Core models
Cohere | Building the Future of AI
Inflection - Human machine interaction just launched Pi
AssemblyAI | AI models to transcribe and understand speech
Training layer
Weights & Biases – Developer tools for ML
Surge AI | Data Labeling Platform
Snorkel AI | Data-centric AI for the enterprise
HumanLoop - Developing/training base model seems to be thriving as well. Each industry have its own base model? No, probably not. Nvidia CEO stated that each industry/country in the future will have their own base model. I agree with country but not with industry. For the case of country, it’s necessary to host a base model within a country with its own language, train and is fed with cultural feedback. For an industry, not really. My hypothesis is base model is just like an average human? (is it? GPT-4 scored shit high in bar exam and Olympiad). Okay so a high performing fresh graduate then. But still, the model will need to be adapted to fit with industrial use case which is usually specific to the business. Isn’t it the same with human? Every job transition, one will need to be trained again. The only thing one can bring across is experience from the previous job.
Deployment
Vellum - Production ready for LLM
AI tools
Modular ETL pipeline for AI chatbots
Search
HeyDay - Remember everything you learnt - Sounds like what I wanted to do for personal AI assistant with remembering everything.
Personal assistant
Lindy AI - Personal AI assistant - Looks promising. Similar to Personal AI
Personal AI - AI that is you - Personal AI focuses on building one’s AI version (like Gilfoyce in Sillicon Valley did). But to me it’s very bloated and not focused on one thing. What I want to do is to create a bot that focus on learning and memory stacking. Personal AI uses their own model that is using personal data for training, but does it have to? A general language model is good enough for my application right? I dont need the bot to sound like me, I just need it to help with my memory and what I learnt.
Creativity
Synthesia - AI Video Generation
Galileo - Generate design for apps
Scenario - generate game assets
Move AI - aids movies with frames moving
Interact with software
Adept AI - I dont agree with Adept, even tho I agree the future of software interaction will be with natural language. Adept does general tasks on every software, which may not be what companies want. Companies want complete control over their data, and what user can do in their site, not through an extension. Big companies like Google, Wikipedia, or even browsers like Edge will eventually have their own LLM integrated directly on their site. It takes them no time to do that. So why bother doing general tasks on those sites? What big companies cannot address is tasks specific to a certain business. Edge or ChatGPT plugins cannot address all tasks a business wants their customers to do. Competitor in this sector will be CommandBar. If I can build this, CommandBar definitely can.
Fixie AI - In terms of use case, I agree with Fixie AI more. But who are the target users of Fixie tho? Seems like Fixie targets businesses as its customers. The flow can be: (1) Businesses can either leverage Fixie agents or create their own private agents, or (2) Businesses integrate Fixie to internal tools [like what?]. Let’s say a company uses Salesforce, or uses Keap, or Stripe, or Intercom. How does Fixie interact with those softwares to gather knowledge and answer questions? The only way is to build agent that makes API calls to those services. Who will delelop those agents? Fixie or the company or the service? I wouldn’t want to develop agents all the times to keep up with internal tools. Fixie cannot build all agents. And the service is busy developing their own business instead of developing these agents. Still a question needs to be answered. No code tool that builds agent, target either individuals or businesses. No code internal tools have been proven to be effective (Retool). 2 parts: One service is to build agents. Another service is to use the agents. The service to use the agents should be an app, or an extension? Isn’t it similar to Adept? Think more
My blog - How to embed this kinda application to basically everyone’s blog. Why needs a UI to navigate between blog posts tho? Still need UI for good reading experience. Maybe do something like this as the first application of what I want to do. Search bar for blog post, search blogs and have the options to read blog out loud. The hurdle here is users get used to have UI on everything, especially on a blog. UI creates uniqueness and personality to a blog. If every blog is only a search bar, bad UX. Need to figure out how to customise, personalise and create a unique feeling for the search bar.
Baselit - Customer facing chatbot. Similar to what I want to do, but this is very narrow in terms of use cases.
Defog - ChatGPT for data - Limited by use cases in data query only. Similar to Baselit.
Markprompt - Generate chat interface for doc site
Raycast - check this out later - look interesting
Not LLM - but rather tools that I find helpful
Command E - CommandBar
Lazy - Control environment using command palette
Developer assistant
TabNine - For application like TabNine, needs to read more on how it handles data.
Buildt - Search codebase - It’s a good one, but again, data privacy. Still need to look into how it handles data
Uptrain - Observability platform using AI
Trigger Dev - Open source Zapier
Grit - Fix Tech Debt Automatically
Second AI - AI developer agents for web apps
Hex - LLM aid for data analysis
GPT-powered for SaaS documentation
Education
The personal study tool that makes confident learners | Glean
Broadn - Seach engine to learn through podcast - I don’t see it has any potentials
Marketing
Automation
Extend - Internal tools using Chrome Extension. Looks interesting.
Research tools
Sprig - Evaluate user feedback
Misc
Moveworks: Solve employee issues
AlphaWatch - Market analysis assistant