- The Strange Review
- AI companies need a "customer moat".
AI companies need a "customer moat".
Google's Bard now links to Google, Docs, and Maps. Vercel releases a UI tool that generates React code,
Happy Thursday! Lots to chew on this week. Investors are increasingly looking for AI companies that have a strong customer moat. Sequoia Capital’s David Cahn goes into detail in “AI’s $200B Question”.
This has been our core thesis at Strange from day one: that the biggest winners in AI will be driven by those that can win customers, and markets.
This is why design is crucial. And why we are doubling down on it at Strange. Design is the process of discovering and distilling user needs, and turning them into products people love.
Enjoy this edition! If you have any thoughts or feedback, just hit reply.
📰 The Long Read
From hate speech to AI music: The YouTube chief trying to leap tech’s biggest hurdles. Alison Lomax, London chief of the video platform, says it is committed to embracing artificial intelligence – but ‘responsibly’. In this interview, she talks about YouTube’s take on the next big technology. (link)
🔥 Latest news
This is a gamechanger. Vercel Labs releases v0.dev, a User Interface designer which generates copy-and-paste friendly React code.
Google’s new Bard extensions link Gmail, Docs, Maps, and more to its AI chatbot. The new version lets users double-check its answers, collaborate with others and integrate with Google’s own apps and services, including Gmail, Docs, Drive, Maps, YouTube and Google Flights and hotels. The idea is to use Bard as a one-stop shop to grab information across these different services. https://tinyurl.com/5n78wuwn
OpenAI hustles to beat Google to launch ‘Multimodal’ LLM. With all the reports of Gemini being released soon and potentially better than GPT-4, Open AI is trying to keep its lead intact. The multimodal features will be launched under the name “GPT-vision.” Also, they are training a multimodal LLM from scratch codenamed Gobi. https://bitly.ws/Vh44
Microsoft’s AI researchers accidentally exposed 38TB of private data. The researchers accidentally exposed sensitive data, including private keys and passwords, while publishing a storage bucket of open source training data on GitHub. The example shows new data security risks as companies use more AI. Sharing lots of data for AI training can lead to leaks if security teams don't guide researchers properly. Also, AI models themselves can be hacked if they aren't reviewed well, which can spread malware. https://tinyurl.com/2trwp9wp
OpenAI launches the Red Team Network to enhance AI model resilience. The contracted consortium of experts will play a pivotal role in informing the company’s assessment of AI model risks and strategies for their mitigation. While it may not directly remedy issues, red teaming serves as a vital mechanism for identifying biases inherent in models such as OpenAI’s DALL-E 2. Notably, this model has been scrutinized for exacerbating stereotypes related to race and gender. Additionally, it can pinpoint pitfalls that lead text-generating models, including ones like ChatGPT and GPT-4, to inadvertently disregard safety filters. https://tinyurl.com/y5f34r9d
Wowza. The Daily AI creates a Star Wars concept film using Pika Labs and Midjourney in just a few hours https://tinyurl.com/yyth7v2b
Created a Star Wars concept film using Pika Labs and Midjourney in just a few hours. It's wild how easy it is to animate the scene using these tools.
— The Daily Ai (@The_DailyAi)
Sep 20, 2023
Enjoy this thread of miniature worlds sparked by a Midjourney prompt tutorial posted by artist Allen T. https://tinyurl.com/2h7n9975
Let's get small and create miniature worlds 💎
It ain't easy being tiny, but someone has to do it. Let's imagine tiny worlds via Midjourney 🖤
Prompt: Micro tiny sized person next to a macro big sized [Subject], macro photography, beautiful, exaggerated perspective, dynamic… twitter.com/i/web/status/1…
— Allen T (@Mr_AllenT)
Sep 18, 2023
🎓 Recent Academic Papers
Google research introduces MediaPipe FaceStylizer: An efficient design for few-shot face stylization https://tinyurl.com/ms33yxtd
Researchers from MIT and Microsoft introduce DoLa: A novel decoding strategy aimed at reducing hallucinations in LLMs https://bitly.ws/VeBo
Baichuan 2: A series of large-scale multilingual language models containing 7B and 13B parameters, trained from scratch, on 2.6T tokens https://bitly.ws/VeAp
🔧 Cool Tools and Experiments
Shortwave: An AI-powered assistant that lets users ask questions about their email history https://tinyurl.com/bdd78938
Catch+Release launches an AI-powered search for user-generated content https://tcrn.ch/3PpDsY5
Typeset launches AI that rapidly generates presentations, social posts, and more https://bitly.ws/VeoN
AI-powered visual branding platform Catbird, that competes with industry giants like Pinterest and Canva, exits beta with 300K users and $100K in recurring revenue https://tinyurl.com/ye27fzwj
Developer Thomas Simonini shares a tutorial on how to create NPC characters using Unity Sentis and Hugging Face transformers https://tinyurl.com/ynnknsec
💰 Latest Startup Fundings
Social gaming and livestreaming platform Eloelo raises $22M in a pre-Series B round to expand its entertainment app https://tinyurl.com/r4e96pwu
Learn.xyz secures $3M to launch its AI-driven social learning app that combines elements of Duolingo and Wikipedia to generate quick, informative “Curiosities” https://tinyurl.com/yck89s5r
LimeWire acquires BlueWillow, a generative AI image creation platform that competes with Midjourney and Stable Diffusion https://tcrn.ch/3PsOzjd
Babbel acquires language learning browser extension Toucan https://tcrn.ch/3PJGw2H
Did you enjoy this post?