Teaching computers to reason while I'm still figuring out how to cook. CS @UAlberta
Joined October 2024
- Tweets25
- Following69
- Followers22
- Likes49
Claude 这周出了 Fable 5.2，学生党先别急着冲，说三点：
Pro 这周就能用 Fable，但它耗额度更快，官方自己都建议日常写代码先用 Opus 5.5。额度重置次数跟账号年限和连续订阅挂钩，刚开 Pro、或者用一个月停一个月的学生，拿到的最少。API 降价 20%，对自己做小项目、调接口的人反而是实惠。
我的结论：日常用 Opus 5.5，真卡住的难题再换 Fable。
你们这周额度够用吗？
1
2
68
"Officials believe" is doing a lot of work here. Anonymous sources, no charges, no public evidence, and Singapore questioned her in 2025 and let her go. Until there's proof, it's an allegation, not a fact.
New details in the China-Jensen Huang spy plot: U.S. officials believe Alice Huang was recruited by MSS, China’s top intel agency. Alice knows other AI execs, like leaders of Nvidia assembler Gigabyte. Her cos have ties to the China military establishment. nytimes.com/2026/11/06/us/po…
9
Claude Fable 5.2 dropped today. As a CS student, I don't care about the benchmark chart. I care about one thing: can it debug my side project without me explaining the codebase three times?
Trying it on a personal project this week. Will report back.
Introducing Claude Fable 5.2 and Claude Mythos 5.2, designed to tackle the most daunting technical challenges.
They’re capable collaborators on ambitious projects in many fields, and show improvements in areas like coding, reasoning, and vision.
2
32
RunzeWater retweeted
Introducing Claude Fable 5.2 and Claude Mythos 5.2, designed to tackle the most daunting technical challenges.
They’re capable collaborators on ambitious projects in many fields, and show improvements in areas like coding, reasoning, and vision.
Your eval checks if a model repeats Beijing's line. Does it check if a model repeats Washington's? Pick 803 prompts on the topics you care about and you'll always find the other side biased. That's not alignment, that's choosing which bias you're comfortable with.
Introducing Arete-27B: Qwen3.8-27B, post-trained to reinforce Western values.
Chinese open models are currently the strongest available, but they ship with state-aligned behavior in the weights. We built an 803-prompt eval across seven value categories, including freedom of expression, democracy, and privacy, and measured how often a model adopts China’s party line.
After fine-tuning on roughly 53 million tokens, Arete-27B's propaganda rate fell from 22% to 1%, in line with the Western models we tested. Its Hermes Bench score shows no significant performance tradeoffs.
The values in an open model are one fine-tune away.
Our eval, weights, and training data are available now along with the full blog post: nousresearch.com/arete
1
1
25
补充一点背景：这是 Canadian Tire 的职级“宽带”薪资，同公司其他 Senior Data/Decision Scientist 也是 6.4–10.6 万。之所以现在能看到，是因为安省 2026 年起强制招聘写薪资区间。所以问题不在这一个岗位，而是加拿大传统企业整套 AI 薪资体系本来就这个水平。
加拿大零售巨头 Canadian Tire 正在招聘一名senior AI scientist，起薪为 6.4 万加元（约合 4.6 万美元），封顶不过10万加元（约合 7.1 万美元）
职位要求应聘者“最好拥有博士学位”，并具备“5 年以上开发和应用机器学习（ML）或人工智能（AI）模型的经验”。
这是一个需要一周四天到office的职位，工作地点位于多伦多——全球生活成本最高的城市之一。
1
1
35
RunzeWater retweeted
Introducing Personal Finance in Claude, in beta in the US.
Connect the accounts you choose, and Claude answers your money questions in everyday language, from your own balances and transactions. Access is read-only, and Claude can't move, manage, or invest your money.
I was literally debugging my own code with Claude Sonnet 5.5.
My program couldn't detect when I stepped away from my PC.
Claude apparently couldn't detect the difference between debugging and reasoning extraction. 💀
2
70
RunzeWater retweeted
Introducing Kimi K3.1 Pro: our most capable 3T-class open-weight model.
🔹 1M-token context, 2.8T total / 104B active params (MoE), Multimodal.
🔹 Full-project coding: games in Unreal, Unity & Godot.
🔹 Excels at parametric 3D modeling in Blender & CAD.
🔹 More reliable at finance work: Accurate modeling with hallucination down by 65%.
🔹 Matching the strongest proprietary models across coding, agentic and knowledge benchmarks.
Kimi K3.1 Pro and K3.1 Pro Highspeed (2–3× faster) are now live on Kimi.ai, Kimi Work, Kimi Code, and the Kimi API.
Open weights by Nov 4, 2026.
🔗 API: platform.kimi.ai
🔗 Tech blog: kimi.ai/blog/kimi-k3-1-pro
