I've cancelled my Claude subscription and now I can't use even basic chat using Sonnet 5. Is it supposed to be like that?

No no, I mean models that were free before I subscribed are now not available after I cancelled.

So I've realized though Claude Desktop completely disabled all the models for me, Claude mobile app in the same account still has access to free models (I've noticed this just now because I rarely use Claude Mobile).

So, my first thought was that it's some sort of a cache invalidation issue. I restarted the app, didn't help. Logged out and logged back in, didn't help. But somehow it worked when I first logged out, restarted the app and then logged in. I guess that's the future we have to live in now.

“I turn now, good luck everybody else”

I don't think the people that are giving full credit to AI models for AI-assisted research results have ever done any scientific research. It's a bit more complex than "ChatGPT solve this problem, make no mistakes".

“ChatGPT for Academic Researchers” raised a lot of eyebrows when it was announced, and we’re already starting to see why. Although neither Buckmaster or Alpöge was on that program, the situation with Navier-Stokes equations is a clear example of corporate greed that has nothing to do with advancing science.

Spent almost an entire day to come up with the shittiest solution to https://spa-net.com/tasks/gate 💀
Getting rusty…

AI giant CEOs and tech influencers who have pushed for the development of AI “at any cost” are now all of a sudden saying we should slow down and regulate AI properly. I don’t know much about AI or finance, but to me it looks like either this is some sort of a damage control plan against the upcoming AI bubble pop, or more concerningly, something really really bad happened with one of their unreleased models.

Yandex to’lov tizimi 2-3 oydan beri stabil ishlamayotgandi o’zi, butun to’liq o’chibdi shekilli ))

Bizda odatda juma kunlarining ikkinchi yarmi ishga to'g'ridan-to'g'ri aloqador bo'lmagan narsalarni o'rganish, experiment qilib ko'rishga beriladi. 4-5 hafta oldin bir kunni HTTP QUERY bilan experiment qilishga ajratgandim.

Men duch kelgan asosiy muammolar L7 network infrada chiqdi (WAF, reverse proxy, load balancer, etc.), va juda ko'pchiligiga sabab shunchaki HTTP method nomlari softwareda hardcoded va hali query bu list/enumga qo'shilmagan. Masalan, nginxda limit_except'ga o'xshash method-based filterlar hali queryni support qilmaydi. Yoki yana bir misol ingress-nginx'da hali query default CORS allowed methodlarga qo'shilmagani uchun preflight requestlar fail bo'ldi.

Ko'pchilik web backend frameworklarda queryni native support qilish ustida uje ko'p narsa qilishayapti bilishimcha. O'zim ko'p ishlatadigan frameworklar ichida Flask bilan birorta muammo chiqmadi meni experimentimda, FastAPIda "technically" ishlayapti (class-based endpoint generationdan tashqari), lekin Djangoda hali ko'p ish qilish kerak.

Agar eski GET yoki POST endpointni QUERYga convert qilayotgan bo'lsangiz yana 1-2 ta tavsiya:

• Redirectlarni 307/308 qilavering, 301/302 query bilan qanaqa ishlashi kerakligi bo'yicha hozircha standard yo'q bilishimcha.
• GET requestni cache qilish uchun URLni o'zi yetarli, lekin POST bilan QUERY uchun bodyniyam hash qilish kerak. Cache setupni albatta qayta tekshirib ko'ring
• QUERY endpoint idempotent bo'lishi kerak, POST endpointni convert qilayotganda side effectlardan ehtiyot bo'linoradi.

spa-net.com qiziq challange ekan, kun bo'yi "enter" bosishdan zerikkanda bosh qotirib ko'rishni tavsiya qilaman ))