- Synthetic
- Posts
- OpenAI and Google Accidentally Leak Exciting New Models
OpenAI and Google Accidentally Leak Exciting New Models
Plus, What A Second Trump Presidency Could Mean for AI
Steve Brown
November 07, 2024
In partnership with
Subscribe to Synthetic
The week’s most interesting and relevant AI news and analysis
This Week in AI
While OpenAI released a preview of its new o1 reasoning AI model in September, the full version of o1 was not expected until December, potentially as part of a broader model release codenamed ‘Orion.’ Users who briefly gained access to the model were impressed by its capabilities. The model was able to solve complex math problems and a classic image puzzle (shown below.) The model also analyzed a picture of a recent SpaceX launch and shared a full ‘chain of thought’ of its work. Reasoners are a vital step in advancing AI, adding System 2 (slow, deliberative) thinking to the System 1 (fast, reflexive) thinking used in most existing large language models. Why is o1 such a big deal? Reasoning (using slower, more reflective ‘thinking’ that runs at inference time rather than being intelligence built into the model at training time) is an essential foundation for agentic AI, the new major frontier in AI research. Synthetic predicts that 2025 will be the year of autonomous AI agents, and models like o1 will underpin many exciting new agents next year.
Oopsy no.2 this week came when Google’s new Jarvis AI prototype became available on the Chrome Store for a few hours before being removed. Google describes Jarvis as “a helpful companion that surfs the web for you.” The new model will control a web browser to book flights, do online research, make purchases, and “automate every day, web-based tasks.’ Jarvis is rumored to launch in December 2024, perhaps with new branding (Project Jarvis is a codename). Once again, Google finds itself in catch-up. Despite Jarvis being rumored to be in development for years, Anthropic beat Google to market with its model, rather uncreatively named ‘Computer Use.’
Quick Hits
Meta Supplies US National Security and Defense Contractors - Meta has announced that it will let national security agencies and defense companies use Llama, its open-source model. However, it prohibits use for “military, warfare, nuclear industries, and espionage.” Let’s see how that works out. 🪖
Apple Will Upgrade Apple Intelligence Servers With M4 Chip - Starting next year, it is rumored that Apple will upgrade the private cloud servers it uses for most of its Apple Intelligence services to use its latest M4 chip, replacing the M2 Ultra chips used today. M4 brings significant boosts in AI performance. 📈
Google Claims World First As AI Finds Security Vulnerability - Google’s Project Zero and DeepMind created Big Sleep, an AI agent that was able to spot a zero-day, exploitable memory-safety issue in SQLite, an open-source database engine used in many real-world software applications. 🚨
Videos: Building an AI-Powered Robotic Arm 🦾
Atom Limbs uses AI and robotics to build affordable, mind-controlled robotic arms for amputees. Their latest models use AI models to determine the wearer’s intent and help them to perform tasks naturally. Check out their impressive innovation.
(15m 45s)
AI Tech and Innovation
Source: Tim Urban, WaitWhatWhy, 2015
The idea of an “intelligence explosion,” rapidly self-improving AI that quickly becomes superintelligent, is not new. As ex-OpenAI employee Leopold Aschenbrenner outlined in his seminal white paper, Situational Awareness, one of the goals of AI researchers is to automate the process of AI research, thereby leading to an exponential ramp in AI's capabilities. Agentic AIs, built on the latest reasoning AI models (see the lead article on OpenAI’s o1), are a first step. While the mere idea of automating AI research may seem fanciful, AI is already used to conduct AI research today. Earlier this year, Sakana AI published a paper on its “AI Scientist.” Sakana AI was founded by former lead researchers at Google, including one who co-invented the transformer that underpins all the leading LLMs used today. Sakana’s AI Scientist has ‘read’ existing literature, generated novel research ideas, designed experiments to test these ideas, carried out experiments, and written research papers to report its findings. The AI Scientist then reviews its own work, critiquing it as would happen during peer review. It has published several dozen research papers, including this one, and does so autonomously without human input. Its performance is judged at the level of a junior machine learning researcher, so human AI researchers still have a job….for now. It’s worth noting that this AI Scientist model was built before the release of OpenAI’s o1 reasoning technology, so we should pay attention not to what it can achieve today, but what systems like this might soon be capable of.
A colorable argument could be made that AI will sooner learn to automate the job of an AI researcher than to automate the job of a plumber.
NVIDIA has long been a champion for robotics, particularly with its Project GROOT initiative. This week, the most valuable company in the world released a suite of new tools, workflows, models, learning frameworks, and capabilities to accelerate the development of AI-enabled robots and make it easier for robotics companies to build world models, representations of the world and the way it works so robots can accurately predict how objects and environments will respond to a robot’s actions.
AI Insights
De-aged Tom Hanks and Robin Wright, in a scene from “Here”
Director Robert Zemeckis is unlikely to make a more perfect film than “Back to the Future,” though he keeps trying. His latest movie reunites him with Tom Hanks and Robin Wright, the stars of Forrest Gump, another Zemeckis gem. Metaphysic, a special effects company, built generative AI software that uses face-swapping and de-aging effects to show the actors at whatever age was required by the scene in real time. Metaphysics technology was also used to reanimate Ian Holm in the recent Alien: Romulus. 👽
Tech leaders from Amazon, Apple, Google, and Meta expressed optimism that the incoming Trump administration will cut red tape to fuel innovation. Cryptocurrency prices spiked as traders expect crypto regulation to be relaxed. Trump is expected to repeal President Biden’s 2023 executive order on AI safety and responsible development, although it is unclear what regulation (if any) will replace it. Market observers expect state-level AI regulation may emerge to fill the void. AI developers could face a more challenging landscape as they need to comply with a patchwork of disjointed and perhaps contradictory rules and regulations.
Toolkit for the Future
Hirelogic in action
Invite the HireLogic AI to your next phone, video, or in-person interview. It analyzes the conversation, provides specific insights, and highlights the interviewee’s strengths, skills, and any potential concerns. It creates a full interview transcript so you can focus on the conversation. And it’s free to try.
Need a personal assistant? We do too, that’s why we use AI.
Ready to embrace a new era of task delegation?
HubSpot’s highly anticipated AI Task Delegation Playbook is your key to supercharging your productivity and saving precious time.
Learn how to integrate AI into your own processes, allowing you to optimize your time and resources, while maximizing your output with ease.
Notion helps you focus on the things that actually matter. Organize information, thoughts, to-do lists, calendars, articles, tasks, notes, and team discussions in one well-organized space. Use Notion AI to create content, translate languages, and ask your notes questions.
Make a logo in minutes—no design skills required! Auto-create a brand kit with fonts and colors to complement your new logo and create a brand identity. Instantly create branded marketing materials, then use over 300 templates to customize branded business cards, social profiles, email signatures, and more.
Melio is a free and easy-to-use payment solution that helps U.S. small businesses pay any expense by bank transfer or credit card - even where cards are not accepted. No setup or monthly fees.
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 filters unimportant emails into specific folders. It works across every device where you check your email. Get to inbox zero with the award-winning SaneBox.
Recording, editing, and refining audio demands more time than most entrepreneurs can spare. Generate voices, high-quality music, and sound effects efficiently and cost-effectively with a single click. Create multilingual voice tracks in up to 29 languages to reach new markets. Try it for free.
Find phone numbers, emails, and company org charts to reach the people you need to. Fill information gaps in your contacts, auto-generate follow-up emails, and identify new target markets. Get 50 free credits and discover why over 500,000 companies use Seamless to grow their business.