About This Week’s Covers
This week’s cover incorporates two major stories from Google: The first is NotebookLM, a new Google product that transforms any document into a podcast hosted by an AI man and woman. The product is so compelling that people are calling it Google’s ‘ChatGPT moment.’ The second story is about a Google-developed robot that has learned to tie shoelaces. On the cover, the NotebookLM hosts are discussing the robot.
The image was created with Ideogram using the prompt “One male and one female podcast host are recording a podcast. Behind them is a large TV screen showing a video of a robotic arm tying a shoelace.” It was upscaled with Magnific and polished with Photoshop.
For fun I tried a challenge with Flux, MidJourney, and Ideogram, for that same prompt. As you can see, MidJourney is still horrible at complex composition. Flux did OK. Ideogram did the best.
The rest of the covers are single-shot Ideogram prompts, intentionally using minimal descriptions to test the tool’s capabilities for everyday users. Below are some of the stand-outs:
Executive Summaries
Google Turns Any Document Into a Convincing AI-Powered Podcast – Goes Viral
Google’s NotebookLM app can turn your research paper (or any document at all – no matter how boring) into a convincing, nuanced, and compelling podcast chat between two AI hosts. Powered by Google’s Gemini AI, the feature brings documents to life by having the hosts banter, summarize, and make insightful connections in a lively, conversational tone. The hosts use modern lingo and expressions— making the experience feel jarringly human. It’s so compelling (and actually fun) that it’s gone viral. Many are calling it Google’s Chat GPT moment.
Google | labsdotgoogle | theverge
Google DeepMind Teaches Robots Complex RealWorld Like Tying Shoelaces Using Simulations
Google DeepMind has launched two new AI systems, ALOHA Unleashed and DemoStart, to make robots better at handling tricky tasks. These systems focus on improving how robots use their “hands” to complete everyday activities, like tying shoelaces or tightening screws. A key part of this improvement is training robots using simulations, which are virtual environments where they can practice without needing physical setups. ALOHA Unleashed teaches robots to work with two arms, helping them master complex tasks faster by learning from human demonstrations. DemoStart, meanwhile, focuses on multi-fingered robotic hands. It uses simulations to practice and perfect skills like turning bolts or plugging in cords, needing 100 times fewer practice runs compared to real-world training.
Google | RemiCadene
World Labs Launches to Pioneer Spatial Intelligence AI
World Labs, a new AI company co-founded by AI luminary Fei-Fei Li and experts Justin Johnson, Christoph Lassner, and Ben Mildenhall, aims to revolutionize AI with spatial intelligence. The company focuses on developing Large World Models (LWMs) capable of perceiving and interacting with 3D environments, expanding beyond the current text and image-based AI capabilities. Spatial intelligence in AI is a big deal because it moves beyond 2D text and images, enabling machines to understand, navigate, and interact with the 3D world as humans do. With over $230 million in funding from Andreessen Horowitz, NEA, and NVIDIA Ventures, World Labs intends to build AI systems that create and understand virtual and real worlds. To me, this is the most important topic to follow in AI. Keep an eye out for the words segmentation, depth, AR/VR, embodiment, multimodality, Dr. Jim Fan.
Worldlabs | wired
UBS Unleashes Lightning-Fast AI Tool for M&A That Can Scan 300,000 Firms in 20 Seconds
Swiss Bank UBS has rolled out an AI tool that can analyze over 300,000 companies in under 20 seconds, completely transforming how the bank processes mergers and acquisitions (M&A). Dubbed an M&A “co-pilot,” the tool generates acquisition strategies, spots potential buyers, and even flags companies vulnerable to activist campaigns by assessing subtle shifts in management tone during presentations and Q&As. UBS also called out AI’s potential in streamlining legal and data management tasks during transactions, although the bank faces hurdles with compliance and data privacy when using AI for financial analysis. Tangentially, Citigroup recently projected that AI could automate over half of industry roles.
bloomberg
Salesforce Rolls Out Over 100 Industry-Specific AI Agents, Reduces Human Support by 90%
Salesforce launched Industries AI, a set of over 100 AI tools designed to solve problems specific to 15 different industries, like healthcare, education, and retail. The tools come ready-to-use, for businesses to automate key tasks like inventory management, patient services, or matching students to the right programs. A standout feature is Agentforce, AI-powered virtual assistants that reduce the need for human support by up to 90%. Salesforce also launched an AI Use Case Library, which offers resources and examples for getting started with the tools, so companies don’t have to spend time and money building AI models from scratch.
Jowyang | salesforce
Oracle Introduces Over 50 AI-Powered Assistants to Help Businesses
Oracle has launched over 50 new AI-powered tools designed to help businesses automate everyday tasks in its suite of cloud-based apps. Announced at the CloudWorld 2024 event, these tools (or “agents”) can handle tasks like shift planning, managing orders, and improving customer support—all without human intervention. This continues an evolution in AI’s use from simple chatbots to more advanced, helpful assistants that work behind the scenes, potentially displacing call centers and operational jobs.
techtarget
OpenAI Introduces o1, Its First ‘Reasoning’ Model to Tackle Complex Problems
OpenAI has launched o1, the first in a series of “reasoning” models designed to handle more complex tasks than its previous models, including coding and multistep problem-solving. Available now for ChatGPT Plus and Team users, o1 aims to enhance capabilities like math and programming, outperforming GPT-4o in these areas but at a higher cost and slower speed. A smaller version, o1-mini, is also being released. The model uses a new training method based on reinforcement learning and “chain of thought” processes, mimicking human-like step-by-step reasoning. While o1 shows improvements in problem-solving accuracy, it remains limited in areas like factual knowledge and web browsing. OpenAI positions this as a preview, hinting at future models with even greater capabilities as it pursues human-like intelligence.
Openai | theverge
Three Initial Reactions to OpenAI’s o1 Model:
Fun things to do with your limited o1-preview uses that can show you the power and limitations:
🤖Give it an RFP and ask it to just do the work
🤖Give it an academic paper & ask it to offer strategies for replication
🤖Ask it to create an entrepreneurial product that it can build
emollick
ChatGPT o1 preview + mini Wrote My PhD Code in 1 Hour*—What Took Me ~1 Year – YouTube
youtube
It’s so over. OpenAI’s 01 model got 25 out of 35 IQ questions correct, far above what most humans get. And, these questions were never part of 01’s training data, as they have never been posted to the public internet.
rohanpaul_ai
Roblox Unveils Generative AI for Building 3D Environments
Once you know that VR/3D worlds are the key to how real world robots will be trained, this will change your view of “a kids gaming company”. Roblox deserves space next to Jim Fan’s NVIDIA robotics lab. Roblox is launching a generative AI tool designed to simplify the creation of 3D game environments through text prompts, enhancing accessibility for developers with minimal design skills. With commands like “Generate a race track in the desert,” users can instantly create and modify game scenes, adjusting elements such as time of day or terrain types. This AI system “tokenizes” 3D blocks, using a method similar to large language models, but it faces challenges due to the limited availability of 3D data. To mitigate inconsistencies, Roblox employs a secondary AI that uses 2D data to validate and refine the 3D outputs. The company aims to open-source the foundation model to foster innovation and provide solo developers with advanced tools, potentially revolutionizing game design on the platform (and if you ask me, evolve into a bigger company than just games).
Technologyreview
Helpful Visuals and Charts: Week Ending 09/13/2024
“Finally got around to really dig deeper into LivePortrait.
https://x.com/8bit_e/status/1833250941006491876
Our latest advances in robot dexterity – Google DeepMind
https://deepmind.google/discover/blog/advances-in-robot-dexterity
A.I. Can Now Create Lifelike Videos. Can You Tell What’s Real? – The New York Times
https://www.nytimes.com/interactive/2024/09/09/technology/ai-video-deepfake-runway-kling-quiz.html
“Yo @runwayml ‘s new video to video is crazy good! Did a few quick test.
Top 87 Links of The Week – Organized by Category
Agents and Copilots
“wow. Apple might just have fixed Siri. and beat OpenAI to the first AI phone. and commoditized OpenAI with Google. and casually dropped a video understanding model. incredibly well executed. (see @smol_ai writeup below for deltas from WWDC)
“@Smol_AI notable reveals from today’s iphone 16 event, especially Apple Visual Intelligence: – Mail and Notifications will show summaries instead of str[:x] – Siri now knows iPhone, becomes the ultimate manual on how to use the increasingly complicated iOS 18 and can read your texts
Salesforce Launches Industries AI: 100+ Out-of-the-Box AI Capabilities that Tackle Industry-Specific Pain Points Across 15 Industries
“Nvidia $NVDA CEO Jensen Huang just “I think the days of every line of code being written by software engineers is completely over”” / X
“NEW: Plot Your Path: The 2024 AI Agent Ecosystem Map
“We do not realize how crazy this is. You can take any complex document and turn it into a bunch of interactive graphs in seconds. This is beyond magic. Nobody would have thought this was possible a couple of years ago. Check this out:
Agentforce: Customizable AI Agent Builder | Salesforce US
Amazon
Amazon’s Audible to Create AI Voice Replicas of Select Audiobook Narrators – Bloomberg
Amazon is allowing Audible narrators to clone themselves with AI – The Verge
Apple
Apple will start selling AirPods with built-in hearing aids
“Let’s talk about augmented audio! What if the same tech that empowers creators to sculpt sound could also help people hear better? 🎧 It’s TRULY cool is how Apple’s leveraging its “advanced computational audio” to both enhance creative tools AND improve accessibility. The
“JUST IN: Apple just unveiled the iPhone 16, and more details to Apple Intelligence The next generation of iPhones is built to use AI from the “ground up” All the details: – Comes in new colors with aluminum finish – Has Custom Action Buttons directly on the iPhone – A new
Artificial General Intelligence (AGI) News
“Researchers found that AI systems can improve reasoning just by ‘re-reading’ a question This simple and effective prompting technique is oddly similar to how humans think Makes you wonder how many simple prompt techniques we’re overlooking
“This is ABSOLUTELY incredible to see – AI outperforms humans in research ideation novelty🤯 **Key Insights from this Paper 💡**: • LLM-generated ideas are judged as more novel than human expert ideas • AI ideas may be slightly less feasible than human ideas • LLMs cannot
“Automating AI research is exciting! But can LLMs actually produce novel, expert-level research ideas? After a year-long study, we obtained the first statistically significant conclusion: LLM-generated ideas are more novel than ideas written by expert human researchers.
“As LLMs get smarter, evals need to get harder. OpenAI’s o1 has already maxed out most major benchmarks. Scale is partnering with CAIS to launch Humanity’s Last Exam: the toughest open-source benchmark for LLMs. We’re putting up $500K in prizes for the best questions. (read on)
Sergey Brin Is Back. Is Google Working on a ‘God’ AI Model? | Entrepreneur
OpenAI O1-Mini | Hacker News
“OpenAI o1 — our first model trained with reinforcement learning to think hard about problems before answering. Extremely proud of the team! This is a new paradigm with vast opportunity. This is evident quantitatively (eg reasoning metrics are already a step function improved)” / X
“I’ve had access to @OpenAI’s o1 for several weeks. My advice on using it: 1. Don’t think of it like a traditional chat model. Frame o1 in your mind as a really smart friend you’re going to send a DM to solve a problem. She’ll answer back with a very well thought out explanation” / X
“🎉Congrats to @OpenAI for releasing o1: – Economics: @tylercowen asked o1 basically to write a college essay – Genetics: @catbrownstein asked o1 to help her reason through “n of 1” cases – medical cases that nobody has ever seen – Physics: @mariokrenn6240 used o1 to draft and
OpenAI releases new o1 reasoning model – The Verge
“Summary of what we have learned during AMA hour with the OpenAI o1 team today Model Names and Reasoning Paradigm – OpenAI o1 is named to represent a new level of AI capability; the counter is reset to 1 – “Preview” indicates it’s an early version of the full model – “Mini”” / X
Audio News
Amazon is allowing Audible narrators to clone themselves with AI – The Verge
Apple will start selling AirPods with built-in hearing aids
“Let’s talk about augmented audio! What if the same tech that empowers creators to sculpt sound could also help people hear better? 🎧 It’s TRULY cool is how Apple’s leveraging its “advanced computational audio” to both enhance creative tools AND improve accessibility. The
“ok, @suno_ai_ just released a new AI music feature called “Covers” and it’s pure magic It works with your voice. You sing into Suno, give it a prompt, and it transforms your vocals into full songs. It’s like a musical collab with the AI Link to try below
Chips, Hardware, and Infrastructure
“📣 Announcing the world’s fastest AI platform SambaNova Cloud runs Llama 3.1 405B @ 132t/s at full precision ✅ Llama 3.1 405B @ 132 tokens/sec ✅ Llama 3.1 70B @ 570 tokens/sec ✅ 10X Faster Inference than GPUs Start developing #FastAI ➡️
Ethics/Legal/Security
This AI chatbot got conspiracy theorists to question their convictions
This Chatbot Pulls People Away From Conspiracy Theories – The New York Times
Are novelists who worry about the rise of AI really ‘classist and ableist’? | Arwa Mahdawi | The Guardian
Reflection 70B’s performance questioned, accused of ‘fraud’ | VentureBeat
“READ THEIR LIPS WITH AI You can now upload a video of any speaker and identify inaudible speech using our model. Find out what people like Blake Lively, Taylor Swift, Lebron James, and more are saying when microphones aren’t around and discover the inaudible! Link Below
Readout of White House Roundtable on U.S. Leadership in AI Infrastructure | The White House
AI experts ready ‘Humanity’s Last Exam’ to stump powerful tech | Reuters
“With an IQ of 120, AI is now officially more intelligent than most humans! However, they are still extremely lacking in perception and understanding of the environment
Sergey Brin Is Back. Is Google Working on a ‘God’ AI Model? | Entrepreneur
“AI will eat most SAAS 💀 “This news from @Klarna should have every enterprise SaaS company shaking in their boots. If an internal team using AI can replicate 20+ years of work and customization from @salesforce and @Workday, to the extent the company doesn’t feel the need to
BP extends use of AI in five-year deal with spy tech firm Palantir | BP | The Guardian
Mr ChatGPT and other AI power players are going to the White House to discuss AI’s massive thirst for energy | CNN Business
Deepfake Sex Videos in South Korea Seen as Old Misogyny With New Tech – The New York Times
Canada working on military AI, defence minister says | CTV News
China’s AI models lag their U.S. counterparts by 6 to 9 months, says former head of Google China
“Since arriving in China, meetings founders and VCs have left me with a very dim view of the current state affairs. Things are so bad. In 2018, 51,302 new startups were founded. Last year, that number was down to 1,202. It took a while for that number to sink in.
US judge runs ‘mini-experiment’ with AI to help decide case | Reuters
“First Criminal Case Involving AI Inflated Music Streaming. U.S. Attorney’s Office released the document of the case. Here’s how he executed it, based on the information provided 👇 1. Purchase thousands of fake email addresses. 2. Use these fake email addresses to create and
“Omg. People are so quick with it 😂 Completely AI generated news segment on “cats in springfield”by @KUDORadio
Sergey Brin says he’s working on AI at Google ‘pretty much every day’ | TechCrunch
Here’s how 7 news audience directors are thinking about Google’s AI Overviews | Nieman Journalism Lab
Illuminate | Learn Your Way (as usual Google makes multiple products without connecting them… NotebookLM v. Illuminate? Nobody knows)….
Images
“Midjourney Office hours: They’re working on a 3d system that will allow you to move into Midjourney images. Not polygons, not NeRF, not Gaussians. But a new NeRF-like format, with a team lead by someone who played an important part in the rollout of NeRF technology. Sorry
“Midjourney has a hardware team Back in April 2018, Midjourney founder @DavidSHolz published this article unveiling Project North Star, a full augmented reality platform which he built and open sourced at his previous company Leap Motion Reading this today hits different https://x.com/nickfloats/status/1833131470455468508
International
“New DeepSeek-V2.5 model scores 89 on HumanEval, surpassing GPT-4-Turbo, Opus, Llama 3.1 in coding
Alibaba’s Taobao launches AI-powered English version in Singapore
OpenAI News
“As LLMs get smarter, evals need to get harder. OpenAI’s o1 has already maxed out most major benchmarks. Scale is partnering with CAIS to launch Humanity’s Last Exam: the toughest open-source benchmark for LLMs. We’re putting up $500K in prizes for the best questions. (read on)
Amir Efrati on X: “new: ChatGPT is CONSERVATIVELY generating more than $225 million per month right now. that’s some kind of growth, folks. https://t.co/2perQfY2LG” / X – https://x.com/amir/status/1834347880251052203
Tibor Blaho on X: “Summary of what we have learned during AMA hour with the OpenAI o1 team today Model Names and Reasoning Paradigm – OpenAI o1 is named to represent a new level of AI capability; the counter is reset to 1 – “Preview” indicates it’s an early version of the full model – “Mini”” / X – https://x.com/btibor91/status/1834686946846597281
Rohan Paul on X: “The leap of @OpenAI ‘s 01 models on PhD-level tasks are just MASSIVE, specially on PhD-level Physics. GPQA diamond benchmark (PhD-level physics, chemistry, biology): – o1 first model to surpass human PhD expert performance – Outperforms GPT-4o by a wide margin —- 🧠 AIME https://t.co/ghph1OZ5Hq” / X – https://x.com/rohanpaul_ai/status/1834294432214159439
OpenAI O1-Mini | Hacker News
ChatGPT o1 preview + mini Wrote My PhD Code in 1 Hour*—What Took Me ~1 Year – YouTube
“OpenAI o1 — our first model trained with reinforcement learning to think hard about problems before answering. Extremely proud of the team! This is a new paradigm with vast opportunity. This is evident quantitatively (eg reasoning metrics are already a step function improved)” / X
“I’ve had access to @OpenAI’s o1 for several weeks. My advice on using it: 1. Don’t think of it like a traditional chat model. Frame o1 in your mind as a really smart friend you’re going to send a DM to solve a problem. She’ll answer back with a very well thought out explanation” / X
“Fun things to do with your limited o1-preview uses that can show you the power and limitations: 🤖Give it an RFP and ask it to just do the work 🤖Give it an academic paper & ask it to offer strategies for replication 🤖Ask it to create an entrepreneurial product that it can build
“🎉Congrats to @OpenAI for releasing o1: – Economics: @tylercowen asked o1 basically to write a college essay – Genetics: @catbrownstein asked o1 to help her reason through “n of 1” cases – medical cases that nobody has ever seen – Physics: @mariokrenn6240 used o1 to draft and
“Summary of what we have learned during AMA hour with the OpenAI o1 team today Model Names and Reasoning Paradigm – OpenAI o1 is named to represent a new level of AI capability; the counter is reset to 1 – “Preview” indicates it’s an early version of the full model – “Mini”” / X
Reasoning – OpenAI API
Open Source
“📣 Announcing the world’s fastest AI platform SambaNova Cloud runs Llama 3.1 405B @ 132t/s at full precision ✅ Llama 3.1 405B @ 132 tokens/sec ✅ Llama 3.1 70B @ 570 tokens/sec ✅ 10X Faster Inference than GPUs Start developing #FastAI ➡️
“I got fooled by the Reflection 70B announcement. tl;dr – the model performs very badly” / X
“@mattshumer_ @ArtificialAnlys Final report on Reflection-70B: after investigating, I do not believe the a model that achieved the claimed benchmarks ever existed. It’ very unclear to me where those numbers came from, and I hope that Sahil/Matt will shed more light how this happened.” / X
“I am working with @mattshumer_ to get to the bottom of what happened with Reflection. He is providing access to all serving code and weights with the goal of replicating the strong reasoning performance @ArtificialAnlys was able to see over the weekend. I will report back.” / X
Reflection 70B’s performance questioned, accused of ‘fraud’ | VentureBeat
Mistral releases Pixtral 12B, its first multimodal model | TechCrunch
“Mistral released Pixtral 12B Vision Language Model 🔥 Some notes on the release: 1. Text backbone: Mistral Nemo 12B 2. Vision Adapter: 400M 3. Uses GeLU (for vision adapter) & 2D RoPE (for vision encoder) 4. Larger vocabulary – 131,072 5. Three new special tokens – `img`,
“We dropped a new model – Pixtral 12B, our first-ever multimodal model. Enjoy! 🥰🎉” / X
Podcasts/YouTube/Op-Eds
The College Dropout Who Invested Billions to Cozy Up With Elon Musk – WSJ
Oprah just had an AI special with Sam Altman and Bill Gates — here are the highlights | TechCrunch
Publishing News
“Omg. People are so quick with it 😂 Completely AI generated news segment on “cats in springfield”by @KUDORadio
Robotics and Embodiment News
Inside Google’s 7-Year Mission to Give AI a Robot Body | WIRED
Science and Medicine News
Apple will start selling AirPods with built-in hearing aids
“Let’s talk about augmented audio! What if the same tech that empowers creators to sculpt sound could also help people hear better? 🎧 It’s TRULY cool is how Apple’s leveraging its “advanced computational audio” to both enhance creative tools AND improve accessibility. The
Video News
Adobe previews generative AI for videos
Bringing generative AI to video with Adobe Firefly Video Model | Adobe Blog
“Gen-3 Alpha Video to Video is now available on web for all paid plans. Video to Video represents a new control mechanism for precise movement, expressiveness and intent within generations. To use Video to Video, simply upload your input video, prompt in any aesthetic direction
“Omg. People are so quick with it 😂 Completely AI generated news segment on “cats in springfield”by @KUDORadio
A.I. Can Now Create Lifelike Videos. Can You Tell What’s Real? – The New York Timeshttps://www.nytimes.com/interactive/2024/09/09/technology/ai-video-deepfake-runway-kling-quiz.html
Leave a Reply