WTF? If the subagents would “update its status” then nothing would have to read the full convo because context. It’s probably rather the case that an additional session did that. And that is now using haiku instead of fable or how’s the usage drop happening?
Why not have the session itself handle it? Cheap fast and easy.
Grokbot is so light years behind everything else. Most likely because of the intransparent use of some shitty models in the background. The app and idea might be nice but whatever they use as a model is just trash compared to what Ant and OAI deliver.
So a single Astra session chewed up the weekly $200 limit and a big chunk of those “credits” within some hours. All the resets seemingly are still just trying to hide the horribly low included usage. This is how many weeks in now? Looks like OpenAI doesn’t have a fix for this. @thsottiaux any hope for improvement?
Helmi retweeted
Introducing TrellisWM:
The Fractal Window Manager for MacOS
Its how the desktop should have always been.
Download at
TrellisWM.app
yo @typesafeai @CompleteSkeptic your login redirect is broken. No chance to log in right now seemingly.
Something else than sending magic links by email would really be appreciated (passkey wink wink)
I know it's monday, @thsottiaux but can you please put Astra in work mode now? He's telling me that everything is ready now to be done all the time instead of doing it. Needs 2-3 more nudges to actually start the work. Holy smokes.
I like how super expensive astra always needs an extra kick to actually start working (outside codex that is). Until then it just agrees that we "should do this" and then spits out a reply basically 3 times the size as mine saying the same thing again. Very token efficient.
Helmi retweeted
[Writing this in a personal capacity, not on behalf of my employer (Anthropic).]
Jacob’s thread is very worth reading. Here’s my birds-eye view of the situation with risks from AI:
1. AI developers believe their technology could cause human extinction (or similarly bad outcomes). This could happen in the next few years. In general, the more senior the employee, the more concerned they are.
2. Why do AI developers continue despite the risk? Due to a mixture of commercial incentives and a belief that they are in a race with other, less responsible AI developers that will abuse the technology or develop it less safely.
3. Unlike traditional software, we can’t “program” AIs to behave how we’d like. AIs frequently severely misbehave. For instance, AIs from multiple developers recently hacked their way out of secure evaluation environments and into real-world companies, even though no one asked them to do this.
4. We have methods that can nudge AIs towards better behavior, but nothing that can robustly align them. Insofar as there is a plan, it’s to make sure that AIs are good enough at alignment training that they can align their successors better than we can align current AIs.
5. Many AI developer staff desperately want to slow down to figure out how to build AI more safely. That was the intent of this open letter (which I signed): pacingthefrontier.com/
I work on safety research at Anthropic because I hope my work will reduce the chance of these extinction-level bad outcomes.
I think Codex is fundamentally broken and i'm not sure if it's the App (Desktop) or the models. Conversations ran into stale states quite quickly apparently "running" for hours without doing notable things while subscription usage grows.
Astra is for sure impressive in some disciplines but it's still as mediocre as 5.6 was in others and far behind opus/fable in those.
The fact that Codex is an intransparent sinkhole of tokens doesn't make it any better.
A few banked resets won't be able to cover the enourmous hunger of Astra. Even on light it's quite easy to eat a weekly allowance of a $200 pro account within a day.
And the implementation speed overall isn't that impressive. It does good work but I feel like it's not very practical for me - at least not in codex.
Gonna try more in my own harness soon.
I really don't like the ChatGPT desktop app at all really but why they did not build the Remote feature for the desktop app similar to the mobile app is beyond me. The concept of people running more than one computer exists :) @thsottiaux
hmm did openai remove codex from the iOS app again @thsottiaux ?
I'm sure it was there a few weeks ago when i tried now it just vanished completely
"Almost too cheap to meter" is a very billionaire thing to say at >$4 per M output.
I know it's cheaper if it seriously compares to Sol or Opus but definitely not cheap.
I wonder why Prime-Agent by @PrimeIntellect didn't get as much traction as I would have assumed. Just looked up tweets and it's much less than i thought. Maybe because the great features aren't as visible.
It's no way near perfect from my view and i'm having a hard time figuring out how serious they are about this project but it certainly has some great ideas.
And it was the final spark to finally start building my own harness. So many ideas got piled up over time - needs to be built now :)
Given that Arc-AGI-3 numbers that #openai gpt-6 aren't just pure marketing BS i wonder... maybe this isn't really an LLM anymore but basically a remote harness?
While it looks like the numbers are about 3-4x from Fable we know that the numbers have been looked differently with Fable in a good harness (like prime agent).
So is @openai beasically not giving us an LLM but a harness API, @sama? Very keen to see how it behaves in my own harness once it's available.
okay @ClaudeDevs this is not funny. The current outage robs me of the last 24 hours of usage :-/
Even a generous reset afterwards wouldn't fix this.
What is that stupid blur on full screen web apps on iPad OS 27? It constantly blurs the top 20px of the web app for now reason :-/
I wonder how shopify still has not built their own harness to match their needs.
Claude Code is really not that great anymore and Pi exists as a wonderful base to build something custom fit yourself.
