plus ça change, plus c’est la même chose
Earth, Solar System, Milky Way
Joined June 2024
- Tweets6.8K
- Following116
- Followers32
- Likes15.6K
graphmanic retweeted
Was reading the score-centering RL paper to add some notes on it the my book, and they have a great table in the appendix on the different importance sampling based train-inference mismatch methods in RL. These have been the biggest focus of RL algorithm research since the GRPO takeoff, imo.
graphmanic retweeted
Progress on ProgramBench is crazy! When it was launched, every model had a flat 0% score and we were worried enough to introduce weaker metrics ("almost resolved" etc.), but things have certainly changed!
graphmanic retweeted
Open models will be used for cyber attacks, just like closed models have been. At the same time, open models aren’t the super high leverage tool magically increasing cyber crime by 100x — those bottlenecks lay elsewhere.
Was I wrong about open models & cyber risks? I don't think so. The AI community is focusing on the wrong lens - which type of model is causing harm - rather than how cybersecurity maintains an equilibrium, who enacts harm, and how they use new tools.
interconnects.ai/p/was-i-wro…
graphmanic retweeted
Was I wrong about open models & cyber risks? I don't think so. The AI community is focusing on the wrong lens - which type of model is causing harm - rather than how cybersecurity maintains an equilibrium, who enacts harm, and how they use new tools.
interconnects.ai/p/was-i-wro…
graphmanic retweeted
You know what else is only used 5% of the time?
sam: local inference is a terrible allocation of scarce resources
“a lot of people want local inference, and somehow they assume that the problem is solved now because we have open-source models, we're going to be able to run them on our computers, and so we're not going to need all the stuff in the data centre
but from a global allocation perspective, actually it's quite terrible to have everybody, like, have a bunch of the world's limited supply of DRAM sitting in their own computer and being used five percent of the time
you want maximum utilisation in the cloud if you're trying to deal with this very supply-constrained environment
somehow many people's intuition is the opposite of that”
graphmanic retweeted
Customer B has been revealed. Anthropic is SpaceX's largest single customer, overtaking NASA and the Pentagon for the first time. 34.5% of SpaceX's revenue last quarter was from compute-as-a-service to Dario Amodei, and Anthropic has committed to spending up to $120.5 billion through 2029. He will be paying Elon $3.1 billion a month starting in December 2026.
graphmanic retweeted
This is very needed. Talking to models like Claude about anything that has an overlap with an area like legal, healthcare, etc is mind numbing. They’re so desperate to hedge and steer away from grey area that they often are really avoiding the real question.
Meta's @Muse answers questions very differently from ChatGPT or Anthropic.
It cites more sources but uses far fewer words and tables, and doesn't try to ask a follow-up.
This short-form content is designed for dealing with tasks rather than continuing a conversation, and frankly, it's far less exhausting to read. This speaks to a really well designed AI experience and its so different from the base LLMs.
Great study by the AirOps team 👇
gg @alexandr_wang
graphmanic retweeted
中国社区逆向拆解了Claude的封号机制。以下是他们的发现
Claude使用积分制。注册时账号获得初始分数，之后根据行为加分或扣分。低于阈值直接封号。回分很难
影响分数的因素:
注册
> 干净IP是最重要的因素。不一定要家宽，但必须干净
> Gmail已经不是最优选，有权重更高的邮箱服务
> 注册时直接弹KYC的号可以扔了
> 用手机App注册会加分
支付
> 刚注册就订阅会扣分
> 直接上Max会大幅扣分。最好等两天先订Pro
> 支付时弹KYC说明账号在考察期，但不一定会封
> 国外实体银行卡加分。Google Play也行。虚拟卡扣分
使用
> 不要一次把限额用完。宁可两个号轮换也不要一个号撸到底
> Desktop客户端有少量加分。尽量少用网页端
> 手机端也有少量加分
> 开启记忆功能加分。记忆文件越多分越高
IP和设备
> 稳定性是关键。不要跳IP，尽量一直用同一个
> 代理必须全局模式并开启TUN
> 多设备登录大幅扣分。理想状态是一台电脑一个客户端
> 频繁换IP、频繁登录、高并发都会大幅扣分
账号年龄
> 老号有巨大的容错空间。2023年的号明显更高
> 分数随时间增长，但很慢
其他
> Windows和macOS没区别。Linux会扣分
> 时区和语言基本不影响。不会因为北京时间就封号
> 跟Claude说自己是中国人没问题，记忆不会触发封号
> 大量使用Fable会扣分
> 升级Max大概率进入观察期，除非是高权重老号
观察期和封号
> 接近阈值时账号可能进入观察期。典型特征是掉登录
> 观察期继续扣分就会封号
> 高分号可以扛过多次观察期
关注我，拆解你每天在用的服务背后到底怎么运作的
graphmanic retweeted
I feel a little bad ranting at Gemini about the unreleased model, as going through the unofficial channels to get a frontier model out today I hear is a good bit of a mess and definitely not easy. That's why they say patience is a virtue.
graphmanic retweeted
I feel a need to defend the AAv5 index and open models at the same time. Benchmarking is messier, more opinionated, and trickier than most people give credit to.
graphmanic retweeted
Open models' demise has been greatly exaggerated, but the previous AA index did flatter them. As an industry we need more specific intelligence indices, as LLMs are used in vastly different tasks.
My meta-review on the benchmarking industry.
interconnects.ai/p/artificia…
graphmanic retweeted
People underestimate how much money is left to get thrown at sovereign ai
Today we announce the launch of the Giotto family of models. We are unveiling Giotto Alto and Giotto Solo, the smartest European sovereign AI models on the market.
Giotto Alto is our flagship model built for advanced reasoning capabilities, free for everyone and available on the App Store and Google Play. It is hosted on European cloud infrastructure, so your data stays in Europe. Giotto Solo is the smartest model that runs on a single GPU, so organisations can keep their AI on their own infrastructure.
We are a Swiss lab with a European heart and we built this so that Europe can run advanced AI on its own terms.
Use Giotto: cloud.giotto.ai/
Readers added context they thought people might want to know
Giotto's graphic shows self-reported benchmark scores comparing to specific variants of other models; independent leaderboards as of late Oct 2026 list no Giotto and different leading scores on AIME 2026, GPQA-Diamond and HLE.
giotto.ai
benchlm.ai/benchmarks/aim…
benchlm.ai/benchmarks/gpq…
benchlm.ai/benchmarks/hle
graphmanic retweeted
Simplest explanation is just a narrow RL training reliant on their internal infra (which isn't good design for an open weight model)
While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected.
Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions.
Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent.
But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models?
We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem.
Read the full report: cohere.com/blog/concerning-b…
graphmanic retweeted
My first open model lecture, covering all the many pieces that build the ecosystem, and the open questions/debates determining where it goes next.
How does distillation fit in?
How do we manage risks?
How do we see access to models changing in the future?
I'm excited to revisit these questions about every 6 months, to give a state of the union on the field.
Chapters
00:00 Introduction
01:36 What is an open model?
08:47 Why should we care about open models?
18:52 How competitive are the best open models today?
25:28 How are open models built? The role of distillation
34:52 Are people actually using open models?
41:47 What comes next with open models?
47:32 Conclusions
What questions did I not answer that you have?
graphmanic retweeted
A Chinese job-recruitment app just released an open-weights 33B model that beats Qwen 27B on most agentic benchmarks.
Nanbeige is the research team of BOSS Zhipin (think WhatsApp meets LinkedIn).
Strong open models in China now come from labs, cloud providers, quant funds... and job boards. China's open-weights roster keeps getting deeper.
🚀Excited to announce Nanbeige4.3-33B!
To push the performance ceiling of small agents, our strategy in Nanbeige4.3 starts with building a powerful teacher first.
Nanbeige4.3-33B delivers SOTA agentic performance among open models at ~30B parameters:
* Strong results on agentic coding and cowork benchmarks.
* When integrated into mainstream agent harnesses, it tackles real-world tasks with cost-efficient inference.
* It will serve as a teacher for upcoming smaller Nanbeige models, starting with Nanbeige4.3-3B.
huggingface.co/Nanbeige/Nanb…
huggingface.co/Nanbeige/Nanb…
#LLM #Agent
graphmanic retweeted
Bypass Paywalls Clean new release v4.4.8.0
+ The Economist
v El Confidencial
v Funke Medien
v Gannett Group/USA Today
v LaNacion-com-ar
v Nw-de
v The Columbian
v The Daily Beast
v The Diplomat
& more
Chrome: gitflic.ru/project/magnolia1…
Firefox (delayed): gitflic.ru/project/magnolia1…
graphmanic retweeted
I've been looking into public benchmarks and am astounded by the poor quality vs. how much we rely on them as an industry.
@xeophon pointed me to this excellent site epoch.ai/benchmarks/ which is an independent third party which scrutinizes benchmarks and "looks at the data" 💪
For example, here are some popular benchmarks we all rely on. Before trusting a benchmark its worth running them and inspecting examples the grader failed carefully, to get a sense of the quality.
I'm continuing to look more into this, and will write something up eventually once I've collected more data, but I think this is an important subject.
graphmanic retweeted
Interconnects Podcast #20: We got together to debate hard questions and messy realities after a crazy, somewhat intimidating, week in the AI world:
- Why did Kimi K3.1 Pro have a mixed reaction with such incredible benchmarks? Will it be adopted?
- Were we wrong about the risks of open-weight models and cybersecurity?
- Why do cyber experts discuss recent events so differently than AI experts?
- What comes next in the few months? What will change our minds?
Chapters:
00:00 Kimi K3.1 Reaction
08:15 Kimi’s open-weight cyber principles
11:23 You can hack everything era
15:54 PHP2Shell, LiteLLM, and the new internet
24:00 Abliterated APIs, safeguards, and trusted access
31:41 Anthropic not hiring juniors employees
36:16 Florian’s Muse addiction
45:30 Wrap-up
More of these soon!
graphmanic retweeted
The complete reasoning-from-scratch video series is now up (and linked) here: magazine.sebastianraschka.co…
(Taking a short break but will be back with LLM model and research updates in November!)
