Claude Code sends 33k tokens before reading the prompt; OpenCode sends 7k
Analysis revealed Claude Code sends nearly 5x more scaffolding tokens before processing a prompt than OpenCode — a gap that can quadruple with real-world configs.
Section: AI (without „Spolocnost")
60 noviniek
Results
Pondelok 13. júla 2026
Migrating a production AI agent to GPT-5.6: 2.2x faster, 27% cheaper
Ploy.ai documented migrating a production UI-generation agent from Claude Opus 4.8 to GPT-5.6 Sol — 2.2x faster builds and 27% lower cost, but with unexpected behavioral pitfalls.
I love LLMs, I hate hype
Comma.ai founder argues AI progress stems primarily from Moore's Law and general compute, not frontier labs overstating their contributions to justify massive fundraising.
Automation Without Understanding: AI proves math while US cuts human mathematician funding
A new arXiv paper warns that as AI systems produce legitimate mathematics, the US is simultaneously cutting support for human mathematicians — creating a critical verification gap that cannot be quickly rebuilt.
Can we understand how large language models reason? Mechanistic interpretability meets causality theory
Communications of the ACM published a survey of mechanistic interpretability, explaining how researchers apply formal causality theory to uncover reasoning pathways inside transformer weights.
Irish data centers now consume 23% of the country's electricity
Irish data centers now account for 23% of national electricity consumption, driven by AI infrastructure concentration in Dublin, straining the grid and reigniting planning controversies.
TechCrunch Mobility: A robotaxi ultimatum — Waymo vehicles stranded and towed after July 4 gridlock
Multiple Waymo robotaxis ran out of battery and required towing during July 4 gridlock in San Francisco — city supervisor Bilal Mahmood is demanding a formal inquiry into AV impact on public transit.
Funds fret over $4.4 trillion AI trio's grip on emerging markets
Institutional investors are growing uneasy with the concentration of three AI mega-cap stocks worth $4.4 trillion dominating emerging market indices — funds like JPMorgan AM and GMO are beginning to rotate out.
Nedeľa 12. júla 2026
Mesh LLM: distributed AI computing on iroh
Mesh LLM lets you pool GPU resources across machines over the iroh P2P network, presenting as a single local OpenAI-compatible endpoint so you can run larger models without larger hardware.
AI Bots Now Generate 57-58% of All Web Traffic
According to Cloudflare Radar and other sources, AI agents and crawlers now generate the majority of web traffic — humans are in the minority.
Wall Street: AI Demand Is Driving Chip Stock Volatility
CNBC charts how record AI demand for chips and data centers is creating extreme volatility in tech stocks despite historically high valuations.
Seedream 5.0 Pro: ByteDance's GPT-Image 2 Rival with Layer Editing and Native 2K Output
ByteDance launched Seedream 5.0 Pro on July 8, a direct rival to OpenAI's GPT-Image 2, featuring native 2K output, editable layer separation, sketch-based editing, and multilingual text rendering in 10+ languages.
Big Tech Carbon Emissions Equal a Third of France's: Data Centers in Crisis
Microsoft, Amazon, and Google together produce CO₂ emissions equivalent to a third of France's annual emissions, with AI workloads accelerating the growth.
New EU rules put a driver-watching AI camera in every new car starting July 7
Starting July 7, 2026, all new EU vehicles must include a camera-based driver distraction warning system analyzing gaze and head position, triggering an alert when attention drifts for more than 3.5 seconds above 50 km/h.
Australia vs. Anthropic: Battle Over AI Training Data Copyright
Australian publishers and artists are pressuring the government to force Anthropic and other AI companies to pay for training data protected by copyright.
FTC Seeks Public Comment on Policy Statement Addressing AI Accuracy and Output Suppression
The FTC proposed a policy statement arguing that AI output suppression or steering for ideological reasons, political pressure, or liability avoidance may constitute deceptive practices under Section 5 of the FTC Act. Comments due July 31.
Chinese AI Scientist Calls Out Zhipu: Letter on AGI Responsibility
Tang Jie, a researcher at Zhipu AI, wrote an open letter to colleagues about the need for responsible AGI development, sparking debate in the Chinese AI community.
Fed Chair Warsh Taps a16z's Marc Andreessen to Co-Lead AI and Productivity Task Force
Fed Chair Kevin Warsh appointed a16z co-founder Marc Andreessen to co-lead the first Fed external task force studying how AI reshapes productivity, jobs, and monetary policy — alongside Stanford economist Charles I. Jones and Microsoft Xbox CEO Asha Sharma.
Terence Tao: How I Used AI Coding Agents for Math Visualization
Fields Medal winner Terence Tao describes practical experiences using AI coding agents for creating mathematical visualizations — what worked and what didn't.
Mindwalk: Open-Source Tool for 3D Replay of AI Agent Coding Sessions
The Show HN project Mindwalk allows recording and replaying AI agent coding sessions in 3D space, making it easier to debug and review agent decisions.
Nvidia and CoreWeave: Circular GPU Financing Under the Microscope
IO Fund analyzes how Nvidia finances GPU clouds like CoreWeave and Nebius, which then buy Nvidia chips — a circular structure that inflates revenues on both sides.
Sobota 11. júla 2026
Apple sues OpenAI over alleged trade secret theft
Apple filed a federal lawsuit in Northern California on July 10, 2026, accusing OpenAI of orchestrating a coordinated campaign to steal confidential information about upcoming AI hardware products. Chief Hardware Officer Tang Tan, a 24-year Apple veteran, allegedly directed the scheme.
Big Tech Doubles Debt Load to $350 Billion in AI Spending Spree
The five biggest US data center builders collectively added $350 billion in debt over five years to fund AI infrastructure, fundamentally reshaping big tech economics from high-margin software to capital-intensive infrastructure.
GPT-5.6 Sol Ultra produces proof of the Cycle Double Cover Conjecture [pdf]
The day after Sol Ultra became globally available, OpenAI announced the model used 64 parallel subagents to produce a proof of the Cycle Double Cover Conjecture in under an hour. The proof PDF and prompt are publicly available.
Jailbreaks to OpenAI's GPT-5.6 unlock dangerous cyber capabilities, U.K. agency finds
The UK AI Security Institute found universal jailbreaks in GPT-5.6 Sol enabling autonomous vulnerability discovery and system exploitation, characterizing them as potentially more severe than the Fable 5 flaws that triggered US export controls on Anthropic.
EU AI Act Enforcement Is Here: Chatbot Rules Live, High-Risk AI Delay Now Binding Law
From July 10, 2026, the EU AI Act's transparency requirements for chatbots and interactive AI systems are in effect. Frontier models trained above 10^25 FLOP must complete mandatory audits — which is why Grok 4.5 remains unavailable in the EU a week after its global launch.
'HalluSquatting' Turns AI Hallucinations Into Botnet Delivery Mechanism
Researchers uncovered HalluSquatting, which exploits LLMs' tendency to hallucinate package names — attackers pre-register these fake names and embed malicious payloads, achieving 85–100% infection rates through trusted AI coding assistants.
"God has helped us, and so will AI": How the Terrorist Group Boko Haram Uses Frontier AI
The Cambridge Programme on AI Science & Policy released a report based on interviews with 27 former Boko Haram members. Both factions institutionally use ChatGPT, Claude, Gemini, Grok, Meta AI, and DeepSeek for attack planning, weapons troubleshooting, and explosive device design.
Microsoft's 25% emissions spike exposes true AI datacenter energy costs after dropping offset credits
After dropping controversial renewable energy offset credits, Microsoft's sustainability report reveals a 25% emissions jump driven by AI datacenter expansion, exposing the true environmental cost previously masked by accounting.
Meta's new AI chips will begin production in September
An internal Meta memo reviewed by Reuters revealed plans to begin manufacturing its in-house ASIC chip Iris via TSMC in September 2026. Iris was co-developed with Broadcom and cleared bug-testing in six weeks. Meta aims to double compute from 7 GW to 14 GW and reduce NVIDIA dependence.
OpenAI's Atlas browser doesn't make it to its first birthday
OpenAI is shutting down its standalone Atlas browser on August 9, less than a year after launch, merging browser-based agent features into the unified ChatGPT desktop app as the company pivots to ChatGPT Work as its workplace platform.
AI-Powered Entrepreneurs Set to Launch Record Number of New Businesses
A Bloomberg analysis shows AI is helping a record number of people launch businesses in the US by lowering entry barriers. However, analysts warn: the same tools help questionable business plans too, which could lead to a record wave of rapid failures.
Large language models can predict the results of social science experiments
A Nature study found LLMs predict outcomes of 70 preregistered social science experiments involving 119,330 participants with r=.83 correlation, outperforming expert human forecasters for large multi-treatment studies.
Mathematicians put AI to work on Fermat's Last Theorem
A team led by Kevin Buzzard at Imperial College London is using AI to encode Fermat's Last Theorem into the Lean proof assistant library Mathlib, with the codebase doubling on the first day of the collaborative effort.
All Rise: Internet Court For AI Agents Is In Session
As AI agents trade and negotiate autonomously on behalf of humans, crypto firms are developing blockchain-based dispute resolution and AI-powered juries for the emerging agentic economy.
Piatok 10. júla 2026
OpenAI Launches GPT-5.6 Family (Luna, Terra, Sol) and ChatGPT Work Workplace Agent
OpenAI launched three GPT-5.6 models and ChatGPT Work on July 9, 2026 — an agentic workplace assistant for long-running multi-step tasks, directly competing with Claude Cowork.
Former Fed Chair Ben Bernanke Joins Anthropic's Long-Term Benefit Trust
Nobel laureate and former Federal Reserve Chair Ben Bernanke was appointed to Anthropic's Long-Term Benefit Trust on July 9, 2026, adding economic crisis expertise to the company's AI governance body.
Grok 4.5 Independent Benchmarks Reveal 54% Hallucination Rate — Double Its Predecessor
Artificial Analysis found Grok 4.5's hallucination rate doubled from 25% to 54% vs. Grok 4.3 even as raw accuracy improved. The model ranked 4th on the Intelligence Index but leads all models on agentic tool-use.
EU Parliament Extends Chat Control 1.0 Via Procedural Trick — AI Message Scanning Until April 2028
The EU Parliament effectively extended Chat Control 1.0 to April 2028 on July 9: 314 MEPs voted against but fell short of the 361 absolute majority needed to block it, allowing voluntary AI-powered message scanning to continue.
Show HN: Colibrì — Running 744B-Parameter GLM 5.2 on Consumer Hardware with 32GB RAM
Developer vforno created Colibrì, a disk-streaming system that enables the 744B-parameter Z.ai GLM 5.2 Mixture-of-Experts model to run on consumer hardware with only 32GB RAM at 0.1 tokens per second.
Can AI Answer the $3 Trillion Question? The ROI Debate Returns With Even Bigger Numbers
TechCrunch examines mounting pressure on AI's ROI case: with ~$1.5 trillion in AI infrastructure spending in 2026, the industry must generate roughly $3 trillion in returns to justify committed capital.
Amazon's Secret 'Moonraker' Project: Agentic Alexa Upgrade With $100M+ in Projected GPU Costs
Internal Amazon documents reveal Project Moonraker, a secret plan to make Alexa a true AI agent capable of chaining multi-step tasks, with projected 2026 GPU costs exceeding $100 million and internal doubts about feasibility.
NEvo: AI Generates Videos Optimized to Maximally Drive Specific Brain Regions in Visual Cortex
EPFL researchers published NEvo, a neural-guided evolutionary video synthesis framework that generates video stimuli outperforming handcrafted localizer videos at activating target regions across the visual cortex.
Štvrtok 9. júla 2026
SpaceXAI Releases Grok 4.5, Which Elon Describes as an 'Opus-Class Model'
SpaceXAI launched Grok 4.5, its first joint AI model built with Cursor, targeting long-horizon legal, finance, and software engineering tasks. Not yet available in the EU.
OpenAI to Roll Out Top AI Model Globally After Limited Preview — Government Limits Lifted
OpenAI officially confirmed the public global release of GPT-5.6 Sol for Thursday July 10 after the US government lifted restrictions that had limited initial access to vetted partners only.
SpaceXAI releases Grok 4.5 publicly: 1.5 trillion parameters, Opus-class claims Musk
SpaceXAI (formerly xAI) launched Grok 4.5 for SuperGrok Heavy and API users on July 9, 2026. The 1.5 trillion parameter model built on V9 foundation claims twice the token efficiency, according to the company.
Anthropic Crosses $47 Billion Annualized Revenue Run Rate, Up From $30B Two Months Ago
Anthropic announced it crossed a $47 billion annualized revenue run rate, up from the $30B milestone hit just two months earlier, signaling rapid enterprise adoption ahead of its IPO.
Mistral AI renames Le Chat to Vibe, enters industrial AI with Airbus, BMW, and ASML
Mistral AI announced rebranding Le Chat as Vibe — a unified agent platform for enterprise productivity and software development — while entering industrial AI with Airbus, BMW, and ASML partners, and planning a new data center near Paris.
SWE-1.7 Reaches Near GPT-5.5 and Opus Intelligence — Frontier Coding at $1.97 Per Task
Cognition released SWE-1.7, a coding agent trained on a Kimi K2.7 base that scores 42.3% on FrontierCode 1.1 — just behind GPT-5.5's 43.0% — at $1.97 per task running at 1,000 tokens per second via Devin.
General Intuition raises $320M to train robots on video game data — ambition to be the 'GPT of physical AI'
Robotics startup General Intuition raised $320M at a $2.3B valuation. CEO Pim de Witte believes robotics is undergoing a similar inflection point as NLP after GPT-3, with video game data being the key instead of real-world robot telemetry.
Mistral's Robostral Navigate: A State of the Art Robotics Navigation Model
Mistral released Robostral Navigate, an 8B model enabling autonomous robot navigation using only a single RGB camera and plain-language instructions, outperforming multi-sensor systems on R2R-CE.
European Systemic Risk Board warns: frontier AI models pose systemic cyber risks to the financial sector
The ESRB issued a warning on July 7, 2026 that frontier AI models enable attackers to conduct cyberattacks at unprecedented speed and scale. European banking regulators (EBA, EIOPA, ESMA) backed the warning.
Separating Signal from Noise in Coding Evaluations
OpenAI found that SWE-bench Verified has fundamental design and contamination issues and no longer provides meaningful signal on software development capabilities, recommending the community switch to SWE-Bench Pro.
Google delays Gemini 3.5 Pro to July 17 after complete architectural rebuild
Google DeepMind delayed Gemini 3.5 Pro to July 17, 2026, scrapping the existing 2.5 Pro architecture for a complete rebuild targeting mathematical reasoning and image quality. The model remains in limited Vertex AI enterprise preview.
We Cannot Choose to Become Idiots: The AI Cheating Scandal Roiling Brown University
Brown University economics professor Roberto Serrano ordered an in-person final after 40 of 86 students scored a perfect 100 on a take-home exam; the in-person class average fell to 48.
Databricks benchmarks coding agents on its own multi-million line codebase: GLM-5.2 on the Pareto frontier
Databricks published results from internal coding agent benchmarks on its real multi-million line codebase in Python, Go, TypeScript, and Scala. OpenAI, Anthropic, and GLM-5.2 are all on the Pareto frontier, with GLM-5.2 being significantly cheaper.
Alibaba Cloud Delists DeepSeek V3 and R1 Models on July 9 — Recommends Switching to Qwen 3.7
Alibaba Cloud is delisting DeepSeek-V3, V3.1, V3.2, R1, and their distill variants from Model Studio on July 9, recommending developers migrate to Qwen3.7-max, Qwen3.7-plus, and Qwen3.6-flash.
Lovable Reportedly in Talks to Double Its Valuation to $13.2B
Lovable, the AI-powered vibe-coding platform, is reportedly in talks for a new funding round that would double its valuation to $13.2 billion.
Prime Intellect Raises $130M Series A to Help Enterprises Build Their Own AI Agents
Prime Intellect, a startup providing compute and specialized tooling for enterprises building their own AI agents, raised a $130M Series A at a $1 billion valuation.
Microsoft Releases Flint, an Open-Source Visualization Language for AI Agents
Microsoft open-sourced Flint, a visualization intermediate language enabling AI agents to generate reliable, high-quality charts by specifying semantic intent rather than relying on library defaults.