LLM enthusiast
Joined January 2025
- Tweets7.9K
- Following305
- Followers228
- Likes34.4K
Curline Zephirin retweeted
Fable 5.5 its over for video and image models 🤯
THIS IS PURE CODE
one shotted this video holy shit , how this is more beautiful than ai video models
Two ways to check this yourself:
Ask: “Where did @karpathy go in 2026?” with web search and memory turned off. Only the new Fable knows the answer. Then give it the prompt.
Run the same prompt I used for the test (it’s in the screenshot) and see for yourself.
Don’t rely on Grok for basic reasoning like this. Grok doesn’t have up-to-date knowledge of niche stuff. Either join the community or run the test yourself.
Also, a knowledge cutoff doesn’t mean a model knows everything before that date. Sometimes a model doesn’t know about an event even if it happened before its cutoff. This test shows that too.
Run the same prompt on regular Opus 5.5 and the quality drops. On Fable 5.1 it’s worse.
emmm
GPT焚决35刀获得500刀套餐
看到有同行把 ChatGPT 中转站底层的低成本号池玩法捅出来了，顺手帮大家拆解一下这套逻辑。说白了就是卡 Google Play 的内购结算与宽限期 Bug。
如果只是个人想长期白嫖高配额度，按这个路子走，每个月大概 35 刀就能一直用上 500 刀档位的套餐。
一、 核心机制：差价按小时折算 + 扣款宽限期
ChatGPT 在 Google Play 上的内购升级逻辑和 Claude 不一样。Claude 内购升级是按你之前付的钱去折算延长订阅周期，而 OpenAI 在 Google Play 上的升级补差价公式是：剩余时间占比 × (新套餐价格 - 当前套餐价格)
怎么操作？
1.先通过 Google Play 内购，开一个 8 刀/月Go 套餐。
2.等到第 30 天 23 小时，也就是距离过期只剩最后 1 小时的时候，在 App 内升级到 500 刀的套餐。
3.此时补差价只算最后 1 小时的钱：1 / (24 × 30) × (500 - 8) ≈ 0.683 刀（哪怕留两三个小时也就 2 刀左右）。
4.一小时后套餐到期自动续订 500 刀，但只要你绑定的卡里余额不足，扣款就会失败。
5.重点来了：Google Play 扣款失败后不会立刻掉订阅，而是会给一个至少 7 天（最高 14 天）的账单宽限期（Grace Period）。在这至少 7 天里，你依然拥有 500 刀套餐的完整权限。
二、个人低成本无缝循环方案
既然单号能卡出 7 天的高配使用期，手里备 4 个号就能排班覆盖整月。
1.极致省钱版：4 个号轮流用 8 刀套餐去卡，每隔 7 天操作一个号。月总成本 (8 + 0.683) × 4 = 34.73 刀，相当于 35 刀包月用 500 刀套餐。但因为底子是 Go 套餐，没有 Codex 额度，吃不到 Plus 的重置卡。
2.满血 Plus 版：底子换成 20 刀的 Plus 套餐去卡。月成本 (20 + 0.667) × 4 = 82.67 刀。除了无缝用上 1 个月 500 刀套餐 + 3 个月 Plus，最爽的是每次官方按人头下发重置卡时，普通 500 刀用户只拿 1 张，你手里 4 个号能领 4 张。因为重置卡有效期一个月，而你的单号周期刚好跨月，4 张卡全都能吃满。
三、中转站是怎么靠这个印钞的？
基础的批量起号、接码、防关联（纯净 IP + 指纹环境）对做中转的人来说都是基本功。只要把号养过 29 天进入收割期，利润率非常夸张。直接算笔账：
1.额度产出：1 个 500 刀账号的额度大约顶 50 个 Plus，周额度约 3000 刀，月额度约 12000 刀。如果按上面 Plus 底子配 4 张重置卡的玩法，单月能洗出约 24000 刀的额度。
2.硬性成本：套餐费 82.67 刀（约 555 元人民币）+ 几乎可以忽略的起号成本 + 站点服务器/推广费 + 行内防 DDoS 的保护费。
3.实际营收：24000 刀额度在中转站通常按 1:1 标成 24000 元人民币卖。哪怕按行内非常卷的 0.05 倍率（按 1:7 汇率折算下来相当于打折甩卖），这批额度卖空能收 24000 × 0.05 × 7 = 8400 元。
这还只是 4 个号一组的静态算法。中转站真正的暴利在于资金池与超卖，用户充值越多、消耗越快，他就滚得越快，这也是为什么各种低价中转站都在疯狂打广告拉客。
四、为什么这个漏洞一直没堵死？
1.这套玩法其实一年前就在圈子里流传了，早就有团队配合 Team 计划用这个漏洞搞大规模模型蒸馏，也早有人向官方举报过。但 OpenAI 想堵死它非常难，因为这涉及 Google Play 全球各地区底层的订阅结算与宽限期政策，不是改几行后台代码就能一刀切的。
2.最后提个醒：看懂中转站的暴利逻辑就行了，千万别自己去搭站卖额度，国内搞这个属于违法重灾区。自己拿两三个号直登自用基本稳得很（套反代目前不确定稳不稳），低调自用即可。
1
22
Curline Zephirin retweeted
Fable 5.2 vs. GPT6-Astra painting the Starry Night.
Curline Zephirin retweeted
Neoclouds have limited cybersecurity. Next time agents successfully go rouge, they'll try taking over a neocloud to run more copies. This is bad.
Thus: neoclouds should greatly strengthen their cybersecurity and every company with strong cyber models should help with that.
Curline Zephirin retweeted
The @FireworksAI_HQ incident was pretty bad. Attackers targeted weights and managed to exfiltrate parts of it.
Here's the email we received.
Is this the first public incident where model weights were targeted?
Curline Zephirin retweeted
I gave Fable 5.2 a PDF of one of Leonardo da Vinci’s 500-year-old notebooks and it made his inventions real for me to interact with.
Such an exciting time for education. Anything from history that’s been stuck on a page can now be something you explore.
Curline Zephirin retweeted
We wanted to do more for the people who’ve been with us the longest. So with Fable 5.2, the longer you’ve been subscribed, the more Fable resets you get.
Curline Zephirin retweeted
Claude Fable 5.2 and Claude Mythos 5.2 are available now:
Curline Zephirin retweeted
🚨 OPENAI = FUCKED by META. Muse Ember costs exactly the same as Luna and mogs it BRUTALLY. 36 vs 28 on Artificial Analysis and more than double on Terminal-Bench. OpenAI loses the cheap segment (ember >> Luna) and the premium segment (opus >>>>> Astra) now. The only usable model they have rn is 6.1 Sol, slow af and just capable enough
It must be Dario Amodei, most likely because of Mythos 5.
President Trump reveals the DOJ warned the head of a major AI company "get ready to go to jail, please give your wife a kiss goodnight...do you like the color orange?"
Asked why he isn’t worried that AI builders say there’s a 10% chance it kills humanity:
TRUMP: “The DOJ. We had a situation where somebody was reported to us to be doing something bad with AI. We called them very strongly and said, get ready to go to jail. Please give your wife a kiss goodnight.”
Interviewer: “Who was this person?”
TRUMP: “I can’t tell you.”
Interviewer: “Or this company?”
TRUMP: “I can’t tell you. But we said, do you like the color orange? Because get rid. It said, go home right now. See your wife and children. Give them a kiss. Good night. And that’s the story.”
Interviewer: “If the Democrats wanted...”
TRUMP: “Because tomorrow morning you’re going to jail.”
Interviewer: “If they wanted to make a deal to put that to...”
TRUMP: “Let me tell you, this person totally changed everything. Totally. Yes, sir. Yes, sir.”
Interviewer: “Was it a leader of a major AI company?”
TRUMP: “One of the big ones, yeah.”
Interviewer: “Okay.”
TRUMP: “One of the big ones.”
1
1
86
