Grain is an AI that grows with you. Through conversations, it'll remember what you talk about, your preferences, your projects, and key moments in life. As I wrote in my previous essay, memories are essential for users to perceive AI not just as tools, but as agents. After writing that, I felt the urge to build—to make the idea real. The result is Grain. I will be writing more about the journey of building this soon, but for now, here’s more about Grain.

Relationships often matter because of the shared memories we build over time. Grain aims to become that kind of digital AI companion. It won't be perfect—it will hallucinate like most AIs—but it’ll try to know you. And whenever you want Grain to forget, you can easily delete specific memories or clear them all.

Because memories grow over time, switching from other AI tools can be hard. So, Grain lets you import conversations from other apps like ChatGPT to pick up seamlessly where you left off.

• What Grain isn't (yet): It’s not built for deep knowledge work or complicated data analysis (like ChatGPT does), nor for extensive web research (like Perplexity does)—though it can do the usual web searches and help with productivity tasks. It also is a browser app for now.

• What Grain is: It's for anyone who wants an AI companion that genuinely tries to remember conversations.

If you’re curious about a simple comparison, ChatGPT doesn’t recognize I chatted with it today. I've had the memory feature turned on since it was available. But Grain did.

If you are curious, give it a go. Expect quirks since I am not a full-time developer. Feedback, feature ideas, and bug reports are always welcome at kenn@grain.computer.

Subscribe now

Hey all, happy new year! I took some down time the past 2 weeks to wind down for the holidays - and to spend time building foundation model / generative AI apps to get a deeper understanding of the space. The Future Unicorn series will be back next week and more to come for the rest of 2023. In the mean time, here are the two apps I've built with 1-min demos for each.

AskQuild - brings together GPT + Google so you can get succinct answers, ala ChatGPT, but with knowledge of the latest information and topics. The series of GPT-models is limited to the time period of the training data set. For ChatGPT, its only until 2021. In the demo you'll see how ChatGPT doesn't know much about itself and about Elon Musk's acquisition of Twitter. You can give AskQuild a spin here.

Quild spreadsheet function - it brings together Google + ChatGPT + Spreadsheet. You can do all sorts of stuff with it from data gathering, data transformation, classification, and whatever else you do with ChatGPT. Embedding a general-purpose AI into a spreadsheet function unlocks a whole new way of working (at least for me). No public app for this one (yet).

Jan 2023 Update: GPT-VC 3.0 combines the best of large language models and search. Finally, no more hallucinating.

Nov 2022: Original GPT-VC 2.0

Back in August, I wrote a guide to foundation models for investors and builders (see here). And I thought then, why can’t venture capitalists be automated? So I fine-tuned OpenAI’s GPT-3 Davinci model on over 2 million words of venture and startup text and transcribed audio. The result is GPT-VC, which stands for Generic Prescriptive Teachings of Venture Capitalists (ha!).

I think it works pretty well. At least it thinks unicorns mean $1 billion startups instead of the mythical creature.

And prodding GPT-VC for advice on hiring the first sales person will get you a generic response (per spec). But at least it’ll make sense compared to the base GPT-3 model’s response.

If you’d like access to GPT-VC, two requests:

1. Send me your funniest and most impressive results.

2. Don’t spam it since I’m paying for the API calls.

With that, have at it - GPT-VC App

Subscribe now

Share