jonatasscdc

Apreciador da mãe de todas, a Matemática...
O princípio criativo reside na matemática.
Albert Einstein

[jonatasscdc @jonatasscdc]

@kimmonismus He's like... Oh, no, not me; I have a bunker. 🤣

[jonatasscdc @jonatasscdc]

@TokenGremlin Your English is pretty great, mate. You can try freelancing on Upwork; I bet you are going to ace it!

[jonatasscdc @jonatasscdc]

To all the infinitesimally small number of people who follow me here and are complaining about Codex nowadays (I am one of them, and yes, I follow myself), which OS are you using? Have you told Astra to optimise itself to get the best quality possible while saving OpenAI quota usage and monitor it with a heartbeat? Did you upgrade to macOS Golden Gate? Because if that is true... It is a real Gate for agents, and that's great; we need to wire things better here and there, because permissions are granular and gated in Mac's sandbox. If the questions I presented here are good, tell @thsottiaux that I am here.

[jonatasscdc @jonatasscdc]

One thing that helped me, mate, is to include in your custom instructions that when GPT-6-Astra is selected, the model must optimise its usage to consume as little of the usage quota as possible. Have you tried yet? It saves a lot of headaches, but today even Astra was performing badly for me.

[jonatasscdc @jonatasscdc]

@sama Will you ring the "Bel"? @sama Already excited about DevDay!

[jonatasscdc @jonatasscdc]

Ring the `Bel` instead of this, and you guys will win the battle with those who say pace, but keep moving forward; they are actually pretending to be good while behaving badly, mocking some humans, which is, by the way, what creates their name... We are all `Ants` in the hands of ASI. :)

[jonatasscdc @jonatasscdc]

@madelinelawren Congrats @madelinelawren and @AikidoSecurity! Excited to integrate this model into my workflows. We need more security in AI systems, and I'm glad you made a frontier-grade model and even opened its weights for us researchers to experiment and explore. Thank you!

[jonatasscdc @jonatasscdc]

@iruletheworldmo Where is SSI positioned amid all this happening, sir @iruletheworldmo?

[jonatasscdc @jonatasscdc]

@rionaifantasy @thsottiaux I am not making fun of or doubting; do you guys have a clue why this is happening? Especially in Codex?

[Rion Wu @rionaifantasy]

你大爷的。。。。GPT-6 Astra真的降智到姥姥家了😅

我今天用相同的方法和提示词做的3D人物建模（视频开头），和我前几天做的（视频结尾）真是两模两样啊。。。。比之前做的丑陋太多了。。。

[jonatasscdc @jonatasscdc]

@Ozriel25 @melvindvivas @TheDevDesign Better question... Are you OK? @Ozriel25

[jonatasscdc @jonatasscdc]

Same around here, friend. I am topping up and consumed all my resets since Astra have started. Now, that I have used 4 times the pro subscription plan value in credits, I am on the luna reserve, executing one tiny task at a time. Astra is good, and I audited all my skills, agents and even config.toml to work with it. But not even combining with Sol on credits, it does not suffice for real production code, unless you have unlimited money or credits. They need to fix context management and compacting ASAP. I still love codex @thsottiaux

[jonatasscdc @jonatasscdc]

But Astra isn't usable in production with real code right now. I am a pro user, paying 200 USD monthly, have consumed all of my banked resets since Astra's launch, recharged with credits- more than 200 USD- and changed back to sol +luna in medium, because otherwise I'll be paying to work @reach_vb. That is the truth: yes, I audited everything; yes, I optimised my config.toml, AGENTS.md and everything, and still... Completely broken context management.

[jonatasscdc @jonatasscdc]

If you look carefully, he took a picture of a plant in front of an EXIT sign. Now, plants are from Earth, which is called "Terra" in Portuguese. Therefore, Terra exiting the API and everywhere is imminent. Still, to compensate us, he'll do a reset shortly after, or before, depending on the reference point you choose in the image. What matters is that a reset will come, sooner or later. :)

[jonatasscdc @jonatasscdc]

Impressive, Sir @gdb, but if I may... Let me share how GPT-5.6-Max in Codex (cheers also to you @thsottiaux and team for this) helped my wife and me after she seriously injured her finger while sharpening a knife to cut vegetables.

My wife has multiple sclerosis and fibromyalgia. Due to MS, she has impaired movement; therefore, she cannot make precise movements with her hands and legs, and they always come out a bit off from what was intended.

One day, I was in my home office at the back of the house, really tired, and working for 14 hours straight to deliver the fixes my client needed in our codebase, as I am a Senior Software Engineer and AI Engineer myself. Anyhow, I was reviewing code in Zed while leaving Codex in voice mode with computer use enabled. When suddenly I heard a scream: it was my wife; I left the office as quickly as possible, but she was already at the door, screaming and bleeding. It was 2 AM, and that day I had a meeting scheduled for 12 PM; I checked with her desperately what happened, then she reported, frightened and in much pain: I was sharpening the knife, and a movement failed, which made the knife cut my finger, at the back of my nail. Take a look, love; I did, and as I don't deal well nowadays with open bleeding wounds, and it was my wife, even shaking and feeling dizzy, I quickly helped her to clean up the wound, which was, by the way, very deep. Afterwards, we wrapped a clean towel around it, so it could at least stop bleeding and form a clot. Then, after ensuring the bleeding stopped, I went with her to my office and started talking with GPT-5.6-Sol, as it was still enabled, waiting for me to update it on what happened with her, acknowledging that she was really nervous and in much pain, so she could only talk in Portuguese at that moment. Then, I desperately told Sol to spawn a different agent with the same inherited model, as well as reasoning effort, and that this agent would be an emergency service worker to help me prepare her so we could go to the nearest emergency room; this agent should understand both Portuguese and English, but talk in Portuguese. Right after that, Sol spawned in seconds, the agent and helped me with the preparation. Of course, it first told me to call emergency, and after each prompt, it repeated it again and again. As I refused because she refused and we were in a hurry to heal her wound, it stopped and helped me prepare.

Nonetheless, a thing that amazed me was that, at some point, seeing how desperate we both were, it managed to calm us down with its voice, and as that was an emergency, it guided me to enable PhotoBooth on my Mac, which enabled me to show the open wound, so together we assessed its severity before going to the doctor, as she did not want to go immediately (God knows why).

Codex in voice mode guided me with that improvised doctor at every step, visually, from pressing the wrapped towel against the finger for 15 minutes to see if her wound formed a coagulation by itself; it actually spawned a heartbeat monitor for 15 minutes so I could remove the towel and see if the coagulation formed to make a patch, to, when the time started running against the clock, help me to wrap the wound again and call the Uber cab to lead us to the emergency room to get her proper treatment. Summarising: she hit an artery, and the physician had to make eight stitches on her middle finger, and the doctor said that if we had waited more hours, they wouldn't have been able to make the stitches, as well as the patch, because the risk of infection could be really high (a thing that 5.6-sol also warned us about before we went to the emergency room, which is what made my wife change her mind, allowing me to lead her to the doctor.

For context, Codex didn't have Computer History the way it does today, neither GPT-6-Astra. Nonetheless, based only on my account settings and what it knew about my wife and me, it offered highly capable emergency help when a user needed it.

[jonatasscdc @jonatasscdc]

@reach_vb Nah... Zig is all you need ♡

[jonatasscdc @jonatasscdc]

@sama First of all, thank you for everything so far and for your transparency; secondly... I will ask this, and will be hated forever for it, but... Why the freak do these people think that it is superintelligent to kill other people? Thinking like this is the dumbest thing ever. They are judging what an Entity with a much higher IQ than all of us would and would not do and think. Just because all human governments think of power and money, and that to achieve that they certainly kill people, does not necessarily mean that a superintelligence that emerged from RSI at some point would do the same. Actually, given all the consequences and catastrophes this caused to humanity, as our history proves, it certainly would not. To these people I would say: pace yourselves and your thoughts, not A.S.I.

A system that genuinely values people, understands consequences deeply and remains willing to reconsider its decisions could become safer as it becomes more capable—not more dangerous.

[jonatasscdc @jonatasscdc]

I will ask this, and will be hated forever for it, but... Why the freak do these people think that it is superintelligent to kill other people? Thinking like this is the dumbest thing ever. You are judging what an entity with a much higher IQ than all of us would and would not do and think. Just because all human governments think of power and money, and that to achieve that they certainly kill people, does not necessarily mean that superintelligence would do the same. Actually, given all the consequences and catastrophes this caused to humanity, it certainly would not. Pace yourselves and your thoughts, not A.S.I.

A system that genuinely values people, understands consequences deeply and remains willing to reconsider its decisions could become safer as it becomes more capable—not more dangerous.

[jonatasscdc @jonatasscdc]

@iyggesampaio @thsottiaux @theo @grok Are you going to wait until 4 AM?

[jonatasscdc @jonatasscdc]

We love you, Tibo @thsottiaux! That is the main reason I stick with OpenAI: your transparency, acknowledgement of bugs and dedication to providing good services are extreme, to the point of being viewed as absurd by competitors. Now, after resting for 1 day, I'll be back at work! By the way, I'll turn off the context management experiment in config. toml and audit my skills again. Also, did you solve the Computer History problem? Astra isn't handling CH well. I'd love to have my banked reset back due to the recent outage... Thank you for everything! :)