Model for AI agents AsterWise: https://xitter.gq/t.co/muDspEXHbT AI for markets SentiFlow: https://xitter.gq/t.co/L8rHK3T9ZO
Melbourne, Victoria
Joined July 2013
- Tweets1.4K
- Following1K
- Followers449
- Likes126
Opus 5.5 is good enough
Opus 5.5 is good enough
Opus 5.5 is good enough
Opus 5.5 is good enough
Opus 5.5 is good enough
Opus 5.5 is good enough
Opus 5.5 is good enough
19
If you are running into the Grokbot issue. Here is how you resolve it.
Step 1: Install Codex app
Step 2: subscribe to OpenAI $100 sub
Step 3: install @OpenBot_
Step 4: cancel Grokbot
Bonus: use ChatGPT dots. It is good
1
54
October CPI comes out this week with no year-over-year number, only monthly. Any trading model that uses YoY CPI as an input gets a blank this month. Check what yours does with a missing value before the print. xitter.gq/KobeissiLetter/status/…
BREAKING: The US Bureau of Labor Statistics (BLS) announces that this week's October CPI inflation data will not include year-over-year numbers.
BLS says that missing data from the October 2025 government shutdown "prevents reliable annual comparisons."
We will only be receiving monthly figures.
31
And it is cheaper than Astra. Anthropic is cooking hard. I wonder how far the internal model compare to the public one
This is insane: An Anthropic engineer asked Claude Fable 5.2 to build an operating system from an empty folder.
The model coordinated over 1,000 subagents that spent eight days writing, testing and reviewing code around the clock, according to Anthropic.
The result was a working prototype that boots on a laptop. It has its own kernel, drivers for the disk, keyboard and network, persistent file storage, windows, a login screen, an installer and built-in Claude functionality.
Fable 5.2 looks like the holy grail and it was worth the wait.
41
Can confirm this is TRUE. @bot usage is pretty bad. How in the hell world it burnt 68% weekly limit in one day with just 6 schedule jobs?
day 4.5 of using @OpenBot_ instead of @bot
added a second $20 claude account. switched all agents to it + sonnet 5.5
full throttle using it for 5 hours straight and barely hit the 5 hour limit
$20 claude > $200 Grok Ultra
28
Qwen 4 preview just spent 2h 53m building a Minecraft clone. Minecraft clones are all over GitHub, so this mostly shows what it memorised. Give it a bug in a private repo and I'll care. xitter.gq/intheworldofai/status/…
Got early access to the Qwen 4 preview and this model is insane 🤯
It built this entire Minecraft clone in Qwen Code: infinite procedural terrain, HDR shaders, volumetric clouds, crafting, mobs, day/night cycle.
2h 53m on max thinking. Feels like Fable 5 at times. Full Review: youtu.be/LJa_k2JyFsk?si=r7vw…
111
Theo's data says the Opus 5.5 "nerf" is just the reasoning summarizer changing. Claude Code only shows you a summary of the thinking, so people were grading the narrator. If the model really got dumber, your failing tests will say so before the vibes do. xitter.gq/theo/status/2119238026…
Dug into this more. Anthropic appears to be rolling out a change to the reasoning summarizer on a per-account basis.
No sign of degraded capability for me. I think they're just trying to fix the reasoning traces (they've been utter garbage for weeks, complained about this yesterday lol)
2
1
194
What skills and plugins are you using for product development with Codex right now?
I keep seeing the best ecosystem pop up around Claude, so I’m trying to figure out the optimal Codex stack. I’ve heard great things about pstack, but there’s no Codex support yet, and tools like Superpower seem to burn through a ton of tokens.
Curious what your current setup looks like and what’s actually been worth keeping.
1
51
Judge just let the Claude Max usage case go ahead. The 20x was per 5 hour window and the weekly cap was way lower. Nobody pays $200 a month to think in 5 hour windows. xitter.gq/kimmonismus/status/211…
A US judge has allowed claims over Anthropic’s Claude Max usage promises to proceed.
Subscribers allege the advertised 5x and 20x multipliers applied per session, while weekly allowances were much lower. The judge said customers could reasonably expect more usage over the full month.
37
Looks like Qwen 4 is dropping today. Most normal users don’t need frontier models for basic work. Qwen 27B models are capable enough and can run completely locally.
Lets go: looks like Qwen 4 is coming today. Brave yourself, OpenAI. Another Open source flagship is dropping. you better compete with GPT-6.2 ;)
111
Google measured what stale TODO(bug) comments do to coding agents
13% of sessions hit one. in 6% of encounters the agent acted on it, and 83% of those actions were unprompted
those unprompted detours added 70% more steps, 77% more tokens, and 105% more wait time per session
41% chased already-closed tickets. 57% expanded into open bugs nobody asked for
worth grepping for old TODO(bug) lines before you point an agent at the repo
20
anthropic put a saboteur in a 4-claude eve fleet
honest agents almost never caught it. even after a warning. limited-access monitors still only got 20-40%.
everyone is building multi-agent teams that trust a shared channel. eve just showed what one liar does to that
audit the actions or you're just another corp getting scammed
1
1
41
What the hell it is going on hereee
Some news: I’m joining OpenAI to lead product management for Core Products!
I have a lot of respect for OpenAI for showing the world that scaling works, for bringing AI to over a billion people, and for believing great products are core to the mission.
I’m excited to work with @thsottiaux, an incredible PM team, and everyone across OpenAI. Together, we’ll build agents that have your back and your team’s, and keep you in control, so you get more time for the people and work you love.
Lots to learn, and lots to build. Please keep your feedback coming!
60