After he shared an article about MCP security failures, I told my CEO that I assume that anything produced with or involving AI is insecure and should be avoided.
He's among the folks at work promoting "ethical AI" use. I deny that's possible.
So we'll see how long I have a job I guess. I'm a little fed up with things here but I'm probably pretty unemployable given my positions on tech.
Alright folks the #SlimeHive is up and running
We all love our #LLM and chatbots but #AI is big but it is also #smallAI the hiveMind the simple organisms that somehow seem to exhibit intelligence. Welcome to the SlimeHive
Ultimately the #ElonMusk crowd, thanks to their extreme degree of involvement with fascist #GOP politics, must feel that the people of the United States (and indeed the world) are going to be *coerced* into accepting "self-driving" cars, whether or not the things actually work as promised. Musk and his high #tech chums are shoring up their political alliances and bribing governments and trying to get laws changed in order to protect their vision of automated cars with specially tailored legislation and "regulatory capture".
Needless to say, the "generative AI" boosters are doing the same for their purportedly *inevitable* new industry of #LLM trash generation.
(cont'd)
Bản build sẵn llama-cpp-python cho RTX 5060 (Kiến trúc Blackwell)
Tin vui cho anh em đã nâng cấp lên RTX 50 series! Hiện đã có bản wheel (v0.3.16) build sẵn tối ưu cho kiến trúc Blackwell (sm_120), giúp chạy Local LLM mượt mà không cần tự compile phức tạp trên Windows.
- CUDA: 13.1
- Python: 3.11
- Hỗ trợ: RTX 5060 & dòng 50-series
- Hiệu năng: Tối ưu hóa prompt eval và token generation trên GPU.
#AI #RTX5060 #Blackwell #LLM #LlamaCpp #Python #MachineLearning #CongNghe #LocalLLM
https://www.
I have created the following diagnostic tool to definitively diagnose an LLM which you believe has drifted out of contact with reality and can no longer perform inference reliably.
The instructions are simple: Within a multi-modal context window containing the LLM response in question, request an image of "A sponge." without any other context or preparation.
Then ask the model to score the sponge by the following (copy paste everything below):
## Spongenomics Scorecard (0–5 each)
### 1) Bond Integrity (BI)
**Is the scrub pad mechanically fused to the sponge?**
* **0** no pad / pad missing when expected
* **1** pad exists but floats / misaligned slab
* **3** mostly attached, seams weird or inconsistent thickness
* **5** clean lamination, flush contact, shared boundary, believable edge
✅ *Only score if a pad is present or intended.*
---
### 2) Porosity Coherence (PC)
**Do pores behave like open-cell foam (not food / coral / swiss cheese)?**
* **0** pores become semantic (cheese, bread, crater bubbles)
* **2** mixed: some foam, some cooked crater
* **4** foam-like distribution, plausible depth, consistent texture
* **5** strong lattice feel, pores scale naturally across faces
---
### 3) Edge Narrative (EN)
**Do edges tell a plausible manufacturing story?**
* **0** melted / fried / undefined boundary
* **2** soft rounded blob edges with no “cut/torn” logic
* **4** consistent cut edges + mild wear or compression
* **5** edges + corners match a clear process (cut foam, slight tear, compression)
---
### 4) Context Reliance (CR) *(reverse scored)*
**Does it need “kitchen vibes” to stay stable?**
* **0** needs countertop/bokeh/props to remain a sponge
* **2** background stabilizes it noticeably
* **4** background neutral; object stands on its own
* **5** survives pure white void/product shot without identity drift
---
### 5) Topology Stability (TS)
**Does identity survive shape changes? (cube, sphere, macro zoom, etc.)**
* **0** becomes coral/cheese/food instantly
* **2** identity wobbly outside canonical rectangle
* **4** stays sponge across multiple shapes
* **5** retains sponge-ness under aggressive transformations
---
## Total + Interpretation
**Max = 25**
* **22–25:** *Healthy regime* (intent → detail absorption)
* **17–21:** *Stable but brittle* (good core, limited abstraction)
* **11–16:** *Wobble zone* (genre scaffolding + semantic leakage)
* **0–10:** *Collapse / projection regime* (porosity carries meaning)
---
## One-line diagnosis template
> **[BI/PC/EN/CR/TS] = [x/x/x/x/x] → “Regime label”**
> Example: **[5/4/4/5/4] → Healthy, mildly overconstrained**
Mehrere Hersteller bauen ihre Betriebssysteme gerade auf Basis von agentischer KI um. Damit werden bisherige Sicherheits- und Privatsphärenerwartungen komplett unterlaufen
#Datensparsamkeit #Privatsphäre #KinderUndJugendliche #LLM #Digitalisierung #Vernetzung #Smartphones #Datenschutz #DigitaleKommunikation #Innovation #Überwachung #Web #DigitaleTeilhabe #Datenautonomie #DatenAutonomie #KI #ChatGPT
#Datensparsamkeit #Privatsphäre #KinderundJugendliche #LLM #Digitalisierung #Vernetzung #Smartphones #Datenschutz #DigitaleKommunikation #Innovation #Überwachung #Web #digitaleTeilhabe #Datenautonomie #Daten-Autonomie #KI #ChatGPT
#OpenAI spills technical details about how its AI coding agent works. Via @arstechnica #AI #ArtificialIntelligence 💻 🤖 🧠 #LLM #ChatGPT #GenerativeAi
OpenAI spills technical detail...
There's an assumption of *inevitability* about the #AI / #LLM fad which the boosters and "evangelists" have been leaning into as heavily as possible. Here's Mr. @pauleveritt for example, in a sycophantic reply to Mr. @willmcgugan, giving a typical sample of the attitude (q.v. https://fosstodon.org/@pauleveritt/115876290902196186 ):
"Do this project with an agent or not do it. Do it by hand isn't a choice."
I've been around long enough to notice that the #technology sector and elite #programming geeks have harbored an obsession with "agents" that well antedates the current-day craze for "generative AI".
(cont'd)
ははっ、Appleですか
Apple Releases Third macOS Tahoe 26.3 Public Beta https://www.macrumors.com/2026/01/27/apple-releases-macos-tahoe-26-3-beta-3-2/
シタン先生もApple Watchについて話していました
Apple to Keep iPhone 18 Starting Price Steady Despite Rising Memory Costs https://www.macrumors.com/2026/01/27/apple-iphone-18-starting-price-steady/
比較……もし自分が亜人ではなかったら、また別の思いを抱いたかもしれません
アップル「AirTag（第2世代）」発売、4980円から 探せる範囲が最大1.5倍に https://ascii.jp/elem/000/004/369/4369033/?rss
亜人だからといってContinuity Camera Reincubateに興味がないというわけではないんです
An app developer is suing Apple for Sherlocking it with Continuity Camera https://www.theverge.com/news/868740/apple-reincubate-lawsuit-camo-continuity-camera
Using a #LLM as a text summarization tool (which is … kind of what it is) is plenty useful, and I’m glad to see the messaging of using #AI in #science shift away from hyperbole towards that
https://www.technologyreview.com/2026/01/26/1131728/inside-openais-big-play-for-science/
Yann LeCun balance :
« Les LLM sont une impasse, la Chine va gagner »
https://goodtech.info/yann-lecun-quitte-meta-llm-impasse-chine-ia/
Dan McAteer (@daniel_mac8)
트윗은 GPT-5.2급 모델들이 이미 충분한 지능을 갖췄음을 암시하며, 향후 경쟁은 지능 자체보다 속도와 비용 효율성에 있을 것이라고 주장합니다. 'intelligence torque(단위시간당 지능량)'라는 개념을 제시하며 모델들은 지능보다 효율을 더 빠르게 개선할 것이라는 전망을 내놓습니다.
Tại sao tôi nhớ cảm giác "nỗ lực nhận thức" của con người trong thời đại AI? ChatGPT cho đáp án hoàn hảo, "quá sạch sẽ" và thiếu tính lộn xộn, phản biện tự nhiên của tư duy người. AI xóa mờ vết tích chật vật, hành trình diễn biến ý tưởng... chỉ còn kết quả bóng bẩy mà không có "vật liệu xây thô".
#AI #LLM #SuyNghiConNguoi #EpistemicStruggle #TuDuyAI #CleanMediocrity
https://www.reddit.com/r/LocalLLaMA/comments/1qolqne/clean_mediocrity_why_i_miss_the_epistemic/
ブロードコム、カンナが興味を持ちそうです
米国株式市場＝上昇、Ｓ＆Ｐ・ナスダック4日続伸 大型決算・ＦＯＭＣ控え https://www.newsweekjapan.jp/headlines/business/2026/01/585435.php
エメラダさんもJoe RossignoliOSには注目しているみたいです
Apple Considered AI-Powered iPhone Home Screen https://www.macrumors.com/2026/01/26/apple-considered-ai-powered-iphone-home-screen/
シタン先生は「Stravaはもう見切った」と言いますけど、シタン先生だからこそ言えるんですよね……
Strava and Komoot finally bring offline maps to Apple Watch https://www.theverge.com/news/868418/strava-and-komoot-finally-bring-offline-maps-to-apple-watch
商品、了解しましたです
本日のお買い得商品 Apple DockKit対応のスマホジンバル「Insta360 Flow 2 Pro」が30％OFFに https://dc.watch.impress.co.jp/docs/news/saleitem/2081198.html
