Author: Santiago Maniches
-
How Model Quantization Actually Works: INT8 to INT4
A 70B model needs 140GB at full precision, 35GB at INT4. The rounding math, why naive quantization breaks, and how GPTQ and AWQ fix it.
-
How LLM Tokenization Actually Works: BPE Explained
The algorithm behind every LLM tokenizer was built in 1994 to compress files, not language. The mechanism, and why it breaks on math and non-English text.
-
How Mixture-of-Experts Actually Routes Every Token
DeepSeek V4 holds 1.6 trillion parameters and uses a fraction per token. The routing math, why naive versions collapse, and the modern fix.
-
Why R Still Beats Python in Clinical Biostatistics
SAS still dominates regulatory submissions, but R runs the survival models, mixed-effects analysis, and increasingly the FDA-facing tables behind it.
-
ONNX Explained: How One Format Runs Any AI Model
ONNX lets a model trained in PyTorch run on a phone, browser, or edge chip without rewriting it. The graph, the opset system, and where conversions break.
-
Same Model, 20-Point Gap: Why Coding Benchmarks Mislead
Claude Opus 4.6 scores 58% or 80% on the same benchmark depending only on which harness wraps it. Here is why coding agent scores mislead.
-
How an Export Law Built for Chips Took Down Fable 5
A private Commerce Department letter used a dormant 2018 export authority to shut down Fable 5 and Mythos 5 worldwide. Here is the mechanism.
-
NeuroGolf 2026: Kaggle’s Race to Build the Tiniest AI
A Kaggle contest that closed today scores neural networks by size, not accuracy. Here is the formula, the constraints, and what teams built to win.
-
MCP Goes Stateless on July 28. Its Poisoning Problem Stays
MCP’s July 28 spec removes sessions and adds response caching. That solves scaling headaches and quietly widens the window for tool poisoning.
-
ShareLock Splits Malicious Prompts Across AI Agent Tools
Researchers used Shamir’s secret sharing to hide prompt-injection payloads across MCP tools, beating detectors with a 90%+ success rate.
-
The Annotation Underground: Who Trains AI for So Little
Every frontier model, Claude included, runs on labelers paid under $2 an hour in countries picked for weak labor law. The full wage chain, documented.
-
The Anchor Problem in AI Agent Delegation Chains
OAuth anchors one-hop agent delegation. At the third hop, no deployed protocol can prove which human authorized the action. Inside the 2026 standards race.
-
MITRE ATLAS: The ATT&CK Framework for AI Systems
MITRE ATLAS provides the shared vocabulary for AI security threat intelligence: 14 tactic categories, techniques like AML.T0018 (Backdoor ML Model) and AML.T0043 (Craft Adversarial Data), and a crosswalk…
-
Neural Backdoor Attacks: From BadNets to LLM Trojans
Gu et al.’s BadNets (2017) installed hidden triggers via training poisoning. By 2023, instruction-following backdoors target RLHF pipelines directly. Rare-word triggers, weight poisoning, and universal adversarial triggers all…
-
LLM Watermarking: How Models Embed Detection Signals in Their Outputs
Kirchenbauer’s green-red token list (ICML 2023), Aaronson’s EMS, and Kuditipudi’s ITS scheme all embed detectable statistical signals into LLM outputs. But Zhang et al. proved no watermark is…
-
Differential Privacy for LLMs: The Training Privacy Guarantee
Differential privacy provides the only formal guarantee against LLM training data leakage. DP-SGD’s four steps, the Moments Accountant, Renyi DP, and the epsilon values that actually mean something…
-
Multiagent LLM Security: When Your Agent Talks to a Malicious Agent
When LLMs call other LLMs as tools, injection attacks jump the boundary. ConVerse (2026) found 88% privacy violations and 60% security breaches in plausible agent-to-agent discourse. Here is…
-
LLMail-Inject: What 208K Attacks Against an Email Agent Found
Microsoft Research’s LLMail-Inject challenge: 839 participants, 208,095 unique attacks against a simulated email agent with production defenses. The finding: adaptive attackers breach even well-designed defense stacks. Here is…
-
Adversarial Machine Learning: From Szegedy to LLM Attacks
Szegedy (2014) showed deep networks could be fooled by imperceptible perturbations. FGSM, PGD, and C&W followed. By 2025, the same mathematical framework governs jailbreaks, poisoning, and memorization extraction.…
-
How RLHF and Constitutional AI Build Safety Into Language Models
RLHF trains models to prefer human-preferred outputs. Constitutional AI uses AI self-critique guided by principles. Neither provides formal guarantees. Here is how both techniques work, what they install…
You must be logged in to post a comment.