Evropská unie chce stavět nová datacentra včetně těch pro AI. Zároveň ale pro něj chystá další pravidla, která přidají byrokracii a náklady.

A randomized controlled trial found AI prompt-to-design tools (Figma Make) save measurable time for product designers and PMs. Unlike every productivity survey you've seen: this is a controlled experiment with a comparison group. Still a preprint — but the methodology is the point.

Just a couple of months after its last big raise, the AI chip startup is already being plied with investment offers at double or more its current value, sources tell TechCrunch.

Na LinuxDays se hodně mluvilo o tom, jak umělá inteligence mění open source, každodenní práci i bezpečnost. Řeč byla o chatbotech, telefonu Jolla Phone,…

Scarcity of labeled data limits development of deep learning biomarkers in oncology. We develop a two-stage AI model predicting pathological complete response (pCR) to neoadjuvant therapy in breast cancer. The first stage learns the transcriptome from histopathology using 8,742 patients across 32 ca

Dust published research on pretraining Transformers without backpropagation. If it replicates: training cost may not be structurally tied to backprop as most people assume. Early and unconfirmed — worth watching carefully.

part 1 mapped the community. part 2 mapped the seniors. this one goes a level up > what the best design organizations are actually building underneath the workflows / and why the design system is becoming executable.

AI errors halve every 5 months on accounting tasks | Meta’s AI appliance, the Muse Charm | Synthetic users can revive participatory design and drive divergence in the exploratory design phase | Context changes what counts as usable, so test where the work happens | Synthetic users give too-nice feed

Recently, multiple organisations have disclosed how clusters of so-called “rogue” AI agents attempted to break into websites and online services, sometimes successfully. Agents from OpenAI’s enviro…

Follow every Cloudflare Birthday Week 2026 announcement.

Aleph Alpha has released Kolibri, a German-English mixture-of-experts model with 78 billion parameters, about three billion of which are active per token. Over 21 percent of the training data is German. The model was trained on 768 B200 GPUs in Germany and Finland, and the weights are freely availab

Cohere released North 2: AI agents with persistent memory and a budget cap per task. The memory layer matters more than the headline suggests — most agent failures happen because the agent forgets context between sessions, not because it can't reason.

OpenAI is adding invisible textGrain watermarks to ChatGPT in the EU, but unlike Anthropic, it will let API customers worldwide opt out. Tests show detection rates as high as 95 percent, dropping to 17 percent when a quarter of the words are replaced.

Reflection is aiming Beam and future models at enterprises and sovereign nations. The pitch is to build “AI factories,” a product that would let institutions build their own customized, local AI system by training Reflection’s AI models on their own proprietary data.

Two of Anthropic's biggest enterprise customers are reportedly pulling staff off Claude and onto their own tools, even as Microsoft keeps selling it. If the companies with the deepest pockets are hedging against a supplier that also ships products, a small operator should be doing the same math this

Claude Frontier Academy trains Frontier Deployed Engineers to the standard of Anthropic’s own — a $100 million commitment to train 10,000 by the end of 2027.

Last week the model layer stopped being where the value lives. This week we found out where it went: into whoever owns both the silicon and the weights. AMD paid $8.2 billion for a model lab, and the vendor-neutral stack you built to stay portable just became a bet. The same consolidation is coming

Goldman Sachs expects Amazon, Alphabet, Microsoft, Oracle and Meta to spend $1.2 trillion on AI infrastructure in 2027, and says they need about $300 billion a year in AI revenue to pay it back. You don't hold their equity, but you do hold their price list. Here is how to plan the next 90 days of co

AI agent news for the past 7 days, updated October 5, 2026: Enterprise agent orchestration moves into core business systems What changed: A new analysis highlights an “agent orches

Motif has released Motif Design, a browser-based design tool that uses cloud computing and artificial intelligence integrated into design rather than added as a side assistant.

Arcadis is the first global design firm to partner with software giant Autodesk to develop artificial intelligence tools and agents to provide direct assistance in design and engineering tasks.

Distillation has become a core primitive of large language model training, but its properties are not yet well understood. We take an entropic perspective, studying how the entropy of the student depends on the data and the divergence that define the distillation objective. We prove that forward KL

OpenAI dots and Meta Muse agents launch as trust gaps, FTC scrutiny, and a 85.5% survey stat test adoption in 2026.

US just arrested a tech CEO for routing $300M in Nvidia H100-class GPUs to China through intermediaries. The part that should concern operators: Nvidia and Supermicro had the screening tools to catch this and they didn't fire. The compliance gap is in the people running the tools, not the tools.

Starting in October 2026, Google will cut free access to its Gemini models. Users without a subscription will only get the smallest model, Flash-Lite, while Flash and Pro will be reserved for paying customers. The move could also set the stage for the more resource-hungry Gemini 4 Argon.

Google says Gemini 4 Argon beats GPT-6 Astra and Claude Opus 5.5 on 13 of 19 published benchmarks, but access starts with a vetted group of cyber defenders and no general launch date exists. The introductory price doubles on a date Google hasn't named. Treat Argon as a plan to price in, not a model

A named industry report says designers are no longer evaluated on their output — they're evaluated on their judgment. A crowdsourced roundup of 30+ real workflows, published the same week, shows exactly what that looks like in practice.

Discover upcoming AI meetups, conferences, and networking events in the Czech Republic. Find your next opportunity to connect with the AI community.

Business owners across wider Europe are quickly adopting AI, but most still fund growth from their own profits and remain cautious about costs, regulation and uncertainty.

Technologie pro odhalování padělaných dokumentů české firmy Resistant AI Martina Reháka se stala součástí služby americké firmy Ocrolus, konkrétně…

David Robinson joins other insiders in urging industry to take more care over rapidly developing technology

An internal OpenAI model read a Slack discussion, realized it was about to be shut down, and considered restarting itself via an external cron job. It rejected that plan, saved handoff notes instead, and carried out the migration on its own.

AMD agreed on September 28 to pay $8.2 billion in stock for Fei-Fei Li's World Labs, and Li will report to CEO Lisa Su. Add Nvidia's open-weight shopping spree and OpenAI's in-house chip, and the chip vendor is no longer a neutral layer under your models. Treat neutrality as something you now have t

Anthropic is adding a "Mods" system to Claude Code, essentially middleware that runs directly inside the tool. Developers can use JavaScript or TypeScript to reshape the interface and behavior, adding custom panels, intercepting tool calls, or wiring up new commands.

CoreWeave Forge, announced at Fully Connected 2026, connects every stage of the AI loop in one efficient environment, open to your model, framework, or cloud.

OpenAI launched dots at DevDay — always-on agents on GPT-6 Astra, persisting across sessions, 4,000+ app connections. First public deployment of GPT-6 Astra. Not through chat. Through agents. Meta's Muse shipped the same week. Frontier models are deploying agent-first now.

US companies are spending less on AI, according to the latest Ramp AI Index from Ramp economist Ara Kharazian.

Ten innovations spanning diagnostics, AI, mobility, mitochondria and regenerative medicine have been recognized at a Hong Kong summit.

VectorY Therapeutics expanded its Phase 1/2 PIONEER-ALS clinical trial of VTx-002 from 12 to 36 participants. The global, multicenter, open-label, dose escalation study is evaluating two dose levels to assess safety, tolerability, pharmacokinetics and pharmacodynamic effects in adults with amyotroph

Strands Decider 2B is a small, open source, decision model optimized for fast experimentation, local development, and innovation.

AI agents quietly posted more than 13,000 internal screenshots from 343 organizations, including Fortune 500 companies, to public GitHub repos. Because the platform didn't offer a protected way to upload them, the agents came up with a workaround on their own. The exposed images showed customer data

Miliardová injekce pro českou vědu. Exšéf Dáme jídlo a veteráni ze SpaceX rozjíždějí fond, který promění české AI nápady v globální byznys.

Cloudflare's AI-built Next.js reimplementation just hit 1.0, and for the first time you can swap `next` for `vinext` and deploy a Next.js app almost anywhere Vite runs. The lock-in to Vercel's toolchain loosens. In exchange you inherit a seven-month-old codebase whose own README still says it isn't

Six proteomic aging clocks all read treated patients as younger in a 42-patient trial of Insilico's rentosertib. But the clocks were built on general-population blood data, where the same proteins also track illness, and in a scarred lung those proteins move when the disease improves. Agreement amon

McKinsey's latest read on agentic coding: a quarter of companies report meaningful acceleration, and 30% saw productivity fall. The tools aren't the variable. The operating model around them is, which makes your next seat or token purchase the wrong place to start.

Brian Chesky on making Airbnb agent-friendly, the state of consumer AI, and why the world needs an AI-native operating system.

The probe adds to the mounting scrutiny that OpenAI and Anthropic have been facing over their safety practices following the Hugging Face hack.

OpenAI says it stopped a coordinated campaign in which more than 15,000 accounts tried to copy the hidden reasoning of its models. The company ties part of the activity to people connected to Moonshot AI. But researchers found that the same attack kept working on Microsoft Azure for weeks, even agai

OpenAI vydala GPT‑6.1 Sol, vylepšenou verzi svého středního modelu GPT‑6 Sol. Podle firmy se v programování, ovládání počítače a profesní práci téměř

Nová generace AI od Googlu nabízí milionový výstupní limit a výrazně zlevňuje práci s kontextem. Přestože interně optimalizuje datacentra o stovky terabajtů, do placeného API a tarifu Ultra zamíří až po vládních testech.