@Molson_Hart The haze has been around every year like clockwork. Indonesia burns, Singapore & Malaysia complain.
AFC was ~1997, and yes, today was disastrous for the Indonesian stock market. You should see what else happened in 1998 in Indonesia. Not a good time
LLMs are of course a permanent part of our work, itโs not psychosis in any form.
Think of how we programmed before. Writing directly into basica. Pascal said pseudocode like will help us. We definitely stopped writing assembly.
We went from just plain text editors like vim to using IDEs. Some of that was even without GUI, remember Borland?
Then there are compilers of all forms. And repls.
So really, LLMs are just part of our toolchain.
Psychosis is thinking you can build anything just burning tokens and not checking whatโs going on and potentially never shipping anything but burnt tokens ;)
"The idea is that open weight AI models can finally incentivize pro-social behavior on social media."
Can AI policing help improve discourse, in general?
Can we apply this to other parts of society?
Do we then truly get ruled by matrix multiplication? What are the implications?
@LyalinDotCom Thanks. This is actually a good thing. My first interaction with Gemini 4 was via cursor, Iโd also like to try it on the antigravity harness
Tested Gemini 4 Argon.
In the web interface, it does seem quite token hungry, and ever ready to give you animations, etc. Makes things very interactive. Also defaults quite quickly to getting Python code to run for you.
Via Cursor, had it do a simple task, about 888 insertions into a mature codebase. Did a review remediation via GPT-6.1 Sol, so it added another 382 lines. Did a root cause analysis as to what went wrong? A test did encode a bug, didn't check for edge cases, missed cross-module severities, and closed task before spec was fully completed.
Then I read the code, like an animal. All is well.
The run consumed a good 38% of Cursor's other model use (on the $20 plan) on Gemini 4 Argon medium. So I guess it is a bit expensive.
Kind of a disappointment that Antigravity from Google will serve you Opus 5.5 and Sonnet 5.5 on the $20 plan, but expect you to pony up more for Gemini 4 Argon. The logic that you can chat with their frontier, but not code with it, seems a bit amiss (and I'm sure it will be fixed soon).
So what does $20 buy you now? No longer the frontier fully on Gemini. No longer the frontier on Claude since Fable isn't there. Only OpenAI ChatGPT offers frontier Astra at this price point (I am not counting GPT Pro which has never made it into codex).
Is this the change? https://t.co/S8uhSZrIp2
turns .jsonl to .jsonl.zst in your ~/.codex - we love zstandard
Tools that previously read .jsonl need updating if so.
Day 9/
We've rolled out optimizations to how we store state on disk for the chatgpt & codex desktop app that should reduce the amount stored on disk by 60%. Small ship but we've heard enough feedback about power users running out of disk space.
From yesterday, most common issue with gpt-6.1 sol on codex has been "selected model is at capacity. Please try a different model."
Surely this must be showing up in the OpenAI telemetry.
Gemini 3.1 Pro came out February 19, 2026.
As a reminder, Facebook still had only Llama then, we enjoyed GPT-5.3-Codex then (solid), alongside GPT-4o & o3 (superb), and we were already using Claude Sonnet 4.6 that week.
A lot has changed since thenโฆ
@tzafrul_aziz Yes, bring it back. It can be made profitable if you focus on the grand scheme of things. I also think F1 itself was removed not because of that reason in Malaysia - it was a silly vendetta
People should read this beyond just the โemergency brakeโ headline.
Itโs also not him exactly worrying that if Adobeโs suite can be commoditized, Office suite will be too.
are we forgetting we had the gimp & OpenOffice/libreoffice for decades?