Interesting, I would love to hear how well those worked.
Grammit uses the Prompt API for the local LLM, which currently uses a version of the Gemma 3n model on Chrome.
Grammit uses prompting instead of fine-tuning or custom training. Simplified, it has a system prompt along the lines of: "Rewrite this text, correcting any grammar or spelling mistakes," combined with a prefilled conversation containing a number of examples showing an input sentence and the corrected output.
There is a Prompt API in development that's available in both Chrome and Edge to give access to a local LLM. Chrome extensions have access to it and I believe websites can request access as part of an origin trial.
The model is fully managed by the browser. It's currently the Gemini Nano model on Chrome, and they are testing a version of the Gemma 3n model in beta channels. Edge uses phi-4-mini.
I've created a few DSL's and I've regretted not closely emulating an existing language that users already know.
When creating a new DSL you should find the language that is most familiar to your users. That could be Python, Javascript, or SQL. There is a good chance though that it is spreadsheet formulas.
Whatever that language, align capabilities you add to your DSL with the syntax and function naming of that language only diverging where needed to add the custom capabilities which motivated the DSL.
The closer you align with what your users already know, the more easily they will be able to adopt it. An added benefit of this that emerged in the past year is GPT's will be able to write your DSL with less prompting.
Disagree on using Bayesian statistics. Frequentist statistics are perfect for A/B testing.
There are so many strong biases people have about different parts about UI/UX. One of the significant benefits of A/B testing is that it lets you move ahead as a team and make decisions even when there are strongly differing opinions on your team. In these cases you can just "A/B test" and let the data decide.
But if you are using Bayesian approaches you'll transition those internal arguments to what the prior should be and it will be harder to get alignment based on the data.
It's a similar situation with Chrome extensions. Extension reviews are generally pretty quick (an hour or so), but every once in a while you get hit with a longer one (days+).
In this type of environment, you need to ensure every release is as rock-solid as possible. For our extension, we have beta extension with a sub-group of opted-in users that we test on for a week or so before doing a production release. Then we roll out the extension to production incrementally starting with 1% of users and slowly ramping that up to 100% (it seems Android has a similar staged rollout feature).
We built Text Blaze [0] a Chrome Extension [1] that supports a lot of similar capabilities.
It lets you do text expansions on any website using hotkeys, include dynamic values like the date a week from today in your snippets, build form UI's [2], and include dynamic logic using formulas and if-statements [3] (it uses a dynamic reactivity model for formula's similar to spreadsheets).
One thing we are really excited about currently is using Text Blaze snippets to connect any website to an Airtable/Monday.com like spreadsheet using a lightweight SQL variant we created [3]. Combined with forms and formulas, this allows you to build mini-apps you can use on any site with a hotkey.
The Bard model (Bison) is available without region lock as part of Google Cloud Platform. In addition to being able to call it via an API, they have a similar developer UI to the OpenAI playground to interactively experiment with it.
Firebase uses Google Cloud Logging. Taking a quick look at the blog post here, Google Cloud Logging already seems to support everything it describes.
Is there something in it that makes it a better solution in some way than what Google is already providing? (Note that Supabase Logs appears to rely on Google BigQuery so you'll be running on Google either way.)
We've explored usage based pricing for our app which has a core unit-of-work that delivers value to the user. We've talked with various customers about adopting usage based pricing in place of our standard cost per user pricing model. We thought it would be great to align user costs with value delivered.
Turns out no one wanted the usage based pricing even if it could save them money based on their predicted usage. Companies preferred the consistency, clarity and control of a user based pricing.
We didn't try out a cost cap like the $20 here, so maybe that was a missing ingredient in our tests.
My company has an Airtable competitor that has API support for SQL. It has pretty good SQL capabilities solving the issues with pagination you have in Airtable.
No joins yet between tables yet, but they are planned. (You can use lookup fields to achieve them to some extent, but that requires planning ahead and is limited).
Google Groups is heavily used within Google's ecosystem. Specifically it is:
- a system for managing group-based access for Google Workspace users, e.g. share a Google Doc with a group to give all members access
- a system for managing group-based access to Google Cloud Platform resources, e.g. share a BigQuery table with a Google Group to give all group members access to it.
- a simple way for Google Workspace users to create shared email addresses. e.g. [email protected]
I created a fairly popular system dynamics modeling application back when I was a grad student [0].
When using modeling tools like system dynamics, it's useful to keep in mind George Box's quote that "All models are wrong, but some are useful".
When using a modeling tool to describe any form of social system you're creating an imperfect copy of it. This imperfect copy embeds what the modeler (rightly or wrongly) views to be important and how they believe the system to work.
The resulting model, though always wrong to some extent, may be useful. It may help you obtain a better understanding of a system and cause positive change in an organization. On the other hand it may not be useful, it may even mislead you.
You can think of modeling a lot like various software engineering practices like Agile. Sometimes these help teams, sometimes they don't. At the end of the day though, it's really about the teams using them not the specific techniques.
"There is nothing, according to me, that deserves more to attract our regard than the intellectual and moral associations of America....
In democratic countries the science of association is the mother science; the progress of all the others depends on the progress of that one."
[0] https://press.uchicago.edu/Misc/Chicago/805328.html