content generation a breeze. The AI will create lessons, exams, assessments, and more. Free to try.
News for humans, by humans.
Today's news.
Edited to be unbiased as humanly possible.
Every morning, we triple-check headlines, stories, and sources for bias.
All by hand with no algorithms.
Need to expand internationally but unsure where to start? Oyster makes it easy to find, hire, and retain local talent. They handle the details (country-specific labor laws, international tax laws, compliance, and global payroll) so you can focus on finding the right talent and ramping their impact.
Focus on the email that matters. Deep Clean automatically identifies old, unimportant emails so you can delete them. Snooze emails until a better time and easily track emails that need a response. Automatically filter unimportant emails into specific folders. It works across every device where you check your email. Get to inbox zero with award-winning SaneBox.
Need fresh talent that can harness the latest AI tools to increase the speed and volume of marketing content creation? Unlock value with fully-vetted, top marketing people with experience in your industry. The marketing-as-a-service model allows you to hire talent on demand, part -time or full-time, in 30-day blocks of time.
Try this easy-to-use, AI-powered alternative to DocuSign to write proposals, create quotes, get eSignatures on contracts, build online forms, collect payments, and much more. Great for sales, HR, legal, finance, and IT. Free 14-day trial.
Relieve pressure on your support team and serve customers better and faster with a support automation platform. AI answers 90% of queries and escalates the rest to your team. It automatically updates its knowledge database so it’ll never escalate that same issue again. Supports SMS, social, email, and phone communication.