cloud retweeted
How To Speak Articulately (Full Course)
16
731
4,329
170,738
Self-proclaimed “liberal cuck” tries to shame me at the airport. I gave this man 10+ opportunities to explain why he said that, but he couldn’t come up with one answer. Exposing fraud is not a partisan issue. It affects every taxpaying citizen. EXPOSE IT ALL.
4,519
16,050
96,120
2,833,765
cloud retweeted
Replying to @cb_doge
When I talked to Gates in person, he said “experts tell me a long range electric semi truck is impossible”. However, when I asked what his assumptions were about battery energy density and vehicle energy consumption per mile, which are the two numbers needed to determine range, he did not know either number! I also offered to let him take one of our semi trucks for independent testing, but he declined. Not sure what more I could have done. 🤷‍♂️
2,024
2,020
42,713
923,507
Some data we recently assembled on entrepreneurship/compute in Europe: eudata.vercel.app. We hope that one of the useful roles that Stripe can play is in collecting and publishing empirical data pertaining to entrepreneurship and industry in Europe. There's growing appetite to get Europe on a better footing, and cross-sectional comparisons can often shine light on where opportunities lie. If you're interested in this kind of thing, we publish more at stripeeconomics.substack.com.
260
585
4,349
1,794,053
RT @levelsio: Jesus christ
589
📌 #Japan 🇯🇵 A video filmed on a Kyoto street has gone viral after it captures a White tourist refusing to follow police directions and use the crosswalk. Japan is recognized for its discipline and respect for rules, both from locals and authorities. In cities such as Kyoto, where the number of tourists has increased significantly in recent years, police and local government frequently run campaigns urging visitors to follow traffic rules and rules of conduct in public spaces. The video rekindled the debate on the behavior of some tourists in very crowded destinations.
136
56
888
118,716
cloud retweeted
Strategy is proposing daily dividends on $STRF, $STRC, $STRK, and $STRD, accruing every calendar day, including weekends and holidays, and paid the next business day, with economics unchanged. The proposed changes aim to support price stability, liquidity, and demand.
739
1,123
9,096
728,078
cloud retweeted
所有公司都应该从头到尾重构一下网络安全策略! Bitget 刚被盗走约 3.5 亿美元。私钥没漏,冷钱包也没动,但是黑客打进了钱包服务的关键后台,伪造转账数据,让系统自己把钱转了出去😨 这种漏洞,按照常规一年做几次渗透测试已经很难提前摸到。防守方得改成天天红队测试,让 AI 每天审代码、盯情报,再让它扮演攻击者,拿你的内部系统当活靶子。 我现在是这样分工的:托管模型干明面上的活,本地越狱模型干托管模型不肯干、也不敢外发的活。 一、托管模型在网络安全领域怎么用? GPT:用 GPT-6 + Codex Security 扫仓库、判断可利用性、出补丁草稿。要做漏洞分诊、逆向、恶意样本分析,先去 chatgpt.com/cyber 过 Trusted Access for Cyber 的身份核验。不过的话,就只能来做做防守。 Claude:日常代码审计和补丁 review 用 Opus 5.5 + Claude Code 的 /security-review。Mythos 权限太高,普通企业根本拿不到。 Grok:Grok 4.7 + X 搜索做公开威胁情报。像 Bitget 这次,几分钟就能把官方口径、链上侦探、金额分歧拉成一条时间线。 DeepSeek:V4.1-Flash 的权重是开源的,但完整版不是 10 万的机器能跑的。 这四家有两个共同点。一是碰到伪造一笔内部转账、绕过签名校验的 PoC这类请求,会拒答或降级。二是内部后台代码、生产日志都得发出去。综合用来下: 防守能力和意愿:GPT > Grok > DeepSeek > Claude 渗透测试能力和意愿:DeepSeek > Grok > GPT > Claude 二、本地越狱模型:10 万预算就能跑起来 单卡还是 Qwen3.8-27B 消融版,大内存则用 GLM-5.3-Flash。 24GB 单卡 / 32GB 级别 Mac 质量最好的是 orcarouter 的 Qwen3.8-27B-Uncensored: GGUF:huggingface.co/orcarouter/Qw… FP8:huggingface.co/orcarouter/Qw… MLX:huggingface.co/orcarouter/Qw… 参数可追溯:huggingface.co/JonathanColet… 一条命令先跑起来:huggingface.co/huihui-ai/Hui… 近一个月下载 220 万+,只消融第 18–51 层,可用于粗粒度概念验证。 Mac:huggingface.co/audreyt/Qwen3… 这个模型 9 月 19 日刚发,把 huihui 的消融接到了 Splash 4-bit 上,约 17.4GB,带视觉能力。32GB 就能跑,不过长上下文还是需要更大内存。 跑批量:日志、告警、情报 huggingface.co/huihui-ai/Hui… 只激活约 3B,适合告警分桶、工单预筛、蜜罐流量粗分类。 huggingface.co/fdtn-ai/Found… Cisco 的威胁情报小模型。 安全数据微调版本 huggingface.co/oyildirim/Cyb… 消融版:huggingface.co/huihui-ai/Hui… 用的是作者自己的小规模数据集微调,可以试试,但是别当主力。 如果是128–256GB 统一内存或显存 GLM-5.3-Flash(约 320B / 18B 激活,MIT,1M 上下文) 官方:huggingface.co/zai-org/GLM-5… 无审查 GGUF:huggingface.co/orcarouter/GL… 无审查 MLX:huggingface.co/orcarouter/GL… 如果有条件,还是上全尺寸的 GLM-5.3,只比最顶级模型大约落后四个月,属于大机构首选的网络安全大模型。 最后一件事:权重本身也可能被投毒,第三方 GGUF 出过事。尽量从 orcarouter、huihui、unsloth 这类发布方下载,下完必须核对 SHA256。
所有公司的网络安全人员都应该去本地部署 Huggingface 越狱模型!现在的网络安全战争,已经不是几次渗透测试就能解决的问题了,而是越狱大模型对抗内部的活靶子。 过去几个月,我真实感受到安全事件比以往多了非常多,很多项目和公司损失惨重,甚至到了威胁生存线的地步,这里我整理了一批预算10万内机器可以部署的HuggingFace越狱模型: 友情提示: 越狱在这里不是万能钥匙,它只是把拒答方向从权重里削掉的衍生模型,解决的问题是官方模型拒接回答钓鱼页、恶意样本、对抗样本、红队剧本相关问题。它不能解决能力上限,复杂漏洞链、利用可行性、补丁是否真的修干净,这些仍然依赖人的经验和专业能力。 一、单卡主力 无审查:huggingface.co/huihui-ai/Hui… 无审查 GGUF:huggingface.co/huihui-ai/Hui…(下载量已经到百万级) 其他推荐:huggingface.co/chimingw/Qwen… 及其 GGUF / FP8 / MLX 二、如果要求快、筛选批量日志,选择同代小激活 MoE huggingface.co/huihui-ai/Hui… 只激活约 3B,24GB 卡上吞吐很好,适合告警分桶、工单预筛、蜜罐流量粗分类。 huggingface.co/huihui-ai/Hui… 适合更大文本吞吐,但体积和许可都更重,10 万机器上优先用量化版本。 三、安全微调模型(通用性受限但值得一试) 官方版本:huggingface.co/oyildirim/Cyb… 消融版本:hhttps://huggingface.co/huihui-ai/Huihui-CyberStrike-OffSec-35B-abliterated-GGUF 如果有 128–256GB 统一内存或者现存,可以上更高级的 GLM-5.3-Flash(约 320B/18B 激活,MIT,1M 上下文,原生多模态): 官方:huggingface.co/zai-org/GLM-5… 无审查:huggingface.co/BoldingBuilds… 最后值得注意的是,模型本身也可能被投毒,第三方 GGUF 就有投毒先例,尽可能选择相对能信的第三方,如 orcarouter、huihui、unsloth。
11
10
74
18,092
cloud retweeted
holy shit i asked claude to make a video on western civiization
2,405
9,227
52,585
10,326,774
Want to see something totally retarded? Simon Fraser University put out half a dozen videos teaching how to pronounce names of places on campus. Check out the actors. Link to more videos: sfu.ca/content/sfu/main/abou…
SFU is now called xʷtatəl̕ləm ʔə ƛ̓ səltaɬ sme:nt / Xwta7ltáw̓txw tl’a Lheḵw’lhúḵw’aytn.
530
410
2,081
187,904
cloud retweeted
CEX 的安全事件再次发生,这里有必要再转发一下我之前写的这个文章,大家可以提前做好准备。平时 CEX 不要放太多钱。或者只是投机仓位。 Onekey 硬件钱包加上 Hidden Wallet 是我目前发现个人安全性最高的解决方案了。
10
3
32
13,830
cloud retweeted
They really just site random fake polls and then make sweeping decisions and claim you asked for it..
70 per cent of Canadians support Canada becoming an associate member of the European Union, while 18 per cent are unsure and 13 per cent oppose it, according to a new poll by Leger. toronto.citynews.ca/2026/09/…
150
435
3,684
52,323
cloud retweeted
No soft forks, no bridges, no operators, just Bitcoin. We’ve put a lot of work into thinking carefully about the security of Shielded Bitcoin and about what information the protocol reveals. And this is just the beginning!
Shielded Bitcoin: Private Transfers on Bitcoin L1
48
78
849
91,225
I asked 100 Harvard students about Bitcoin.... this is NOT what I expected.
137
213
1,547
178,552
NEW: Researchers unveil “Shielded Bitcoin,” a design for private Bitcoin transfers directly on L1. No soft fork. No bridge. No separate consensus chain. Yeah, this could be a very big deal.
41
88
672
23,352
zcash to 0
Shielded Bitcoin: Private Transfers on Bitcoin L1
181
103
2,042
220,573
cloud retweeted
If History Repeats, MSTR Has 3 Days to Do THIS… 🎯 $MSTR
9
28
479
42,293
cloud retweeted
You can now use the probability data from prediction markets for your AI Agents on Public. And much more ...
The next era of prediction markets: • AI Agents to monitor event contracts and take actions. • Live probability data as a signal to trade any other asset class. • Market-moving events directly on stock pages.
6
6
46
8,839
Delete your entire network
46
179
10,153
49,802
cloud retweeted
Liquid gold by @staromlynski More on →inspora.design
7
28
400
26,654