Topic Cluster

#LLMs

Navigate language-model releases, assistant comparisons, answer verification and local-model choices.

39 articles

Where to start

A model name, a chat subscription and a local installation describe different things. This collection helps you separate those layers before comparing capability or cost. The release guides are dated snapshots; the workflow articles explain how to decide which facts matter for your task.

  1. Compare assistant workflows

    Choose evaluation tasks before choosing a subscription.

  2. Read the Gemini release-status analysis

    Distinguish an announcement from a specification or a rumor.

  3. Inspect the Grok 4.7 release guide

    Review access and evaluation questions without assuming identical account availability.

  4. Plan a local-model trial

    Connect the shortlist to memory limits and a repeatable check.

Start with an assistant comparison for everyday work. Use model guides to check a particular announcement and follow their primary sources for current details. If privacy, hardware or offline use motivates the choice, move to the local-model shortlist. Always verify an important answer against its source rather than trusting a model label.

Common Questions

Stop Repeated AI Agent Tool Calls

Diagnose repeated AI tool calls, define observable progress and set application-enforced limits without mistaking a higher step cap for a fix.

Marcus Vance·3 min read·
Common Questions

AI Transcription Invents Words in Silence

Trace invented transcript text to silence, segmentation or earlier context, then test speech detection without removing quiet real speech.

Julian Thorne·3 min read·
Common Questions

Choose LLM Quantization for Your Task

Compare smaller quantizations using your own failure cases, memory limits and correction effort instead of assuming one format always wins.

Marcus Vance·3 min read·
Common Questions

Check Whether Your Local AI Works Offline

Audit generation, embeddings, OCR and tools, then test a complete offline workflow without assuming that a local model makes every step local.

Marcus Vance·3 min read·
MCP server stdout carrying JSON-RPC messages and stderr carrying separate diagnostic logs.
Common Questions

MCP Invalid JSON: Check Your stdout Logs

Find startup banners and debug logs contaminating MCP stdio messages, then separate protocol output from diagnostics safely.

Marcus Vance·3 min read·
Common Questions

MCP Tool Schema Fails in One Client

Compare MCP tool schemas and identical inputs across clients to isolate compatibility failures without disabling validation or broadening access.

Marcus Vance·3 min read·
Common Questions

RAG Misses French or Arabic Documents

Test multilingual RAG at extraction, retrieval and answer stages, using paired questions instead of assuming fluent output proves good search.

Julian Thorne·3 min read·
Common Questions

Ollama Out of Memory in Longer Chats

Find why a local model fits initially but fails in longer chats, using a context and concurrency test that separates memory causes.

Marcus Vance·6 min read·
Common Questions

Why the First Ollama Response Is Slow

Separate model loading from prompt processing and streaming delays to diagnose why the first local AI response takes much longer.

Marcus Vance·3 min read·
Common Questions

Valid AI JSON Can Still Be Wrong

Validate AI output in three stages: JSON syntax, schema and real meaning, with explicit handling for missing facts and failed validation.

Marcus Vance·3 min read·
Illustrative Ollama terminal output highlighting the PROCESSOR column.
Common Questions

Ollama Not Using GPU: Diagnose CPU Fallback

Diagnose Ollama CPU fallback by checking model placement, GPU visibility, context and memory before changing drivers or buying hardware.

Marcus Vance·3 min read·
Common Questions

Why RAG Still Mentions a Removed Document

Trace old document facts through chat history, retrieval collections and caches to understand why removed material still influences answers.

Julian Thorne·3 min read·
Common Questions

Can a Document Redirect Your AI Agent?

Use a harmless document fixture to test prompt-injection handling and enforce tool permissions outside the model's interpretation of text.

Julian Thorne·3 min read·
Evidence checkpoints through document extraction, indexing, retrieval, prompt assembly and answer generation.
Common Questions

RAG Cannot Find an Answer: Trace the Failure

Trace a missing RAG answer through extraction, indexing, selection and retrieval instead of repeatedly uploading the same document.

Julian Thorne·3 min read·
Scanned invoice beside extracted text missing the invoice identifier.
Common Questions

RAG Cannot Read a Scanned PDF: What to Check

Diagnose scanned-PDF failures by checking OCR, reading order, indexing and retrieval before changing prompts or replacing the language model.

Julian Thorne·3 min read·
A source table compared with extracted values that have lost their row relationships.
Common Questions

Why RAG Gets Table Totals Wrong

Find whether incorrect AI table totals come from parsing, missing rows or arithmetic, then choose a retrieval or calculation workflow.

Julian Thorne·3 min read·
Claude Opus 5.5 announcement thumbnail showing the Anthropic Claude logo with headline text 40% Cheaper, 30% Faster, Stronger
AI

Claude Opus 5.5: Features, Pricing and Access

Anthropic's Claude Opus 5.5 leads in agentic coding at 40% lower cost than Opus 5 in vendor testing. Explore its features, pricing and access options.

Elena Rostova·6 min read·
Best local LLMs for 16GB RAM featuring Qwen, Gemma, and Llama model logos above computer memory hardware
Guides

Best Local LLMs for 16GB RAM: 4 Models Compared

Compare four local LLMs for 16GB RAM by download size, hardware and task. Use ten test prompts to check answer quality, memory use and response speed.

Marcus Vance·13 min read·
3D editorial illustration of an autonomous AI agent with tool use and memory loops
Explained

What Is an AI Agent, Really?

Everyone's building AI agents. But the term means different things to different people. Here's what actually qualifies — and what doesn't.

Marcus Vance·4 min read·
Retrieval-Augmented Generation (RAG) architecture explained simply
Explained

RAG Explained Without the Jargon

Learn how retrieval-augmented generation finds sources, adds context and produces an answer, with a worked example and the limits you need to check.

Marcus Vance·4 min read·

The Weekly Breakdown

High signal AI & software stories.
Direct to your inbox. No hype.

Independent analysis of AI models, developer tools, and computing architectures. Delivered every Sunday morning. 100% free.

Zero spam·One-click unsubscribe·Sunday delivery