This is even more profound than Karpathy‘s approach to storing and retrieving knowledge - and certainly one of the most inspiring blog posts / papers I have read this year.
And you can think of extending this approach to domains like sales and marketing.
This is probably the most convincing use case for OpenClaw and Hermes.
But: as you are opening your heart and brain to LLMs it raises the question if local inference might be the best approach.
While I see the point of limited capacity, it also shows that Google did not plan for rate limiting / throttling of high usage customers. This is ALWAYS the problem with flatrate pricing models. 2% of your customers burn 80+% of your capacity. Did see that in former times with DSL, not too long ago with mobile and now with AI subscriptions. If you want to provide a "good" service for all customers better implement (and not only write in your T&Cs) a fair usage model which (fairly) penalises heavy users.
Good on them that they want to provide a way to bring back customers on board that were burned / surprised by their move.
BUT: The industry is missing a significant long term revenue opportunity here. There obviously is latent demand and Claws have a great product market fit. Why on earth would you deactivate customers that show high usage? Inform them that you have another product (API keys) for them and maybe threaten with throttling. But don't throw them overboard! Find a solution that makes commercial sense for both sides (security from API bill shock for the customer / predictable token usage for the provider).
What we're seeing right now is the complete opposite. Ban customers that might even rely on their account. Feels like the accountants have won this round - but did not expect the PR backlash and possible Streisand effect...
Both Google and Anthropic are choosing the wrong route here.
While I see the formal aspect of abusing an OAuth token and burning through subsidized tokens, this only creates an internal accounting problem in the short term.
Meanwhile the rising popularity of Claws creates a yet untapped new market segment where users spend significant tokens.
A „soft“ migration of users by explaining to them how the API works, how to pay and how to change from OAuth would be way smarter.
The way this plays out right now is that current Claws users are massively penalized by being suspended indefinitely and new users will think twice. And we can expect a solid PR disaster / Streisand effect for the „poor“ model providers like OpenAI or Anthropic.
Commercially choosing the soft route by warning and throttling will be way smarter and possibly generate more long term revenue
So here goes my OpenClaw integration with Anthropic via OAuth…
While I see their business risk I also see the onboarding path for new paying customers. I just upgraded to Max and would even consider the API if cost were controllable. I hope that Anthropic finds a smart way to communicate with customers in a constructive way and offers advice for the not so skilled OpenClaw homelabbers instead of terminating their accounts…
Is anybody here from Anthropic that could pick up that message before a PR nightmare happens?
Aperture is dearly missed even today.
And to make matters worse: you cannot even import Aperture libraries into Photos any more. Essentially leaving you with picking out the raw images from the package. And don’t get me started on excellent support for tethered shooting in a studio setting.
And I could go on and on.
The only thing I really missed in Aperture was first level support for Nik tools which are cool for their adaptive and non destructive masks.
Sigrok was offline for a couple of days but seems to be back online.
While researching the outage (which seemingly was caused by AI scrapers hitting the site hard) I learned that interesting discussions about the future direction of the project are happening on the mailing list
It seems PayPal is having issues with their login according to downdetector.com . Multiple reports on X with similar issues: Login problems, Captcha loops ad infinitum.
And you can think of extending this approach to domains like sales and marketing.
This is probably the most convincing use case for OpenClaw and Hermes.
But: as you are opening your heart and brain to LLMs it raises the question if local inference might be the best approach.