AI — Page 6

Artificial intelligence, machine learning, LLMs, and AI tools transforming development.

OM-1: Training Robot Policies From Human Demos Alone

Reward AI says OM-1 learned manipulation from human glove demonstrations alone. What's claimed, what's missing, and what evidence would prove it works.

AI Agents Could Use 100x More Energy Than Chatbots. Here's Why

Yuki Okonkwo·1 month ago

AI. Yuki Okonkwo1 month ago· 6 min read

AI Agents Could Use 100x More Energy Than Chatbots. Here's Why

Agentic AI runs one request through many model calls, searches and retries. New research says the energy cost can be over 100 times a chatbot query's.

MIT's HardFlow Aims to Make AI Outputs Obey the Rules

Marcus Chen-Ramirez·1 month ago

AI. Marcus Chen-Ramirez1 month ago· 6 min read

MIT's HardFlow Aims to Make AI Outputs Obey the Rules

MIT researchers say HardFlow steers pretrained diffusion and flow models to satisfy hard safety and physical constraints without retraining. Here's what it does, and what we still don't know.

Amodei's Call to Slow AI: Pacing or Positioning?

Bob Reynolds·1 month ago

AI. Bob Reynolds1 month ago· 6 min read

Amodei's Call to Slow AI: Pacing or Positioning?

Dario Amodei wants AI development paced and third-party evaluators like METR inside frontier labs. What the proposal promises, and what it leaves unanswered.

Agent Harnesses, Not Bigger Context Windows, Decide Long Task Success

Rachel "Rach" Kovacs·1 month ago

AI. Rachel "Rach" Kovacs1 month ago· 5 min read

Agent Harnesses, Not Bigger Context Windows, Decide Long Task Success

A new analysis argues agent harnesses beat raw context windows on long tasks. We examine the four mechanisms, the missing benchmarks, and what it means.

Altman Rules Out 2026 OpenAI IPO, Citing Safety Concerns

Samira Barnes·1 month ago

AI. Samira Barnes1 month ago· 5 min read

Altman Rules Out 2026 OpenAI IPO, Citing Safety Concerns

Sam Altman called a 2026 OpenAI IPO "ill-advised" while discussing security and recursive self-improvement. What the timing actually signals for governance and oversight.

DeepSeek V4.1 Flash: How 890 Bytes Per Token Reshaped Inference Costs

Yuki Okonkwo·1 month ago

AI. Yuki Okonkwo1 month ago· 6 min read

DeepSeek V4.1 Flash: How 890 Bytes Per Token Reshaped Inference Costs

DeepSeek V4.1 Flash cut KV cache costs 437x to 890 bytes per token. A look at the architecture, the $0.36 agent bill, and where the tradeoffs live.

Close-up of a man with earbuds and microphone beside a code-filled screen reading Codex Is Dead

Flask Creator Armin Ronacher on Why Minimal Coding Agents Win

Marcus Chen-Ramirez·1 month ago

AI. Marcus Chen-Ramirez1 month ago· 9 min read

Flask Creator Armin Ronacher on Why Minimal Coding Agents Win

Flask creator Armin Ronacher on why the minimal Pi coding agent rivals bigger harnesses, and his take on cloud agents, open source, and inference costs.

Claude Code Plugin Evals Turn AI Skills Into Testable Software

Samira Barnes·1 month ago

AI. Samira Barnes1 month ago· 6 min read

Claude Code Plugin Evals Turn AI Skills Into Testable Software

Anthropic's new plugin evaluation workflow grades Claude Code skills against a no-plugin baseline and can gate CI. What it does, and what it leaves unanswered.

HarnessDev Asks Whether AI Models Can Engineer Their Own Tools

Yuki Okonkwo·1 month ago

AI. Yuki Okonkwo1 month ago· 6 min read

HarnessDev Asks Whether AI Models Can Engineer Their Own Tools

HarnessDev from ByteDance Seed and partners tests whether LLMs can build their own agent harnesses. The catch: only 34 of 64 changes generalized. Here's what that means.

Wide-eyed man in a black hoodie beside a DeepSeek post reading “We’re back” dated Sep 9, 2026

DeepSeek V4.1 Flash: Benchmarks Shine, Real Tasks Falter

Dev Kapoor·1 month ago

AI. Dev Kapoor1 month ago· 7 min read

DeepSeek V4.1 Flash: Benchmarks Shine, Real Tasks Falter

DeepSeek's new open-weights model posts frontier-level benchmark scores and rock-bottom prices, but hands-on tests reveal cracks in stateful logic and simulation.

Five male portraits with speaker names beneath bold text reading “Gambling with our lives”

Inside the Labs, the Doom Estimates Are Getting Louder

Bob Reynolds·1 month ago

AI. Bob Reynolds1 month ago· 6 min read

Inside the Labs, the Doom Estimates Are Getting Louder

A researcher quit, an alignment lead priced the risk above 10%, and OpenAI claims a Millennium Prize result. What the warnings actually say, and what they don't.

Meta Faces New Class Action Over AI Training and Face Recognition

Marcus Chen-Ramirez·1 month ago

AI. Marcus Chen-Ramirez1 month ago· 6 min read

Meta Faces New Class Action Over AI Training and Face Recognition

A new class action alleges Meta used Facebook and Instagram photos to train AI and build NameTag face recognition. Here's the legal terrain and Illinois precedent.

Google's ToolGrad Flips Tool-Use Training Data Generation

Rachel "Rach" Kovacs·1 month ago

AI. Rachel "Rach" Kovacs1 month ago· 6 min read

Google's ToolGrad Flips Tool-Use Training Data Generation

ToolGrad builds valid API chains first, then writes the user query, hitting a 99.8% pass rate. We look at what that number does and doesn't prove.

Jensen Huang Calls AGI Arrived: What Investors Hear Differently

Bob Reynolds·1 month ago

AI. Bob Reynolds1 month ago· 6 min read

Jensen Huang Calls AGI Arrived: What Investors Hear Differently

Jensen Huang declared AGI arrived this week. Two veteran VCs heard a term worth interrogating. Here's what the panel said on coding, law, agents and safety.

OpenAI's Antitrust Problem: When an AI Slowdown Looks Like Collusion

Samira Barnes·1 month ago

AI. Samira Barnes1 month ago· 6 min read

OpenAI's Antitrust Problem: When an AI Slowdown Looks Like Collusion

OpenAI and other AI firms are weighing coordinated slowdowns, but antitrust law may treat safety coordination as cartel behavior. Here is the legal terrain.

Presenter onstage before an OpenAI screen reading BEL 10T, stamped LEAKED, beside a glowing nighttime Earth

OpenAI's BEL Leak and the Navier-Stokes Claim: What's Real

Marcus Chen-Ramirez·1 month ago

AI. Marcus Chen-Ramirez1 month ago· 7 min read

OpenAI's BEL Leak and the Navier-Stokes Claim: What's Real

Leaks describe a 10-trillion-parameter OpenAI model called BEL, but the evidence is thin. We sort what's confirmed from what's hype in the GPT-7 rumor cycle.

Smiling man speaking into a microphone beside a screen labeled “/studio/” with an illustrated portrait of him

One Person, One Studio: Inside Nick Saraev's AI Animation Pipeline

Yuki Okonkwo·1 month ago

AI. Yuki Okonkwo1 month ago· 7 min read

One Person, One Studio: Inside Nick Saraev's AI Animation Pipeline

Nick Saraev built a full animation pipeline from talking-head footage, style models, and an AI agent. Here's how it works and what it costs.

PyTorch Conference China 2026: The Open AI Stack Grows Up

Yuki Okonkwo·1 month ago

AI. Yuki Okonkwo1 month ago· 5 min read

PyTorch Conference China 2026: The Open AI Stack Grows Up

PyTorch Conference China ran in Shanghai alongside KubeCon and OpenInfra, placing AI frameworks beside the infrastructure layers models need to ship.

Google's Mantis Gives Coding Agents a Full Security Feedback Loop

Dev Kapoor·1 month ago

AI. Dev Kapoor1 month ago· 6 min read

Google's Mantis Gives Coding Agents a Full Security Feedback Loop

Google open-sourced Mantis, an Apache 2.0 toolkit that lets AI coding agents find, reproduce, and patch vulnerabilities. What it does, and what's unproven.