SteddiPace

Someone who shares interesting AI discoveries. Coding at a steddipace.

[SteddiPace @steddipace]

@Yuriko8866 @onijiang00 那是因为国际象棋有困毙算和棋这种规则，你在中国象棋也这样规定，你看没有老将车马能单杀对面不。

[SteddiPace @steddipace]

The safeguards of Fable 5 are almost unreasonable. Even not setting aside cybersecurity and biology, it can block requests just or handling sentences that contain swear words.

They should lift the block for some relatively safe scenarios. Allowing it to follow the user's request to output swear words would probably not ruin the world.

[SteddiPace @steddipace]

Years ago we paid the same price and enjoyed better and better hardware.

Now we pay the same price and get the better and better AI models.

[SteddiPace @steddipace]

When I used the formerly popular tool, Antigravity for coding. I found that Claude would often change the test to fit the source code even when the bug was actually in the source code itself.

It simply modified the test and said, " Congratulations! All the tests pass," without explaining what it has done.

Only when you asked, "I saw some tests failed, so what happened?" Did it finally admitted normally that it changes the test to fit in the source code as if nothing is wrong.

I built a small tool to insert comments at the top of each file, telling Claude not to change the tests just to make the source code look correct. But it didn't work well. Claude still kept ignoring them and made a lot of mistakes when writing tests.

Do you have similar problems when coding?

[SteddiPace @steddipace]

Right now , eight of the top 10 most-used model on OpenRouter are Chinese models. All of the top five are Chinese models.

[SteddiPace @steddipace]

For Minecraft playes

What mod do you often use to make your life easier when you're trying to beat the game?

I have tried so many times to beat the game directly. Although I nearly succeeded a few times, I eventually give up.

Once I was lucky enough to enter the end. I accidentally died there. Later, when I tried to get back to the battlefield. I found myself trapped on the small obsidian platform without a pickaxe. End stone is so hard to mine by hand that eventually I gave up.

After that, I have never managed to reach the End again.

So beating the game is so hard for players like us. What mod do you usually use to make it easier?

[SteddiPace @steddipace]

Then will GPT 5.6 get permission to be released soon?

[Anthropic @AnthropicAI]

We’ve received notice that the Department of Commerce has lifted export controls on Claude Fable 5 and Mythos 5.

We'll begin restoring access tomorrow, and will share an update soon.

We’re grateful to our users for their patience, and to everyone who worked with us on redeploying the models.

[SteddiPace @steddipace]

The only real way to confront the job crisis caused by AI is to reduce working hours and hire more people. Then everyone would benefit from it.

[SteddiPace @steddipace]

Today I found an effective prompt for getting Codex to teach you how to code.

"Imagine an engineer staring the screen, wanting to [describe your goal here]."

"Then what should they think next?"

Codex will then give you a step-by-step guide.

[SteddiPace @steddipace]

Everyone is trying to use AI to build programs, but is anyone seriously considering using LLMs to make code more comprehensible？

[SteddiPace @steddipace]

The prohibition is a political issue rather than a technology one, which makes it hard for OpenAI itself to deal with.

[NIK @ns123abc]

🚨 BREAKING: U.S. government will decide who gets access to GPT-5.6

OpenAI will release GPT-5.6 only in a limited preview to a small group of partners.

Sam Altman told staff the government would be "approving access customer by customer."

Commerce Sec Lutnick personally called Altman warning: don't launch without approvals from other agencies.

A de facto licensing regime.

ITS HAPPENING

[SteddiPace @steddipace]

Do LLMs really hit a wall?

Frontier LLMs in the US now seem to be slowing down, while GLM, a Chinese LLM has quickly narrowed the gap.

What can make us delighted is that other lanes such as image and video generation is far from saturated. We may still see a lot of progress in the near future.

[SteddiPace @steddipace]

当你在和AI聊天，感觉对面的AI蠢到想要顺着网线过去抽死它的时候。

可以试试给每一句话都加上喵~，如果对面没有跟着你发，就告诉它，你的回答里面为什么没有喵呢，喵？

听懂了吗各位喵~ 这样试过之后果然就镇静下来了喵~ 马上就心平气和了喵~

（快不认识这个字了）

[SteddiPace @steddipace]

The voice mode of ChatGPT really gave me a pretty bad experience. The instant answers seemed to leave no thinking budget for model, so the answers were shadow. It also failed to follow the clear instructions in chat.

Maybe with today's technology, we cannot release something that can respond seriously, accurately and swiftly at the same time. What OpenAI should do is to add a button that increases the thinking budget at the cost of waiting. I don't think it is unacceptable for users who want a better answer to wait for a while when they finish talking.