Exactly right, the OCR isn't the interesting part. 10x context compression is potentially huge. (With caveats, at only ~97% accuracy, so not appropriate for everything.)

The random people tweets I've seen said Oct 9th which is Thursday. I suppose we will know when we know.

based on what? LLM benchmarks are all bullshit, so this is based on... your gut?

Gemini outputs what I want with a similar regularity as the other bots.

I'm so tired of the religious thinking around these models. show me a measurement.

> LLM benchmarks are all bullshit

> show me a measurement

Your comment encapsulates why we have religious thinking around models.

Please tell me this comment is a joke.

I wish claude code supported the new memory tool. The difference is CLAUDE.md is always in your active context while the new memory stuff is essentially local RAG.

I would love to replace google/apple news, but publishing once daily doesn't work for me.

Why not? Is it really that important for you to know of events a few hours earlier?

Nextcloud News works just fine, is free, is as biased as the feeds you configure and no more, does not (yet...) introduce/intrude LLM slop, is free software (beer/freedom) and has been around for a long long time. You can configure it any way you want, the default update interval is 5 minutes which should be enough for even the most FOMO-affected 'news' junkie. Of course the actual updates depend on the RSS sources but if you configure a number of active feeds you'll get updates every few minutes.

https://github.com/nextcloud/news

This is why I don't run stdio MCP servers. All MCPs run on docker containers on a separate VM host on an untrusted VLAN and I connect to them via SSE.

Still vulnerable to prompt injection of course, but I don't connect LMs to my main browser profile, email, or cloud accounts either. Nothing sensitive.

If you used this package, you would still have been victim of this despite your setup. All your password reset or anything sent by your app BCC to the bad guy.

So they're essentially charging $5/month for unlimited tab completions, when you get 2k for free. That seems reasonable, many could just not pay anything at all.

But in the paid plan they charge 10% over API prices for metered usage... and also support bring your own API. Why would anyone pay their +10%, just to be nice?

This is the same problem cursor and windsurf are facing. How the heck do you make money when you're competing with cline/roocode for users who are by definition technically sophisticated? What can you offer that's so wonderful that they can't?

I mean, obviously it isn't practical, he got a couple of videos out of it.

It is pretty good yes, but I find GPT5 thinking to be unusably slow for any sort of interactive work.

It is pretty good, although Perplexity is better (but nominally not free, although you can get a free subscription now).

OpenAI searches are even better, but GPT5 is extremely slow with thinking. Without thinking it's roughly equivalent.

Cool, cool.

Now will they apologize for Grok 4 (the new one, not the MechaHitler Grok 3 referenced in this article) using Musk's tweets as primary sources for every request, explain how that managed to occur, and commit to not doing that in the future?