financialraen retweetledi
financialraen
280 posts
financialraen retweetledi
your startup is just an openai wrapper
openai is just an intel wrapper
intel is just an intel foundry wrapper
you should invest in intel
zach@blip_tm
your startup is just an openai wrapper openai is just an nvidia wrapper nvidia is just a tsmc wrapper tsmc is just an asml wrapper asml is just a sand wrapper
English
Bro you won't believe what anthropic and open AI does
BridgeMind@bridgemindai
You're paying $1.74/$3.48 per million tokens for DeepSeek V4 Pro to train on your prompts. Read that again. You pay them. They use your code to make their model better. OpenRouter's privacy guardrails literally block it by default. Ranked #14 on LMArena Code. Trains on your data. And you're paying for the privilege. This isn't an AI model. It's a data collection service you pay a subscription for.
English
Cursor is Elon's first purchase, which is a huge mistake.
A coding agent harness is now open source (see Codex and Claude Code). The current design works virtually perfectly, so there's no need for a fundamentally new design.
One can say that agentic coding is solved.
One might argue that he bought Cursor for users, but users, as we know, switch between coding IDEs frictionlessly, so it's not like Twitter, where you cannot leave without losing followers.
Cursor is an emperor with no clothes. A shell without substance. You only know it exists because it's been there before most others.
English
@YannickCrypto @dataalways No but opus 4p7 is really good with mev and 4p6 good at smartcontract hacking decompiled bytecode
English
Meet 0x841: a fresh sandwich bot lurking in the dark forest--currently attacking about 1500 users per day, while going almost unnoticed. What's particularly scary: most of the victims are using private mempools and should be protected.
English
@dataalways Seems like one of the smartest sandwichbots ever.
English
@PhiloGroves Same here and the frustrating part is nobody from OpenAI replies to help
English
I am cyber-verified for OpenAI and an authorized vulnerability researcher for Apple's security program, performing source code analysis on my own system...
And still get hit with this.
English
I think my scenario of OpenAI being ahead by 1-3 months by end of year is more likely after this launch
Before the launch I thought GPT-5.5 would be slightly worse than Mythos overall, but tied in some benchmarks, while costing $60-90/M output
So performance-wise it's on target, but price-wise it's much lower at $30
Anthropic is now stuck on a 4.16X more expensive model that takes longer to respond and train
Not only that but their reasoning RL is also worse as evident on math and hard scientific problems (Opus 4.7 just caught up to GPT-5.2/5.3)
I think it's quite likely now that OpenAI will out-accelerate Anthropic
Lisan al Gaib@scaling01
After some deliberation I think GPT-5.5 is close to Mythos despite being only ~1/5 to ~1/2 the size* VendingBench-2 results are good ARC-AGI results are good FrontierMath 4 results are good CritPt results are very good (sadly no Mythos results to compare) CyberGym results are insane TerminalBench 2.0 results are insane UK AISI cyber range results are insane (these 3 results are all on par with Claude Mythos) but for some reason it absolutely shits the bed on SWE-Bench Pro, which threw me off but should just be discarded as noise or spiky intelligence *I think GPT-5.4 is ~1-2T GPT-5.5 is ~2-5T Mythos is ~10T Mythos pricing does look kind of ridiculous at $125 Mythos might turn out to be Anthropic's GPT-4.5 moment
English
@bridgemindai @RazorSilenzio Read policy first open AI trains on your data
English
Anthropic, OpenAI da aynısını yapıyor ama sana söylemiyor. Tüm AI şirketleri böyle. DeepSeek en azından dürüst davranıp açıkça söylüyor.
BridgeMind@bridgemindai
You're paying $1.74/$3.48 per million tokens for DeepSeek V4 Pro to train on your prompts. Read that again. You pay them. They use your code to make their model better. OpenRouter's privacy guardrails literally block it by default. Ranked #14 on LMArena Code. Trains on your data. And you're paying for the privilege. This isn't an AI model. It's a data collection service you pay a subscription for.
Türkçe
@bridgemindai @RazorSilenzio Alright Albert, this not api but read policy it is used to train
English
@theo @solanapoet Hmmmmmmmmm😧, it would be fun if the solution was a crypto wallet with $100 in it then multiple micro puzzles🤔 think about it @theo
English
@Defi_Scribbler We have ai that can do some of that exceptionally well
English
If you want to make your FACELESS YouTube automation journey easy, do these:
> hire a scriptwriter $50-$300 p/script
> hire a voice actor $100-$400 p/voiceover
> hire a video editor / motion designer $150-$500 p/video
> hire a thumbnail graphic designer $10-$150 p/thumbnail
total budget per video = $300 - $1,300
there’s no guarantee you’ll get returns on your investment per video but it will be easy
English
financialraen retweetledi
普林斯顿博士生 Yifan Zhang 更新了 DeepSeek V4 的技术细节。他 4 月 19 日预告「V4 下周」并列出三个架构组件名称，今晚给出了完整参数表，同时首次披露存在一个 285B 参数的轻量版 V4-Lite。
V4 总参 1.6T。注意力机制为 DSA2，组合了 DeepSeek 此前在 V3.2 中使用的 DSA（DeepSeek Sparse Attention）和今年初论文提出的 NSA（Native Sparse Attention）两种稀疏注意力方案，head-dim 512，配合 Sparse MQA 和 SWA（滑动窗口注意力）。MoE 层共 384 个专家，每次激活 6 个，使用 Fused MoE Mega-Kernel。残差连接沿用 Hyper-Connections。
训练端首次披露的细节包括：优化器用 Muon（一种将 Newton-Schulz 正交化应用于动量更新的矩阵级优化器），预训练上下文长度 32K，强化学习阶段用 GRPO 并加入 KL 散度校正。最终上下文长度扩展至 1M。模态为纯文本。
Zhang 不在 DeepSeek 任职，DeepSeek 官方未对上述信息做出回应。
中文
Chinese anthropic
Zixuan Li@ZixuanLi_
Truly sorry for any confusion or frustration caused by unclear, misleading, or inappropriate rules in our moderation system and on our pages. OpenClaw, Hermes, and SillyTavern are now explicitly marked as supported under the GLM Coding Plan. Other general-purpose tools will be analyzed on a case-by-case basis. A gentle reminder: please do not share your account or use your subscription as an API. If you follow the rules and still encounter 1313 errors, please reach out to us at user_feedback@z.ai.
English
financialraen retweetledi
Anthropic said Mythos was too dangerous to release. Then four random guys in a Discord gained access on day one by guessing the URL...
This is pretty insane:
→ Group in a private Discord guessed the endpoint from Anthropic's naming conventions
→ They figured out the conventions from the leak in the Mercor breach three weeks ago
→ Used a contractor's legit eval credentials to walk in
→ Have been using it ever since to build simple websites
The AI that finds zero-days in every operating system on earth was defeated by address bar autocomplete... big yikes
Bloomberg@business
Anthropic's Mythos has been accessed by a small group of unauthorized users, raising questions about control of the AI model bloomberg.com/news/articles/…
English
financialraen retweetledi
🚀 Meet Qwen3.6-27B, our latest dense, open-source model, packing flagship-level coding power!
Yes, 27B, and Qwen3.6-27B punches way above its weight. 👇
What's new:
🧠 Outstanding agentic coding — surpasses Qwen3.5-397B-A17B across all major coding benchmarks
💡 Strong reasoning across text & multimodal tasks
🔄 Supports thinking & non-thinking modes
✅ Apache 2.0 — fully open, fully yours
Smaller model. Bigger results. Community's favorite. ❤️
We can't wait to see what you build with Qwen3.6-27B! 👀
🔗👇
Blog: qwen.ai/blog?id=qwen3.…
Qwen Studio: chat.qwen.ai/?models=qwen3.…
Github: github.com/QwenLM/Qwen3.6
Hugging Face:
huggingface.co/Qwen/Qwen3.6-2…
huggingface.co/Qwen/Qwen3.6-2…
ModelScope:
modelscope.cn/models/Qwen/Qw…
modelscope.cn/models/Qwen/Qw…
English
financialraen retweetledi
OpenAI just released a new open-source model
it's "a bidirectional token-classification model for personally identifiable information (PII) detection and masking in text"
github.com/openai/privacy…
huggingface.co/openai/privacy…
English