- Ben's Bites
- Posts
- AI transparency protocol
AI transparency protocol
PLUS: Anthropic enters the industry
Hey folks, a lot of highlights today. Snapchat released its own AI chatbot, an insightful thread around OpenAI’s foundry pricing (ahem $1.5m/yr), Anthropic rolling out to startups, AI transparency protocol, Meta adding AI to WhatsApp, Facebook and Insta plus Elon Musk is recruiting AI researchers to compete against OpenAI—a company he was a co-founder of.
Yeh, let’s just get to it, shall we?
🤌 Our Picks
Highlights if you've only got 2 minutes
Snapchat is bringing chatbot AI to the cool kids on the block, with an announcement that they’re rolling out an in-app chatbot called My AI. (My assumption of “cool” is based on the fact that everyone about 10 years younger than me seems to use it…) The Snapchat chatbot will act more like a contact in the app, with its own avatar and the ability to converse with it when you like. Given the young user base of Snapchat, it’s important that this AI is extensively regulated, and they’re off to the right start: it does indeed have strong personality constraints, refuses to answer homework questions, and is extra careful around politics.
Foundry, OpenAI’s upcoming platform for AI model training, has just had its pricing scheme leaked. And now we can see where Microsoft might start clawing that $10billy back: Model instances range from $250k/year to $1.5m/year! The mega-thread in the link above will give you the full low-down, but overall the platform is there to provide robust fine-tuning options, and might be using GPT-4… This is a big step, giving capital-backed companies the chance to really develop/tune their own tools to their own domains, which will be great for adoption and value. Furthermore, a push to in-house fine-tuning might be a move towards reducing hallucinations as we build an ecosystem of expert AI machines.
Anthropic - the ex-OpenAI team with a chatbot called Claude - has reportedly begun to make their AI available to a select number of start-ups. It’s not exactly news, as Quora were open about including Anthropic models in their Poe technology a while back. But the extra reports suggest they might be stepping back from their so-far coy approach to AI development. Given that their tech has made claims to improve on many aspects of OpenAI’s, this is an intriguing and welcome step that I’m excited to follow… But is it just because investors are piling on the pressure? Let’s not forget Google has a $300m stake.
I’ve had an unusually large desire of late to see strong progress on regulation and care in the world of AI development, and OpenAI’s report we shared a few days ago accelerated that further. Kinda feels like we’re all tripping over ourselves and maybe need some slowing down! The Partnership on AI (PAI), with more than 50 large member organisations, have put together an AI transparency protocol following extensive consultation, and some big players have signed up - OpenAI, TikTok, the BBC, and more. It’s designed to protect the consumer to ensure that people know when they’re interacting with AIs, as well as providing clarity about abilities and drawbacks. When large companies sign up for these sorts of things, landscapes can get poised to shift…
A few more (a lot of stuff to highlight today):
Meta revamps AI unit to get generative tech into products. (link)
Elon Musk is recruiting AI researchers to create an alternative to OpenAI's ChatGPT (Musk was one of the original co-founders of OpenAI). The goal of the project is to produce more trustworthy and reliable responses, but there is no concrete plan yet. Musk has previously criticized political correctness and 'woke' culture, and his new project could be part of Twitter or a standalone lab. (link - paywalled - posted a screenshot pdf in discord here. or alternative coverage here)
Keep your AI claims in check - An advisory from the FTC. (link)
Have you checked out Content at Scale?? (Sponsor)
With this new technology, you can upload a list of up to 100 keywords and within minutes, get 100 entire blog posts (2,600+ words!) written. It's so human-like it's actually not AI detectable, saving you against future Google updates and no more waiting around or wasting time on copywriting yourself.
It's not one of these regular AI writers, definitely worth checking out.
🛠️ Cool Tools
Product launches, updates and demos
FGenEds - Turn lecture slides to cheat sheets. (link)
Prompt Vibes - A massive collection of useful ChatGPT prompts. (link)
SwePT - AI sends pull requests for features you request in natural language. (link)
Life Copilot demo - AI assistant that can understand and do things for you. (link)
GitGPT - A natural language command line git assistant. (link)
Mage Space adds ControlNet and Instruct Pic2Pix models for text to image creation. (link)
Shortwave - Intelligent email summaries to read emails in seconds. (link)
Cleanvoice - Remove filler sounds, stuttering and mouth sounds from your podcast or audio recording. (link)
Imagen AI - Generate Variations of any photo, image or illustration. (link)
SupaRes - a blazingly fast engine for automatic AI image enhancement. (link)
Typeface - Supercharge personalised enterprise content creation. (link)
🤓 Miscellaneous
News, podcasts, videos, blogs etc
LLM-powered assistants for complex interfaces. (link)
Snapchat is releasing its own AI chatbot powered by ChatGPT. (link)
How much do customers actually care that your product has AI? (link)
Anthropic begin supplying its text-generating AI models to select startups. (link)
We're worrying about the wrong things about chatbots. (link)
Using large language models effectively. (link)
There are so many "prompt-ops" tools and I'm sold on none of them. (link)
Googlers are begrudgingly testing Bard and making memes out of their new task. (link)
Generative AI’s money game. (link)
How to create, release, and share generative AI responsibly. (link)
Gradient dissent - Better without AI. (link)
OpenAI's leaked Foundry pricing says a lot about what to expect in 2023 - a mega thread. (link)
Keep your AI claims in check - An advisory from the FTC. (link)
👋 Too many links?! I created a database for all links mentioned in these emails. Refer 1 friend using this link and I'll send over the link database.
🔬 Research
Published research papers
Directed diffusion from Google - Direct control of object placement through attention guidance. (link)
Face-TTS from Naver generates speech from a given text script and face image. (link)
Deepmind finds that differentially private diffusion models generate useful synthetic images. (link)
ChatGPT - A meta-analysis after 2.5 months. (link)
SpikeGPT - A generative language model with pure binary, event driven spiking activation units. (link)
Language is not all you need - Aligning perception with language models. (link)
Internet Explorer - Targeted representation learning on the open web. (link)
FedCLIP - Fast generalisation and personalisation for CLIP in federated learning. (link)
ELITE - Encoding visual concepts into textual embeddings for customised text-to-image generation. (link)
📰 Unclassifieds
Short, sponsored links
Use built-in AI to summarize any email, so you can do more while reading less 🚀 Launching today in free open beta - Shortwave.
Grow your business 10x faster with an AI marketing assistant. Instant insights on marketing, SEO strategy, tweet recommendations, and more. Check out the chrome extension here.
Aquabyte is hiring a computer vision engineer. Learn more»
Replit is hiring a full-stack software engineer (AI & data). Learn More»
Have a product, service, job, event, newsletter, app, book, movie, tool, or anything you'd like to share with over 40k subscribers?
Advertise with us | Job board | Join Community
🖼 AI Images of the day
Funny memes and pics from around the web
Send this to 1 AI-curious friend and receive my AI project tracker database! Use this link.
How was today's email? |
Reply