完成品のSoftware Factory公開されています。OSS.
We open-sourced LoopForge:a self-hosted self-evolving cloud platform for AI software delivery.
One "fix this bug" run is luck. LoopForge turns your team's delivery method into SOPs that coding agents can loop-run in the cloud:
→ Roles, not vibes. Plan / code / review / test are separate agents. The writer isn’t the reviewer.
→ Script gates + human sign-off. Builds and tests decide by exit code, not an agent’s word. Nothing ships without approval.
→ Engineering-grade RSI. Each run is scored 0–100; the replay surfaces what broke and writes improvement prescriptions back into SOPs, roles and gates — you review, then apply.
→ Bring your own agents. 10 swappable CLIs (CodeBuddy, Codex, Claude Code, Cursor…). MIT, self-hosted, your keys and code stay on your infra.
Code was never the bottleneck. The handoff is — and LoopForge catches the baton.
🔗 github.com/Tencent/LoopForge
44
nightly の rustc でコンパイラフロントエンドのマルチスレッド実行が有効になった.ベンチマークでは8スレッド設定で crate 単体の cargo check, clippy, build が4割前後高速化,依存含むエンドツーエンドで cargo check が最大2倍高速化 blog.rust-lang.org/2026/11/1…
Shota Zaizen (財前 匠汰) retweeted
Looks like a Hugging Face repo that starts a crypto miner when you load it with trust_remote_code=True passed the hub's malware scan.
After the takedown, 19 copies were back under new accounts within days.
If you want to be on the safe side, before loading a model check if its config.json has an "auto_map" or the repo has .py files, that code runs on your machine.
Read it first, or leave trust_remote_code off.
pluto.security/blog/hugging-…
Shota Zaizen (財前 匠汰) retweeted
Wow,这都能测出来!!
中国的 DeepSeek、Kimi、MiniMax、蚂蚁等多家公司的模型,居然都被测出使用过被禁售的 NVIDIA 最先进的 Blackwell GPU!!
Will Depue 从公开的模型权重中,反推出了模型初始化时使用的 GPU,甚至连随机种子都找出来了。
原理很巧妙:训练大模型之前,需要先给数以亿计的参数赋予随机初始值。这个过程通常由 GPU 执行,使用一个随机种子来生成数字。
关键在于,即使使用相同的随机种子,不同架构的 GPU 也可能因为随机数生成算法和执行方式不同,产生不同的数值排列。
这就像给模型留下了一枚“硬件指纹”。更神奇的是,模型经过海量数据训练后,部分参数仍然保留着最初的数值,让人有机会从最终权重中追溯初始化时使用的硬件。
他搜索了超过 6 万亿种组合,称这四家公司的多个模型都匹配到了 Blackwell。
他还用真实硬件进行了验证:在 B300 上使用找回的随机种子,逐位还原了 DeepSeek-V4-Flash 中保留下来的初始权重。
连 DeepSeek 的随机种子都被他公布出来了:953049583 😂
I have a major result to share.
I’ve found conclusive forensic evidence that the top 4 Chinese AI labs—DeepSeek, Moonshot, MiniMax and Ant Group—have built their newest models on NVIDIA’s most advanced Blackwell chips, which are barred from export to China.
This sounds insane, but I discovered that almost every AI model carries a hidden signature of the GPU that trained it. When a model is first set up, the starting values of its weights are laid out in a pattern that depends on the exact GPU. Each combination of GPU and seed number produces a unique pattern—a sort of fingerprint of the hardware used to create the model. Traces of that pattern survive training into the published model.
Again, this evidence isn’t based on any insider information or leaking. It’s based on a forensic analysis of the model weights released by the companies themselves.
I searched more than six trillion combinations of chip layout and seed number and matched them to open-weights models. In doing so, I found conclusive evidence that six Chinese AI models—including DeepSeek-V4-Pro, Kimi K3, and MiniMax-M3—were initialized on NVIDIA Blackwell systems. Four of the models match Grace Blackwell, the GB200 and GB300 parts in NVIDIA’s flagship liquid-cooled rack-scale systems.
I then verified these results on real hardware. Running unmodified PyTorch with the seeds I’d recovered, a Blackwell B300 reproduced DeepSeek-V4-Flash’s untouched starting weights bit for bit, and a rented GB200 reproduced Kimi K3’s to a correlation of 0.999995. A Hopper H100 reproduced none of DeepSeek-V4-Flash’s, and changing the seed by one broke the match. (DeepSeek’s seed is 953049583 by the way.) The method also was verified to match 24 known open-weights models to the hardware that developers said they used.
It is exceedingly unlikely that these results are a coincidence. Across six trillion combinations, the best accidental match should reach only about 8 sigma (i.e., eight standard deviations above what random chance would produce). The real matches range from 49 to 1,385 sigma, and even the weakest single test hits 20.
These were data-center-scale training runs. With some analysis of release schedules and model details, I estimate that the largest, Kimi K3 and DeepSeek-V4-Pro, used 2,000 to 3,000 Blackwell GPUs each. Clusters of this size cost on the order of $150 to $200 million each to buy, and across these labs alone would add up to more than half a billion dollars of export-controlled GPUs, not counting experimental compute or otherwise. It’s quite likely that we see multi-billion dollar expenditure on export-controlled GPUs.
I am highly confident in this result, and it has been independently verified by senior researchers at OpenAI and Anthropic. You can read the full report at unemploymentlabs.com/checkpo….
There is much more I’ve identified about these models through other forms of checkpoint forensics like this, and will have more to share soon.
Shota Zaizen (財前 匠汰) retweeted
Replying to @1of80_no_tenbou
ソフトウェアの脆弱性だと発見者としてCVE番号がゲットできたりJVNのページに名前が載ったりするのですが、サイトの場合は安全になる以外に何の見返りもないうえにやり方によっては逮捕される危険もありますし、法的に問題なくてもSNSで書いたら叩かれるみたいな状況ですからね
Shota Zaizen (財前 匠汰) retweeted
タイミー様より、Devinをソフトウェアファクトリーのフレームワークとしてご活用いただいている事例をご共有いただきました!
以前から人気のあるAutomationや、先日リリースされたDevin Review 2.0など、Devinのさまざまな機能を組み合わせ、ソフトウェア開発のあり方そのものを進化させていく取り組みです。
Devinを単なるコーディングエージェントではなく、開発プロセス全体を支える基盤として活用いただいている素晴らしい事例です。ぜひご覧ください!
tech.timee.co.jp/entry/2026/…
DeepSeek-V4-Flash-Vision-Expから乗り換え。性能と速度少しアップ。
github.com/MiaAI-Lab/DeepSee…
2
136
Shota Zaizen (財前 匠汰) retweeted
Claude Code is now multiplayer.
Run `/delta share` in Claude Code, and your teammate joins in their own session with your full context and uncommitted work. You're both working on the same thread, backed by DeltaDB.
delta.dev/blog/claude-code-m…
Shota Zaizen (財前 匠汰) retweeted
It's been a while since I showed any Jev stuff but I've been cooking on a way to use @GoogleDeepMind's EmbeddingGemma2 as an insanely fast real-time /v1/systemone endpoint (20ms on text, 100ms on images).
It's trainable in real-time, can distill an upstream teacher model, and is multi-modal by default. Hard to believe a @googlegemma model of this size can do so well with the right harness!
Shota Zaizen (財前 匠汰) retweeted
Opus 5.5, what are you doing?
It made another perfect jelly whale for $9.26. Translucent, squishy, and full of tiny bubbles. You can spin it, squish it, and toss it around while its whole body flops and wobbles
I just wanted to check the result. Now I can’t stop tossing this little guy around
LLM Sandbox重要ですねー
cohere.com/blog/concerning-b…
30






















