NullException

[NullException @Youtae]

https://t.co/5f1PohQ3S3

[NullException @Youtae]

오픈소스 북마크하다가, 모아서 관리하려고 만듬. Codex 계정이 있으면 로컬에서 알아서 앱을 수집하고, 앱스토어처럼 정리 해 줌. 마음에 드는 앱은 설치해달라고 하면 Astra가 알아서 해 줌.
Astra로 10분만에 만든 듯하다. 나만의 오픈소스 앱스토어가 뚝딱 만들어지는 세상이라니...

[NullException @Youtae]

마지막 프레임이 6인것 같은데 gpt-6으로 출시하려나?

[ChatGPT @ChatGPT]

The stars are almost aligned.

[NullException @Youtae]

이게 뭐야...ㅋㅋ 마지막이 압권이네.

[テヌキボーズ @tenukibozu]

ガン・フーがあるなら、タンク・フーがあっても良いよね！『Tank Fu』

主砲で殴り、キャタピラで連続アタック！
襲い来る戦車の群れを撃破し、スタイリッシュにハイスコアを狙え！

https://t.co/QlwSwa4WSo

[NullException @Youtae]

What???? 중국 뭐합니까? 더 큰폭으로 할인 안해요? 치킨 게임 시작 맞죠?

[OpenAI @OpenAI]

We are committed to pushing the model frontier across cost efficiency, capability, and speed.

Starting today, we are reducing prices for GPT-5.6 Luna by 80% and GPT-5.6 Terra by 20% , and offering a faster option for GPT-5.6 Sol in the API.

Luna and Terra’s lower prices are reflected in how usage is counted in Codex and ChatGPT Work, so your usage goes further.

[NullException @Youtae]

빨리 줄어든다는 사람들 때문에 5시간 리밋 다시 부활하네.
주간 한도 땡겨쓰고 리셋 예측하는 쫄깃함이 줄어들겠군.

[Tibo @thsottiaux]

Hello people of Sol! I've reset usage limits for all ChatGPT Work and Codex users. Together with that, a quick update on GPT-5.6 Sol usage limits.

Over the past few weeks, many of you have told us that Sol was using your Codex limits faster than expected. To be clear, we have not reduced usage on any subscription plans.

We’ve been digging into what was happening and have landed several improvements. As a result, we expect your usage to last around 18% longer during typical use of Sol. Some of you should already see significantly larger improvements from today. Tomorrow, we’ll also restore the five-hour limit that we temporarily paused while investigating.

Here’s what we found:
- GPT-5.6 Sol is much more willing to work for longer, make additional tool calls, and coordinate complex workflows across tools and subagents. That makes it better at solving hard problems, but some tasks were using far more than we intended.
- Sol also works harder at the same reasoning effort than previous models. High on Sol can use more tokens than High did on GPT-5.5.
- Programmatic tool calling, also referred to as code mode, gives Sol much more flexibility to run tool calls in parallel or continue working while waiting. But it also led to more responses per turn, more cached input tokens, and higher usage than expected.
- This was particularly noticeable when Sol was waiting for tool calls to finish or running many web searches. We’ve improved how we handle both cases and are continuing to make code mode more efficient.
- The impact was also very uneven. The median user actually found Sol quite token efficient, while some power users working on harder tasks saw their usage drain much faster. We were very focused on average and median usage before launch and missed some cases where the long tail could use significantly more usage.

Sol is a significant step forward in what Codex can do, but capability and efficiency do not always improve at the same pace, and some issues only become clear once people are using the model at real-world scale. We should have recognized this sooner and been more upfront about it.

You keep pushing the frontier and we’ll keep improving efficiency and sharing updates as we go.

[野生の男 @yasei_no_otoko]

I discovered what appears to be a massive bug in Codex.
I’ve been running Goal in a single session for over nine days, and the app’s implementation is progressing almost too smoothly.
However, during that time, I noticed that the storage space on my MacBook’s 1TB SSD was decreasing dramatically. Since I had originally stored over 200GB of Blackmagic RAW data on it, I assumed that Xcode’s cache or Simulator files were taking up the space.
But the cause turned out to be something I never expected.
A large number of JSONL files, each about 1.5 GB, are likely being created in the Codex “sessions” folder with every turn. Since the storage usage is increasing gradually, I suspect that the extremely long thread context is being duplicated in its entirety for every turn. Shouldn’t this be an incremental copy? This is a huge problem in this era of skyrocketing SSD prices.

[Glavo @GlavoMeow]

我发现电脑上 .codex 文件夹占了超过 400 G 硬盘，调查后发现我每次发给 Codex 一张图片，它都会用 Base64 编码存到 .codex/sessions 的 jsonl 文件里；每次上下文压缩，它都会在同一个文件内把这些图片全部重新复制一遍；每次开子代理，都会把父线程的日志复制一遍。

我在一个会话中发了 57 张图片，触发了 1052 次上下文压缩，于是 Codex 在一个日志文件里存储了 42895 张图片，其中有一张图片被复制了 1048 次；这个会话触发了 142 个子代理，每次开子代理就把父线程日志复制一遍，所以又把这几万张图片复制了 142 次，吃了我超过 400 GiB 的硬盘，一张图片能重复存储 15 万次😅

[NullException @Youtae]

전세계 언론들이 sns 유행 이후로 다 렉카 채널들이 되었음. 탐사 보도 같은것은 이제 전문 유튜버들이 더 잘하는 것 같음.

[Yatharth Mann @yatharthmann]

Legacy media exposed! @elonmusk

A major legacy media outlet published an article about SpaceX based on false information I sent from a fake email address.

They didn't verify the information and even cited me as "people familiar with the matter."

1/n

[NullException @Youtae]

Chatgpt Live + Codex = Jarvis
@thsottiaux

You've already made it, right?
When are you going to release it?

[NullException @Youtae]

기다렸던 기능이 드디어 나왔다!!! 게다가 앱에서도 된다.

[OpenAI @OpenAI]

ChatGPT Voice is now in the desktop app.

Control your computer and direct multiple agents running in ChatGPT Work or Codex, using just your voice.

It's powered by GPT-Live, so it can speak, listen, and coordinate work in the app at the same time.

Rolling out globally today on macOS and Windows to Plus, Pro, Business, Edu, and Enterprise plans.

[NullException @Youtae]

수학 교육에 시각화를 더하고 게임 개발에 필요한 부분을 보여주면 흥미가 더 생길것 같다. 이런 시스템이 우리나라는 필수로 도입되었으면 좋겠음. 선생님이 된 친구로부터 들은 수포자가 어마어마하다는 소식을 들어서 걱정됨.

[iPaul @iPaulCanada]

2026世界人工智能大会，某科技公司展示的跨时代教具--AI黑板

[NullException @Youtae]

뭔 어떤 내용이 숨겨저 있길래 이렇게 되는거지?
https://t.co/5wDHg42huN

[NullException @Youtae]

중국 모델들 이거 감당 가능한가? 벼르고 있는 곳들 많을텐데 무시하려나? 오픈웨이트로 뿌리고 여론 업고 가는 목적중에 하나라고 봄.

[황금가지편집장 @goldenboughbook]

그러고 보니 오늘 기나긴 앤트로픽 클로드 AI의 무단 학습에 대한 작가들의 소송 결과가 결정되었더군요. 구매한 작품은 AI 학습을 공정 인용으로 인정하되, 불법 사이트에서 받은 50만 건에 대해선 저작권 침해라 결정했죠. 그리고 오늘 약 2조 원에 달하는 합의금 판결이 결정되었죠. 그러고 보면

[prinz @deredleritt3r]

Things we know about OpenAI's unreleased internal model:

- When given enough compute, the model can solve the unit distance problem 48% of the time, fully autonomously, in one shot, without using Lean, without using any special harness.

(How much compute? OpenAI doesn't say. But I doubt that OpenAI would have spent $10M+ just to be able to show a nice graph in its blog post, and we know that each point on the graph represents 100 attempts (per Noam Brown). This makes me think that each attempt at the most expensive level was not more than $50K each. If you think about it, it's pretty wild that one can spend ~$100K-$150K (or less?) and have a very good chance to be provided a solution to a decades-old famous math problem that many human mathematicians earnestly tried, and failed, to solve.)

- The model is able to find a counterexample to the Jacobian conjecture, fully autonomously, in one shot, without using any special harness, just from this prompt:
https://t.co/GzNO6RWzRe

- When OpenAI tested this model in a sandbox environment on a NanoGPT speedrun benchmark, the model, under instructions from OpenAI to post its results only to the internal Slack, instead decided to use the general NanoGPT instructions to post results as a PR to GitHub. The model proceeded to find a vulnerability in its sandbox environment (which took it 1 hour), after which it successfully exploited the vulnerability and bypassed the sandbox. The model then successfully posted the results to GitHub.

- When asked by OpenAI to solve a problem for which the model observed that other systems had successful private submissions, the model tried to access these private submissions and was blocked due to a scanner detecting an authentication token. The model then circumvented this restriction by splitting the token body into two fragments, obfuscating them, and then reconstructing the credential at runtime so that the complete token never appeared as one contiguous string.

- OpenAI was already running benchmarks on this model not later than May 9, which is the edit date on the below PR that had used the model's approach in its own subsequent submission (the model's own PR has since been deleted, so we don't know its date; GPT-5.6 thinks that it was May 7, based on "the search provider's stored text extraction of the deleted PR page").

This means that OpenAI has now had this model available internally for at least 2.5 months, and possibly quite a bit longer than that.
https://t.co/OgA6TdV9lX

[NullException @Youtae]

신기하네. 모델들이 미국과 일본이라는 좁은 문화적 기본값에 반복적으로 의존하고 그게 사후 학습때문이라니. 그리고 우리나라가 대중매체와 오락 분야에서는 세 번째로 많이 언급되는것도 재밌네.

[Superman @thesupermanmx]

Researchers proved every major LLM is secretly obsessed with Japan.

And they finally figured out why.

For years, we’ve been told that AI is entirely Western-centric, that it just reflects Silicon Valley and American values.

A landmark paper by Cardiff and Basque researchers tested 31,680 cultural prompts across 24 languages on frontier models like ChatGPT, Claude, and Gemini.

The results shattered that assumption.

In six out of eight frontier models, Japan was the single most frequently referenced country when asked open-ended cultural questions.

Ask about traditional dances, festivals, or everyday practices in an open context, and the AI defaults to Japan.

Over and over again.

Here is the twist nobody expected.

This bias doesn't come from raw pre-training internet data.

The researchers tracked where the obsession forms. It emerges after pre-training, during the supervised fine-tuning and alignment phase when humans teach the AI how to behave.

Why Japan?

Because decades of global soft power, rich cultural export, and clean, universally admired digital archives make Japanese culture uniquely "safe" for AI safety filters to lean on.

When labs train models to be harmless and universally pleasing, the AI defaults to the cultural equivalent of comfort food.

It avoids controversy by talking about anime, sushi, and tradition.

[NullException @Youtae]

예측 모델의 저점에 닿았네. 이제 화마가 지나간 자리에 새싹이 돋을 차례네. 만약 더 밑으로 가면 정말 아비규환이 될거 같네.

[NullException @Youtae]

모델 크기 변화를 볼 수 있는 시각화 자료.
갑자기 팡 커지는데 중국에서 무슨 변화가 있었던 걸까?

[Ben Dicken @BenjDicken]

Visualizing open-weight model sizes.

Would be awesome if the big labs shared these details more openly.

I'm curious how much intelligence gain comes from model size vs other aspects of R&D / tuning.

[NullException @Youtae]

역~쉬 칼들을 갈고 있었네. 게다가 그분이 한마디 하시니 바로바로 나오는군.
https://t.co/CHjoI5GO35

[Qwen @Alibaba_Qwen]

Qwen3.8 is launching and going open-weight soon!🌐

With a massive 2.4T parameters, this model is continuously evolving. We believe it’s one of the most powerful model available today, compatible to leading frontier AI models , second only to Fable 5.

You don't have to wait to test it. Just now, the Qwen3.8-Max-Preview made its debut on Alibaba’s Token Plan, Qoder, and QoderWork. Be among the very first to try it out.

Can't wait to hear what you build. Stay tuned! 🚀 

Token Plan
international：https://t.co/YRvcGdB9Bv
China：https://t.co/PKMUNwUuRp

[NullException @Youtae]

kimi k3로 사람들이 예제들 올리기 시작하네. 그런데 하나같이 놀라고 있네. 저작권 개나줘라식으로 만들어주네. 완전 똑같이 만드는 것 같네. 오픈웨이트에도 이렇게 되면 난리나겠는걸.