- ThePrompt
- Posts
- Google: "We have no moat and neither does OpenAI"
Google: "We have no moat and neither does OpenAI"
Open-source is winning the AI race
Hi folks!👋🏻 This is The Prompt! Happy Cinco de Mayo! 🎉
Today is the last day to apply for a free ad spot in The Prompt. Apply here with your apps.
Let's see why open-source is winning the AI race 👇🏻
FEATURED
Google "We Have No Moat, And Neither Does OpenAI" 🛡
In a leaked Google document, a researcher says that Google is not positioned to win the AI arms race, and neither does OpenAI.
Why?👀
Open-source.
Google and OpenAI solve “major problems”, but those same problems are already being solved by open-source models.
They do have one of the best-performing AI models, but the gap is closing — and really fast 👇🏻
And that’s far from obvious.
People are running AI models on their phones at 5 tokens, and you can easily finetune a personal AI on your laptop in one night.
All this happened when Meta’s LLaMa model was leaked to the public in March of this year. The model had no instruction, no reinforcement learning (RLHF), and yet the open-source community built everything in just 2 months.
Plus they basically solved scalability.👇🏻
"The barrier to entry for training and experimentation has dropped from the total output of a major research organization to one person, an evening, and a beefy laptop.
Here’s what else is mentioned
Collaboration with open source should be prioritized
Individuals are less likely to pay for restricted models
Google should prioritize the training of smaller models like LoRA, instead of costly and inefficient big models
Competing with open-source is not a winning position
Google should embrace open-source 🤗
Read the full letter here.
WHAT ELSE IS GOING ON
🦙 Salesforce announced SlackGPT. A native AI experience inside Slack which has the usual features: summarization, emails & connection with 3P integrations.
👀 Bing Chat is now in Open Preview. They also released more multimodal options, a multi-session chat, and opened developers’ access to build on top of Bing.
🏋🏻♀️ Biden met with the CEOs of Microsoft, Google, & OpenA. The goal was to talk about AI “dangers” and the big “r” word = regulation. The administration also announced that they’re spending $140M on an AI Hub — to create seven artificial intelligence research hubs. They’re trying to keep up!
💸Looseleaf (or PDF.ai now) was built and sold in 48 hours. I love everything about this - makers take notes! The app solves one problem - ask questions about your documents. The UI is amazing. Here’s how it all went down.
❌ OpenAI is reaching out to makers who use GPT in their app names. Apparently, it goes against their branding guide to use GPT in your name - have that in mind. You can get more info about their guide here.
🔥 Midjourney is starting to work on a 3D model. Not a clear timeline for this, but apparently they announced this on their office hours.
RESOURCES
The best resources we came across lately that will help you become better at writing prompts & building AI apps.
📚 What is a vector database [simple & very informative]
👋🏻 Daily papers on HuggingFace curated by AK [ great if you want to follow the research]
🎥 How to make AI videos [step-by-step video tutorial]
TOOLBOX
The latest AI tools to use or get inspiration from.
LLMreport: OpenAI analytics for your LLMs usage
AI Ads: Automated ads from your Shopify store
PeopleGPT: First AI search engine for People data
Iconify: App icons designed by AI
ChordCreate: Generate Chords with AI
PLAN: GPT for travel
Grid: AI-powered spreadsheet
IT Comm: Free AI PR service
Drumloop: Generate an original drum loop
PROMPT OF THE DAY
TOOL
Midjourney
PROMPT
1980s film noir mystery film still, low-angle medium closeup two-shot of 2 ancient Egyptian women opening a large gold treasure chest, kohl-lined eyes, bold lip colors, elaborate wigs, and flowing garments, ochre & gold tones, atmospheric lighting, dramatic light and shadow contrast, hieroglyphics, clues, hidden treasure, suspenseful --ar 16:9 --style raw
RESULT
LATEST PAPERS
NeuralEditor: Editing Neural Radiance Fields via Manipulating Point Clouds
Single-Shot Implicit Morphable Faces with Consistent Texture Parameterization
TUVF: Learning Generalizable Texture UV Radiance Fields
NeRSemble: Multi-view Radiance Field Reconstruction of Human Heads
Masked Trajectory Models for Prediction, Representation, and Control
What'd you think of today's edition? |