What Did GPT-6 Astra Do With Raw Model Weights? Results
Astra identified games from raw weights, fixed a stuck car by editing 3.5% of a network, and wrote Atari networks from scratch. Here are the numbers and limits.
AI News & Models
AI Models & Industry
New AI models reviewed and compared, how they work, and what the labs behind them are betting on.
107 articles · Updated Nov 12, 2026
Most read by people in Engineering, Executives and Legal & Compliance.
Tencent Open-Source World Model vs Genie 4: Can You Run It?
Tencent's open-weight world model claims Genie 4-style interactive generation on one consumer GPU, with minute-long keyboard steering. How it compares.
Tencent world modelGenie 4 open source alternativeopen-source world model
What Is a Looped Transformer? NVIDIA Dory Explained
A looped transformer reruns a shared block of layers per token to gain depth without more weights. Here is how NVIDIA's Dory does it, and what it costs.
How Does SenseNova-Loop Get More Depth Without More Parameters?
SenseNova-Loop-9B-Preview reuses 24 layers twice for 56 layer executions from 32 stored layers, with two-stream mHC and 3:1 linear-to-full attention.
What Are Subvocal's Silent-Speech Glasses and Do They Work?
Subvocal's radar-based glasses claim to turn silent mouthing into text at 200+ wpm for $499 pre-order. Here is how they work and what remains unproven.
Subvocal glassessilent speech glassessubvocalization wearable
Will Websites Block Your AI Agent? Three Zones of Access
Banks, hotels and schools are deciding how AI agents get in. How signed agents, Visa and Bill, Stripe and Google AP2 shape which sites your agent can use.
AI agents website accessCloudflare signed agentsagent payments protocol
How Is LFM3's Tri-Hybrid Architecture Built and Trained?
How Liquid AI built LFM3-1.7B-Base: 13 gated short convolutions, 5 sliding-window and 6 global attention layers, and a 22T-token recipe to 256K context.
LFM3 architecturetri-hybrid modelgated short convolution
What Is the Agent Control Plane and Who Will Own It?
Salesforce, Microsoft, Oracle and OpenAI are opening products to each other. Here is what the agent control plane is and why enterprise vendors want it.
agent control planeSalesforce Microsoft Copilot StudioOracle ChatGPT workspace
What Is Dynamic Creatures' Hugo Robot and When Will It Launch?
Hugo is a furry humanoid robot from Dynamic Creatures with 57 joints and 43 actuators, run by a human remotely today, with customer tests planned for 2027.
Dynamic Creatures Hugohumanoid robot theme parkHugo robot
Sonnet 5.5 vs Opus 5.5 vs Fable 5.1: Which Claude Model?
Haiku, Sonnet, Opus and Fable compared on cost, speed and task fit, plus when Fable 5.1 is worth paying for and how the /advisor command helps.
Sonnet 5.5 vs Opus 5.5Fable 5.1which Claude model
What Does Anthropic's S-1 Reveal About Revenue and Losses?
Anthropic's S-1 shows $4.585B 2025 revenue, over $16B in H1 2026, an $80B net loss, and a listing expected near Nov 16 at $1.5T to $2T.
Distillation vs Reinforcement Learning: Can Qwen 4B Match 27B?
Qwen 27B taught Qwen 4B three ways: teacher answers, probabilities, on-policy. All landed near 58% versus the 64% teacher, but one cut tokens 30%.
model distillationdistillation vs reinforcement learningon-policy distillation
What Is Broccoli AI? The YC Home Services Startup at $60M ARR
Broccoli uses AI voice, text and SMS to run the front office for home-service contractors. Here is what it does and how its YC founders found the idea.
Broccoli AI home servicesAI receptionist for contractorsYC startup $60M ARR
Why Does Intel 14A Use High-NA EUV When TSMC Doesn't?
Backside power relaxes front-side pitch, so Intel can print 14A layers in one high-NA pass while TSMC sticks to double-patterned EUV. Plus Intel's PDK lessons.
High-NA EUVIntel 14Abackside power delivery
FinFET vs Gate-All-Around: Why Did GAA Take 34 Years?
How transistors moved from planar to FinFET to gate-all-around, and why GAA, demonstrated in 1988, only reached volume production at Samsung in 2022.
gate-all-around transistorFinFET vs GAAhigh-k metal gate
What Is Clover-35B-A3B? Block Diffusion on a Qwen3.6 MoE
How DiffuSpace's Clover derives from Qwen3.6-35B-A3B: bidirectional GDN attention, confidence-threshold decoding, and block-level expert coreset selection.
Clover-35B-A3Bblock diffusion LLMQwen3.6-35B-A3B
Is NCM Faster Than an LLM? Next Concept Modeling Tested
Stanford's NCM writes text coarse to fine. Independent DGX Spark tests show about 4x speed for one user, slower batching, and a Sudoku result with a caveat.
Next Concept ModelingNCM StanfordNCM vs LLM
What Does Anthropic's S-1 Reveal About Enterprise AI Spend?
Anthropic's S-1 shows over half the Fortune 500 as customers, yet under $1 spent per $1,000 of expenses. Here is what the filing says about demand.
What Does Anthropic's IPO Filing Reveal About Its Finances?
Anthropic's IPO filing cites $16.3B in half-year revenue, up to $750B in compute commitments and a class F share with 50.1% of votes. Key figures.
Did OpenAI Disprove the Exponential Time Hypothesis?
OpenAI says its model disproved the Exponential Time Hypothesis. Here is what ETH is, how it relates to P vs NP, and what the claim does and doesn't mean.
Exponential Time HypothesisOpenAI ETH disprovedP vs NP AI
How Does Medical Billing Work and How Is Candid Automating It?
How US medical billing works, from claims and ICD-10 codes to denials, and how Candid Health uses automation and AI to fix revenue cycle management.
Candid Healthmedical billing automationrevenue cycle management
ALEF Flying Car Price, Pre-Orders and Delivery Date
ALEF's flying car starts around $300,000, has 3,600+ pre-orders and targets first deliveries by year-end after its 1,000th test flight. Here is what we know.
ALEF flying car priceALEF flying car pre-orderALEF vs Tesla Roadster
What Does Anthropic's S-1 Filing Reveal About Its Business?
Anthropic's S-1 shows 900% net revenue retention, 11,900 customers above $100K, multi-chip compute across four vendors, and a "closing the loop" thesis.
How Good Is GPT-6.1 Terra? Coding and Computer Use Tests
Hands-on GPT-6.1 Terra results: build times for a browser OS, skate game, subway FPS and wrestling game, plus a robot arm run, pricing and quirks.
GPT-6.1 Terra reviewGPT-6.1 Terra codingGPT-6.1 Terra computer use