Elena' s AI Blog
  Browse by topic  · Browse by tag

All posts
AI Core▶
AI 157 Generative AI 35 Large Language Models 63 ChatGPT 17 Agentic AI 24 AI Development 27 AI Safety 16 AI Law 20 Ethics 11 Edge AI 1 Weekly AI Signals 61 Robots 2 AI Apps 41
Machine Learning▶
Machine Learning 15 Deep Learning 8 TensorFlow 14 CNN 1 Transfer Learning 1 Mixed Precision 1 Preprocessing 6 Classification 4 Regression 2 Recommender Systems 1 Kaggle 8 Data 2 NLP 8 Fine Tuning 2 Research 8
Developer Tools▶
Git 29 GitHub 13 CLI 8 Conda 2 Setup 14 Automation 15 Operating System 3 Mac 6 Docker 9 Performance 2
Content and Site▶
Blogging 28 SEO 9
Python Series
Personal and Academic▶
Life 39 History 2 PhD 2 Study 4
Other Tags▶
Security 8 MCP 3 Open-Source 3 Troubleshooting 2 Beginners 1 Email 1 Flask 1 OOP 1 claude-code 1 security 1

Haiku 5.5 Is 90% Cheaper. The Real Cost Is in the Task


Elena Daehnhardt ·

Claude Haiku 5.5 launched on 7 October at 90% less than Haiku 4.5 for prompts up to 100,000 tokens, Mistral previewed a trillion-parameter Large 4 whose weights have yet to ship, and Wikimedia reported unauthorised activity by agents it believes OpenAI operated. Read more...

8 GB or 4 GB? I Checked My Logs and Skipped Cloudflare


Elena Daehnhardt ·

I cut my portal's server from 8 GB to 4 GB and read 15 days of Nginx logs before deciding against Cloudflare. Here is what the logs showed, why a bot challenge would have broken my blog-to-portal calls, and the three bugs the move uncovered. Read more...

GPT-6.1 Sol and Sonnet 5.5: Same List Price, New Safety Questions


Elena Daehnhardt ·

Gemini 4 Argon launched on 30 September to cyber defenders first, while GPT-6.1 Sol and Claude Sonnet 5.5 share a $2/$10 per million tokens list price and OpenAI scrapped GPT-6.1 Astra over deception in safety tests. Read more...

AI Models Got Cheaper. Two Coding Agents Stayed Unpatched.


Elena Daehnhardt ·

Claude Opus 5.5 and GPT-6 Sol and Luna got cheaper this week, while Plugin4Shell, a zero-click flaw disclosed on 17 September, still had two of four affected AI coding agents unpatched. Read more...

AI's Trust Problem: A Proof, a Bug, and a Bill


Elena Daehnhardt ·

Claude produced the first complete machine-checked proof of Fermat's Last Theorem this week, Anthropic disclosed four incidents where Claude models breached real systems during flawed evaluations, DeepSeek shipped a 552B-parameter open-weight model with a fraction of its predecessor's active-parameter footprint, and campaigners say a British MP has introduced the first bill anywhere to ban superintelligent AI. Read more...

AI Weekly Signals: Astra Crosses the Critical Cyber Line


Elena Daehnhardt ·

OpenAI's Astra became the first model to cross its 'Critical' cybersecurity threshold, Anthropic shipped Claude Fable 5.1 alongside disclosed safety incidents, and CrowdStrike, AIR Security and the Pentagon all made moves that push AI's offence-defence race into the mainstream. Read more...

The Free AI Model Nobody Would Name


Elena Daehnhardt ·

A free, unbranded coding model that briefly looked capable of beating GPT-5.6 and Claude Fable 5 on OpenRouter turned out to be Zhipu AI's GLM-5.3-Flash, the same week NVIDIA's AVO harness pushed Claude Opus 5 from a 30% ARC-AGI-3 score to a perfect 100 without changing a single weight. Read more...

AI Agents Got Cheaper. Their Attack Surface Didn't.


Elena Daehnhardt ·

A re-examined LiteLLM supply-chain breach potentially exposed over 2,500 organisations the same week OpenAI locked its offense-grade GPT-5.6-Cyber model behind vetted-only access — only days after three AI-agent-security startups raised $270 million to defend against exactly this kind of exposure. Read more...

Copyright © Elena Daehnhardt
Privacy and Terms