The AI Intelligence Hub

BitsMinds

News, guides, tools and reviews from the world of artificial intelligence. Everything AI, in one place.

MICROSOFT BUILD 2026 · TOPOLOGICAL QUANTUM 1,000× more stable. Majorana 2 — a better topological qubit. New materials stack: lead · InAs / InAsSb ROADMAP HALVED → PRACTICAL MACHINE BY 2029 Microsoft MAJORANA 1 parity lifetime: milliseconds MAJORANA 2 ~20 s parity lifetime · up to 60 s BITSMINDS.COM Microsoft's claims · pending independent peer review
Research

Microsoft’s Majorana 2 Claims a 1,000× Leap in Qubit Stability — and Pulls Its Quantum Roadmap Forward to 2029

At Build 2026, Microsoft unveiled Majorana 2, the follow-up to its topological quantum chip, claiming qubits roughly 1,000 times more reliable than the prior generation. A new materials stack — swapping aluminum for lead — pushes the “parity lifetime” of a topological qubit to about 20 seconds, with some lasting a full minute, up from milliseconds. Microsoft says the progress lets it halve its roadmap and target a practical, fault-tolerant quantum computer by 2029 — even as critics remain skeptical of the topological approach.

Tom's Hardware / SiliconANGLERead →
GTC TAIPEI · COMPUTEX 2026 · THE CPU FOR AGENTS Built for agents, not humans. Nvidia's first CPU for the agentic era. 88 Olympus cores · 176 threads · LPDDR5X 1.2 TB/s 1.8× AGENTIC SANDBOX vs x86 · 63% vs GRACE · BEATS EPYC + XEON VERA OLYMPUS · ARM 88-CORE AGENTIC CPU BITSMINDS.COM Source: NVIDIA newsroom · Phoronix
Industry

Nvidia’s Vera Is a CPU “Built for Agents, Not Humans” — 88 Custom Cores Aimed at Intel and AMD’s Last Stronghold

NVIDIA / PhoronixRead →
GTC TAIPEI · COMPUTEX 2026 · PHYSICAL AI OPEN STACK Physical AI, fully open. One model to perceive, simulate, and act. Cosmos 3 omnimodel · reason, then generate action OPEN WEIGHTS · NANO 16B · SUPER 64B · EDGE 2B SOON COSMOS 3First fully open physical-AI omnimodelNANO · SUPER16B (8B+8B) · 64B (32B+32B) · open weightsTHINK, THEN ACTReason block reads scene → action block actsBENCHMARKS#1 on Physics-IQ, R-Bench, PAI-BenchISAAC GR00TOpen 6-ft humanoid · 31 DoF · runs on Thor BITSMINDS.COM Source: NVIDIA blog · The Robot Report
Research

Nvidia Open-Sources Cosmos 3, Its “Think-Then-Act” Physical-AI Omnimodel — and a 6-Foot GR00T Humanoid to Run It

NVIDIA / The Robot ReportRead →

Latest News

All news →
BUILD 2026 · MICROSOFT TAKES AIM AT CLAUDE LOCKED ON MICROSOFT AI MAI-Thinking-1 FIRST IN-HOUSE REASONING MODEL 35B active · ~1T MoE · 256K ctx trained with zero distillation Claude ENTERPRISE CODING DEFAULT SWE-Bench Pro ≈ 53% — level with Opus 4.6 BITSMINDS.COM Microsoft's claims, pending independent benchmarks
Models

Microsoft Aims MAI-Thinking-1 Straight at Claude: a 35B Reasoning Model It Says Beats Sonnet 4.6 and Matches Opus 4.6 on Code

At Build 2026, Microsoft’s AI Superintelligence Team unveiled MAI-Thinking-1, its first in-house reasoning model — a sparse Mixture-of-Experts design with 35B active parameters (~1T total) and a 256K context window. Microsoft says human raters prefer it to Claude Sonnet 4.6 and that it matches Opus 4.6 on the SWE-Bench Pro coding benchmark, and it pointedly trained the model from scratch with zero distillation on commercially licensed data — no OpenAI involved.

Microsoft AI / TechTimes

Trump Signs a Softer AI Executive Order: A Voluntary 30-Day Frontier-Model Review, and No Mandatory Licensing
Industry

Trump Signs a Softer AI Executive Order: A Voluntary 30-Day Frontier-Model Review, and No Mandatory Licensing

President Trump signed an executive order on June 2 asking AI developers to voluntarily hand the federal government access to their most powerful models for up to 30 days before public release, so agencies can benchmark “advanced cyber capabilities.” The order explicitly bars any mandatory licensing or preclearance — a markedly lighter touch than the version pulled in May — and pairs the review with a new Treasury-run cybersecurity clearinghouse for critical infrastructure.

The Register / NPR / The White House

Microsoft Build 2026: A Homegrown MAI Model Blitz, an “IQ” Layer for Agents, and a Quiet Goodbye to OpenAI Dependency
Products

Microsoft Build 2026: A Homegrown MAI Model Blitz, an “IQ” Layer for Agents, and a Quiet Goodbye to OpenAI Dependency

Build 2026’s headline was MAI-Thinking-1 — Microsoft’s first from-scratch reasoning model, claiming Opus 4.6-level coding at 35B active parameters — alongside MAI-Code, Image, Voice and Transcribe, the Microsoft IQ context layer, a GitHub Copilot desktop app, and an agent-governance stack.

Microsoft

GTC TAIPEI · COMPUTEX 2026 · KEYNOTE RECAPJUN 1–2Agentic AI, end to end.“Every token is now a revenue unit” — Jensen Huang40,000 engineers · rack assembly: 2 hrs → 5 minRUBIN SHIPS THIS SUMMER · AWS · GOOGLE · AZURE · ORACLEVERA RUBINFull production · supply chain 2× BlackwellVERA CPUBuilt for agents · 1.8× vs x86RTX SPARKAI PCs this fall · ASUS, Dell, HP, Lenovo…COSMOS 3Open omnimodel for physical AIISAAC GR00T6-ft humanoid platform for researchBITSMINDS.COMSource: CRN Asia · NVIDIA keynote
Products

Nvidia’s COMPUTEX 2026 Blitz: Vera Rubin in Full Production, a CPU “Built for Agents,” and a PC You Just Ask

Jensen Huang’s GTC Taipei keynote put Vera Rubin into full production with HBM4 from Samsung, SK Hynix and Micron, unveiled the agent-first Vera CPU (1.8x vs x86), RTX Spark AI PCs landing this fall, and Cosmos 3 for physical AI.

CRN Asia

SECURITY · AI SUPPORT BOT · ACCOUNT TAKEOVERMETA AI × INSTAGRAM · JUN 1Hacked by asking.Meta’s AI support bot changed account emails on requestNo access to the victim’s inbox needed — 2FA never triggeredSTEP 1 · VPNSpoof the victim’s regionSTEP 2 · CHATAsk bot: add new emailSTEP 3 · CODECode sent to attackerSTEP 4 · RESETNew password · takeoverHit: Obama-era White House · Space Force’s top enlisted leader · OG handles resold on TelegramPATCHED · FIX ANNOUNCED MONDAY · SCOPE UNDISCLOSEDBITSMINDS.COMSource: TechCrunch · The Decoder · Engadget
Industry

Hackers Hijacked Instagram Accounts by Simply Asking Meta’s AI Support Chatbot to Change the Email

A logic flaw in Meta’s AI support assistant let attackers add their own email to any Instagram account and reset the password — bypassing 2FA entirely. Victims included the Obama-era White House handle and the Space Force’s top enlisted leader.

TechCrunch

Anthropic Quadruples Project Glasswing to ~200 Partners in 15+ Countries as Mythos Goes After Critical Infrastructure
Companies

Anthropic Quadruples Project Glasswing to ~200 Partners in 15+ Countries as Mythos Goes After Critical Infrastructure

Anthropic is adding ~150 Glasswing partners across power, water, healthcare and communications — quadrupling the program to ~200 organizations as it warns that Mythos-class cyber capability will be replicated within 6–12 months.

Anthropic

AI INFRASTRUCTURE · EQUITY RAISE · CAPEXALPHABET · JUN 1$80BEquity raise to fund the AI buildoutDemand is exceeding available supplyHOW THE $80B BREAKS DOWN$30B$40B$10BPublic offeringAt-the-marketBerkshireRAISE vs 2026 CAPEXTHIS RAISE$80BALPHABET 2026 CAPEX$180–190BINDUSTRY AI INFRA · 2026≈ $700B across Big TechBITSMINDS.COMSource: TechCrunch · CNBC · Alphabet filing
Companies

Alphabet to Raise $80 Billion for Its AI Buildout — With a $10 Billion Vote of Confidence From Buffett’s Berkshire

Google’s parent will sell roughly $80B in stock to fund AI infrastructure as demand outstrips supply — including a $10B private placement to Berkshire Hathaway, with 2026 capex guided to $180–190B.

TechCrunch

OPEN SOURCE · MIXTURE-OF-EXPERTS · APACHE 2.0JETBRAINS MELLUM2 · JUN 2Mellum212B total parameters · ~2.5B active per tokenMoE · 131K context · 2x faster inference6 variants · Base · Instruct · Thinking (RLVR)THE FOCAL-MODEL THESISFast, specialized parts orchestrated by frontier modelsROUTER · 8 OF 64 EXPERTS FIREOnly ~21% of parameters fire per tokenBITSMINDS.COMSource: JetBrains AI Blog · Hugging Face
Models

JetBrains Open-Sources Mellum2, a 12B Mixture-of-Experts Model Built to Be a Fast "Focal" Part, Not a Frontier Rival

JetBrains released Mellum2 under Apache 2.0 on June 2 — a 12B MoE model that activates just 2.5B parameters per token for 2x-faster inference, pitched as a fast specialized component for multi-model AI pipelines.

JetBrains AI Blog

Anthropic Confidentially Files for an IPO, Beating OpenAI to the SEC at a $965B Valuation
Companies

Anthropic Confidentially Files for an IPO, Beating OpenAI to the SEC at a $965B Valuation

Anthropic submitted a confidential draft S-1 to the SEC on June 1, days after a $65B round lifted it to a $965B valuation — setting up a three-way IPO race with OpenAI and SpaceX.

PYMNTS

Trending

Top Guides

All guides →
BITSMINDS GUIDE · WORKING WITH AI Drive the conversation. Four habits for better — and cheaper — AI answers. 1PROMPTGoal first, be explicit2MEMORYPin rules up top3RESETStart fresh early4TOKENSKeep it lean BITSMINDS.COM Original guide · BitsMinds
Text & Chat

How to Work With AI Effectively: A Practical Guide to Prompts, Memory, and Token-Efficient Conversations

The Agent Loop Each tool call returns to Claude until stop_reason equals end_turn. User request messages array Claude API claude-opus-4-7 + web_search + tools Tool handler validate + write prompt + tools stop_reason: tool_use tool_result returns as next user message BITSMINDS.COM
CodeClaude API

How We Built Our First AI Agent: A Production Guide to Claude API Agents

ComfyUI — generation as a graph Drag, drop, connect. Every step you'd hide behind a UI is a node here. Load Checkpoint sdxl_juggernaut.safetensors CLIP Text Encode "forest cabin, golden hour" KSampler model prompt latent VAE Decode Save Image Each wire is a tensor. Each node is a function. Reorder for ControlNet, IPAdapter, upscalers, video, audio…
ImageComfyUI

ComfyUI: The Node-Based Workflow That Power Users Actually Run

Runway Gen-4.5 Text + image in. Cinematic clip with synced audio out. INPUTS TEXT PROMPT drone shot over a fishing boat at sunrise, calm waves REFERENCE IMAGE harbor-sunrise.jpg 2048×1024 · 1.2MB SETTINGS 10 sec 16:9 audio on 10 SECOND CLIP · 4K · 24FPS SYNCED AUDIO
VideoRunway

Runway Gen-4.5: The Production Standard for AI Video in 2026

Trending AI Tools

All tools →