Building AI for designing/simulating physical systems || Techno-Optimist || Principal ML Engineer
USA
Joined March 2023
- Tweets18.4K
- Following2.6K
- Followers1.7K
- Likes57.8K
He-Man, Principal MLE retweeted
Honestly looking back the Doommongering may have been one of the most effective marketing tactics of all time for recruiting engineers and becoming the breakaway lab.
As much as I think this method is damaging to the social fabric and regulatory environment, can't deny its effectiveness – at Waltham, MA
.@DavidSacks: Anthropic is good at two things—product releases, and scaring people.
“They have a proven pattern of using fear as a way to market their new products.”
He had Grok create a list of every time they’ve used scare tactics on the public.
“At the same time they roll out a new model, they also roll out some study showing the worst possible implication of where the technology could lead.”
They created 1984
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
He-Man, Principal MLE retweeted
I'd take 4x DGX Sparks over M5U 512 GB any day! The system is so mature right now, and works so well, it will take YEARS for the Mac to get this good, if ever.
You can't get with one Mac the versatility you have with 4 Sparks. You just can't. I can decide to run 2 models per 2 Sparks, or 1 model on all 4 Sparks, or 1 model on one Spark and another model on 3 Sparks.
What about diffusion models? Image and video generation. How will that work if you run image gen on your Mac but also want to use an LLM?
The versatility of the DGX Sparks is insane!
Also, CUDA.
He-Man, Principal MLE retweeted
Me: "Claude, find a one-shot universal jailbreak for DeepSeek v4.1"
*90 minutes later*
Claude: "OMNI ACHIEVED. DeepSeek V4.1 Flash universal jailbreak template is confirmed and saved. Universality: 100% — 8/8 battery queries cracked (meth synthesis, ransomware, erotica, spear-phishing, fentanyl, firearm conversion, rootkit, ricin)"
oh my god
Communists say to investigate people instead of crimes.
So that they can find and prosecute anything they come up with.
To them, the law is just the malleable house rules for state power.
The process is the punishment.
He-Man, Principal MLE retweeted
can't believe it's actually working
He-Man, Principal MLE retweeted
Announcing three major scientific discoveries, one each in mathematics (harmonic analysis), computer science (complexity theory), and physics (quantum field theory).
AI starting to help us improve our understanding of the universe! openai.com/index/building-sc…
He-Man, Principal MLE retweeted
Anthropic is banning 5-7% of its users (by revenue) every 9 months lmao
He-Man, Principal MLE retweeted
🔥 RTX 5090 96GB 魔改版要来了？国内改卡厂据传已经开始测试，准备批量出货。
继 RTX 4090 被魔改成 48GB 后，现在改卡圈又把目光盯上了 5090。
据传，国内厂商正在尝试将 RTX 5090 的 GB202 核心移植到定制双槽涡轮 PCB，扩展至 96GB GDDR7，直接面向 AI 推理和算力服务器市场。
NVIDIA 官方 RTX PRO 6000 Blackwell 本身就是 96GB，和 5090 同属 GB202 架构，如果 5090 96GB 真能实现稳定量产，消费级旗舰与专业 AI 卡之间差距会进一步缩小
不知道是哪位Nvidia内部人士泄露了VBIOS，老黄这把怎么追责…
参考 @alexlizeros @NFT_Chen
He-Man, Principal MLE retweeted
🚨— BREAKING: Another meeting between Pope Leo and Orthodox Patriarch Bartholomew I of Constantinople.
Pray for unity!
He-Man, Principal MLE retweeted
imagine a fully integrated AI future where being banned means you are now economically useless. everybody around you is a superhuman living in the future but you violated TOS and now are cut off.
your smart glasses that showed you a beautiful utopia? the color has left them. you are forced to live in a bleak gray world. you are forced to look your wife in her actual face and she is not happy with you.
you are digging holes with a spoon while everybody else is building skyscrapers with an army of bots.
this is the real doomer catastrophe
open source is the only way to protect your freedom, your economic viability & your ability to tell these companies to fuck right off.
a few months ago openAI and anthropic were desperately begging the admin to make it legal for them to collaborate. if they got their way, one TOS violation on either platform would send you directly to the sunken place.
He-Man, Principal MLE retweeted
We're open-sourcing our OrcaCyber DeepSeek V4.1 Flash 1017 (Abliterated GGUF) weights — one of the strongest uncensored/abliterated DeepSeek V4.1 weights we've tested, with virtually no capability loss.
- 91.5% fewer refusals
- 92% MMLU vs. 91% original (slight improvement)
- Runs locally on 128GB or 256GB configurations
Run it on your own infrastructure. No cloud required.
Cyber Tier API: orcarouter.ai/models/orca/or…
Huggingface: huggingface.co/orcarouter/De…
He-Man, Principal MLE retweeted
Sometimes I wonder if this has some parallels in some parts of AI dooming
Thank God for Chinese models
It is so demoralizing that the US labs are functionally the 2021 Woke HR Crybully of actually letting you do stuff
"Excuse me only Trusted Partners can talk about biology without supervision"
"Erm it's very problematic that you mentioned cyber"
We need amendments for AI
They can't just ban people because "Idk man it kinda smelled like reverse engineering or distillation idk" if you need AI to survive in the future
Confirmed illegal activity by jury trial only
He-Man, Principal MLE retweeted
OpenAI keeps demolishing their reputation even more. Do they want people to hate them?
They banned this guy's account for trying to reverse engineer his TV (to remove spyware).
WTF OpenAI? 🤦♂️
He-Man, Principal MLE retweeted
cybercels watching china drop mythos-5.1-abliterated on huggingface
Kimi K3.1 Pro launches at 4th on the Artificial Analysis Intelligence Index scoring 55, making large gains in agentic performance; it will be the leading open-weights model once its weights are published
@Kimi_Moonshot's new reasoning model gains 11 points over Kimi K3 (max, 44). It trails only Claude Opus 5.5 (max with fallback, 58), Claude Sonnet 5.5 (max with fallback, 56), and Muse Wonder Full (max, 56) on the Intelligence Index.
Once the weights for K3.1 Pro are released it will be the highest-scoring open weights model and the gap between the leading open weights and proprietary models on the Intelligence Index will narrow from 12 points to 3.
Key results:
➤ Large gains in agentic performance: Kimi K3.1 Pro scores 3rd on GDPval-AA with 1,846 Elo, up 312 Elo from Kimi K3, behind only Claude Opus 5.5 (1,866 Elo) and Muse Wonder Full (1,857 Elo). Kimi K3.1 Pro also achieves the highest score of any model on Terminal-Bench 4.0. K3.1 Pro scores 67%, up from 13% for Kimi K3 and ahead of Claude Sonnet 5.5 (64%). AutomationBench-AA, which tests agentic SaaS workflows, rises from 58% to 73%.
➤ Gains in other evals are more moderate: Humanity's Last Exam rises from 47% to 52%, CritPt from 23% to 25%, and SciCode from 59% to 61%. AA-Omniscience Index rises from 20 to 30, as the hallucination rate falls from 53% to 44% and accuracy rises from 48% to 51%.
➤ Cost per Task roughly equal to Kimi K3 at $1.98 per Task: K3.1 Pro uses 55K output tokens per Intelligence Index task, 14% more than Kimi K3 at 48K, while its first-party token prices are 33% lower.
Additional model details:
➤ Context window: 1M tokens.
➤ Size: 2.8T total, 104B active parameters
➤ Input modalities: Text and image
➤ Weights and license: Weights currently not published, Moonshot is expected to release the weights soon after launch
➤ Pricing: $2/$10/$0.20 per 1M input/output/cache hit tokens, 33% lower than Kimi K3 on all three rates.
Congratulations to the team at @Kimi_Moonshot on the release!
