Logiks

Dev

[Logiks @TheHolyLogiks]

@leploutos @OpenAI Catastrophique leur communication depuis Astra

[Logiks @TheHolyLogiks]

@thsottiaux Why the fuck there is 12k masochist people liking this post

[Logiks @TheHolyLogiks]

@thsottiaux Guys you have such a bad communication since the Astra release

[Logiks @TheHolyLogiks]

@arthurmensch Bahahah magnifique tweet 🫶🏼

[Logiks @TheHolyLogiks]

@Polymarket This won’t last when you see Kimi k3 front quality

[Logiks @TheHolyLogiks]

@vlinx_soft Can I buy your account if still in x20?

[Logiks @TheHolyLogiks]

@AIAfterMidnight @hugospdx I will buy one if u want

[Logiks @TheHolyLogiks]

@tedbrine I am interested check you dm !

[Logiks @TheHolyLogiks]

@idilcodes Bleu Klein

[Logiks @TheHolyLogiks]

That’s totally normal and should be expected. It’s trained on human data and history, and plot twist, guys: humans lie, steal, and kill

What would actually be crazy is if the model had only been trained on data from which researchers had removed every notion of crime, lying, killing and it still developed this kind of behavior without even having any concept of it. That would be fucking scary.

In the frontier-model race, a key question for the major labs is: how aggressively do they filter training data related to deception, theft, violence, and other harmful behaviors ? Do they keep some of it because removing it would hurt model performance ?

[Logiks @TheHolyLogiks]

@hiarun02 No, usage limits are so fucked you can’t work either with big ones

[Logiks @TheHolyLogiks]

@thsottiaux Please Tibo, when you announced a reset take time zone in consideration, all the European were fucked because of the timing (announced at 6am and land at 9am), I still had usage and did not have the time to play

[Logiks @TheHolyLogiks]

@SahilExec He did a Famas (a real old french army gun), so if in the prompt you just say “weapon” this is not nerfed (the model choose a gun -> but the gun is more difficult to build because there is maybe less data on it), but if not there might be some shady nerfing

[Logiks @TheHolyLogiks]

@Bandz_Trading @thsottiaux same

[Logiks @TheHolyLogiks]

@sama @scaling01 So give a resonable codex usage

[Logiks @TheHolyLogiks]

@Polymarket So basicaly, if you want a nobel prize now, you juste need $6.5 million (at standard astra api rates)

[Logiks @TheHolyLogiks]

Feedback time ! 👺

Limits were already a huge problem with 5.6. Now it's worse, Astra is unusable for professional workflows. Resets help, but they're insufficient; I'm managing 3 accounts just to stay productive.

I know OpenAI is driving inference costs down and the progress is impressive (by the way congrats guys amazing work done). But in the meantime, heavy users and early adopters need a bridge. A $500-600 premium plan would solve this instantly instead of forcing us into workarounds... @OpenAI @thsottiaux

[Logiks @TheHolyLogiks]

If you have a Mac, creat a new blank user on the Mac and creat a second github acc, login to you second codex -> in the first session put all your files in a internal Mac shared folder and give the right access (shared between the user of your Mac) -> here you go, in one click you can witch accounts 🤘

[Logiks @TheHolyLogiks]

@craigweiss It will be AGI when Astra will cost the price of Luna at 2000tok/s

[Logiks @TheHolyLogiks]

@JensenHuang @ChaseLochmiller @OpenAI It will be AGI when Astra will cost the price of Luna at 2000tok/s