Extropic AI Surpasses Quantum

PLUS: Mercedes Pilots Humanoid Robots, Perplexity Challenges Google's Dominance and more.

Sponsored by


AI brews beer and your big ideas

What’s your biggest business challenge? Don’t worry about wording it perfectly or describing it just right. Brain dump your description into AE Studio’s new tool and AI will help you solve that work puzzle.

Describe your challenge in three quick questions. Then AI churns out solutions customized to you.

AE Studio exists to solve business problems. They build great products and create custom software, AI and BCI solutions. And they once brewed beer by training AI to instruct a brewmeister and then to market the result. The beer sold out – true story.

Beyond beer, AE Studio’s data scientists, designers and developers have done even more impressive things working 1:1 with founders and executives. They’re a great match for leaders wanting to incorporate AI and just generally deliver outstanding products built with the latest tools and tech.

If you’re done guessing how to solve work problems or have a crazy idea in your back pocket to test out, ask AI Ideas by AE Studio for free solutions, right now.

Extropic Beff Jezos on AGI Computing | Better than Quantum Computing? | Accelerate or Die

Jeff Bezos, born in '64 and a big-shot CEO, has thrown his weight behind Perplexity AI, a slick AI search engine that smartly summarizes web info. Users have been digging it for various queries, showing its versatility. 

On the tech side, we're hitting a wall where making stuff smaller just won't cut it anymore. We're talking about microchips and their transistors that are now so tiny, they're just a few atoms thick. Gom Verdo, stepping up with Extropic, suggests ditching traditional quantum computing for something even cooler, involving superconductors and what they call "the most efficient neurons." They're onto something big, promising a quantum leap in computing efficiency for AI.

Softbank Considers Investing in Mistral AI

SoftBank eyes investing in Mistral AI, a French startup challenging US AI dominance, potentially valuing it over $2 billion. Mistral, founded by Meta and Google veterans, gained attention with a $415 million funding last year. Microsoft recently invested $16 million, integrating Mistral's model with Azure services. European Commission plans scrutiny amid Big Tech-AI partnerships. 

SoftBank's Masayoshi Son predicts AGI surpassing human intellect soon. Meanwhile, AI's rising popularity, like Mistral 7b, faces concerns over energy consumption and environmental impact. Stay updated with PYMNTS' AI coverage. Subscription recommended for daily AI insights.

Apple researchers achieve breakthroughs in multimodal AI as company ramps up investments

Apple researchers have advanced multimodal AI by training large language models on text and images, enhancing AI systems for future products. The MM1 research, detailed in a paper on arxiv.org, highlights the importance of diverse training data for optimal performance in tasks like image captioning and natural language inference. 

Notably, Apple's increased AI investments aim to catch up with rivals like Google and Microsoft. Projects like the "Ajax" framework and "Apple GPT" chatbot could revolutionize Siri and other services. With the AI arms race intensifying, Apple's secretive advancements hint at a future of pervasive AI integration, shaping the digital landscape.

Mercedes begins piloting Apptronik humanoid robots

Mercedes-Benz partners with Apptronik, piloting humanoid robots for manufacturing tasks. The collaboration aims to automate low-skill, physically demanding labor, enhancing efficiency. This move reflects a growing trend in the robotics industry, with major players like Amazon and BMW also testing similar technologies. 

Apptronik's background in humanoid robotics, including work with NASA, positions them well in this competitive landscape. As humanoids gain investor interest, successful pilot programs could lead to significant industry advancements, shaping the future of automation.

Perplexity is ready to take on Google

Perplexity, an AI search startup, gains attention with high-profile users like Nvidia and Shopify CEOs. CEO Aravind Srinivas drives buzz with hot takes and impressive investor backing. With over $74 million in funding and a valuation exceeding $500 million, Perplexity challenges Google with its user-friendly interface and specific search capabilities. 

Despite its growing popularity, some users, including tech journalist Alex Heath, are not ready to fully switch from Google. Perplexity aims to prioritize factuality and accuracy, setting itself apart from Google's broader cultural focus.


This study dives into crafting top-notch Multimodal Large Language Models (MLLMs), spotlighting the critical role of various architecture components and data choices. By experimenting with different mixes of image-caption, image-text, and text-only data, the researchers pinpoint key strategies for achieving stellar few-shot performance across benchmarks. They uncover that while image encoders, image resolution, and token counts are game-changers, the design of the vision-language connector plays a minor role. Scaling these insights, they develop MM1, a family of cutting-edge multimodal models, boasting up to 30B parameters, achieving state-of-the-art pre-training metrics and competitive performance post fine-tuning. MM1 shines with capabilities like enhanced in-context learning and multi-image reasoning, making it a powerhouse for few-shot chain-of-thought prompting.

This research introduces WebSight, a game-changing dataset with 2 million HTML-screenshot pairs, aimed at transforming web design by enabling the direct conversion of screenshots into HTML code. The team successfully fine-tuned a vision-language model on this dataset, demonstrating its ability to accurately turn webpage screenshots into functional HTML. This breakthrough is expected to streamline web development, especially for those without coding skills, and the dataset is open-source to fuel further innovations in the field.

The paper presents Emu Video Edit (EVE), a new model setting the bar higher in video editing without using supervised video editing data. It achieves this by training separate image editing and video generation adapters, then attaching them to a text-to-image model. A novel unsupervised distillation method, Factorized Diffusion Distillation, aligns these adapters for video editing, enabling EVE to edit individual frames accurately while maintaining temporal consistency. This approach paves the way for further adapter-based capabilities in video editing.

GiT introduces a versatile Vision Transformer (ViT) capable of handling a wide array of vision tasks with a simple yet effective universal language interface. This framework, inspired by the flexibility of transformers in language models, aims to bridge the gap between vision and language tasks without needing task-specific modules. GiT showcases exceptional multi-task performance on five benchmarks without fine-tuning for specific tasks, demonstrating significant improvements over isolated task training. With further enrichment from 27 datasets, GiT achieves impressive zero-shot capabilities, signaling a promising direction for unified architectures in vision and language domains.

Quiet-STaR is a leap forward in language models, teaching them to "think before they speak" by generating rationales for each token to predict future text more accurately. This approach overcomes challenges like computational cost and the initial lack of capability to generate or utilize internal thoughts, leading to significant improvements in difficult question answering and natural text perplexity. The key innovation is a parallel sampling algorithm and an extended teacher-forcing technique, enabling zero-shot improvements on benchmarks without task-specific fine-tuning, making language models more general and scalable in reasoning.


Manychat - Drive more sales and conversions on Instagram, WhatsApp, and  Messenger using automation

Nara - Al-powered digital sales associate that helps your online store sell more and keep clients happy on all chat channels

Dextra - AI-Powered Assistant for Smarter and Faster Business Engagement

Mubert - offers a unique platform for app and content creators to generate royalty-free music tailored to their specific needs

BookPecker - 14509 books summarized in 5 bullet points

Speakz - Translate your media across languages, keeping style, voice and ambient sounds.

Pi - your personal AI available as a desktop app

Superhuman - turn an idea into a fully written email, improve your writing, fix your grammar, write in any language, and more.


Top websites block Google from training AI models on their data. Nowhere near as much as OpenAI, though.

Top websites are blocking Google from using their data to train AI models, a move prompted by concerns over AI monopolies. Google's response, a new tool called Google-Extended, allows publishers to opt out. While some major players have adopted it, uptake remains lower compared to similar tools like GPTBot. BUSINESS INSIDER

Deci announces new AI dev platform and small model Deci Nano

Deci, a contender in the AI scene, unveils Deci-Nano, a compact yet powerful language model, alongside a comprehensive Gen AI Development Platform. Departing from solely open-source offerings, Deci's shift hints at a blend of commercial and open-source models. Despite being closed source, Deci-Nano boasts high performance at a fraction of competitors' costs. VENTUREBEAT

India drops plan to require approval for AI model launches

India backtracks on requiring government approval for launching AI models after facing backlash. New guidelines advise labeling under-tested AI models instead. Critics slammed the earlier plan, with a venture firm partner calling it "a travesty." India aims to regulate AI to prevent misuse and ensure content integrity. TECHCRUNCH

UK And Germany Double Down On Joint AI, Clean Energy R&D Efforts

The UK and Germany have bolstered their collaboration in science and innovation, committing to joint efforts in AI, clean energy, and more. They plan to allocate funds and establish a task force to facilitate projects. This partnership aims to address global challenges and enhance research outcomes. FORBES

This Startup Cleans Up The "Ugly Underside" Of AI

Unstructured, a startup addressing the messy data problem in AI, raises $40 million in Series B funding. Led by Menlo Ventures, the investment values the company at $230 million. Unstructured's tool helps convert various file formats into one suitable for AI models, with clients including the US military and health insurance companies. FORBES

What'd you think of today's edition?

Login or Subscribe to participate in polls.

What are MOST interested in learning about AI?

What stories or resources will be most interesting for you to hear about?

Login or Subscribe to participate in polls.

Join the conversation

or to participate.