Agregátor RSS

Sony u letošních novinek slibuje, že přinese kino do vašeho domu a nový komplet Bravia Theatre System 6 pro domácí kino to plní do puntíku.

Byla vydána verze 1.86.0 programovacího jazyka Rust (Wikipedie). Podrobnosti v poznámkách k vydání. Vyzkoušet Rust lze například na stránce Rust by Example.

Necelé tři měsíce od startu se operátor první české zpravodajské družice Saturnin-1 pochlubil černobílou fotografií Prahy. Svým zpracováním připomíná starší snímky z dalšího českého satelitu Vzlusat-2, obě zařízení proto nejspíše používají obdobnou technologii. [********************] ...

Espionage? Botnets? Trying to exploit a zero-day?

Updated  Someone or something is probing devices made by Juniper Networks and Palo Alto Networks, and researchers think it could be evidence of espionage attempts, attempts to build a botnet, or an effort to exploit zero-day vulnerabilities.…

The North Korean threat actors behind Contagious Interview have adopted the increasingly popular ClickFix social engineering tactic to lure job seekers in the cryptocurrency sector to deliver a previously undocumented Go-based backdoor called GolangGhost on Windows and macOS systems. The new activity, assessed to be a continuation of the campaign, has been codenamed ClickFake Interview by Ravie Lakshmananhttp://www.blogger.com/profile/[email protected]

** Nové hodinky ukazují, jak může za pár let vypadat elektronika v Evropě ** A nezůstávají jen u výměny baterie ** Časem můžete hodinkám upgradovat snímače nebo třeba displej

The rules have changed. Again. Artificial intelligence is bringing powerful new tools to businesses. But it's also giving cybercriminals smarter ways to attack. They’re moving quicker, targeting more precisely, and slipping past old defenses without being noticed. And here's the harsh truth: If your security strategy hasn’t evolved with AI in mind, you’re already behind. But you’re not alone—[email protected]

ProtectEU plan wants to have its cake and eat it too

The EU has shared its plans to ostensibly keep the continent's denizens secure – and among the pages of bureaucratese are a few worrying sections that indicate the political union wants to backdoor encryption by 2026, or even sooner.…

Žádná služba dosud nenabídla tolik kultovních filmových sérií. Star Wars, Predátor, Vetřelec, X-Max… Disney tyto značky vlastní, a tak může nabídnout více než pohádky a animované filmy, které si uz něj člověk hned vybaví. K tomu přidává výborné muzikály, válečné filmy, ale i další dosud nedostupnou ...

Ikea má až do konce dubna nebo do vyprodání zásob v akci kancelářskou židli Markus . Běžně stojí 3990 Kč, během covidu cena šplhala až téměř k pěti tisícům, ale nyní může být vaše za 2990 Kč . Stejně stojí i méně obvyklá světle šedá varianta . Židle má nosnost až 110 kg. Lze jí regulovat výšku ...

When Google unveiled a new neural network design in 2017 called the Transformer architecture, it probably had no idea this would threaten Google Search’s dominance within seven years.

When OpenAI applied Google’s Transformer architecture to generative language models, the GPT was born. (GPT stands for “Generative Pre-trained Transformer.”)

OpenAI then took the GPT concept and created a chatbot by incorporating reinforcement learning from human feedback, ranked responses, dialogue optimizations, and safety measures. The result was ChatGPT, which OpenAI made public on Nov. 30, 2022. 

ChatGPT changed the world. 

Now, AI chatbots like ChatGPT, Meta’s Meta AI, Microsoft’s Copilot, Anthropic’s Claude AI, Perplexity AI’s Perplexity, xAI’s Grok, and even Google’s own Gemini are changing how people search for information online. Instead of relying on Google, some users are turning to AI chatbots.

To stop bleeding users, Google has been planning to integrate its own AI technology into the Google Search Experience for the general public, first with AI Overviews in May 2024, which now reportedly has more than 1 billion users, and more recently with AI Mode (March 2025).

AI Mode is an “experimental” service available to people who signed up for Google Search Labs. It’s based on a customized version of Google’s Gemini 2.0.

Note that AI Overviews, which many users found initially problematic (recommending glue for pizza recipes or suggesting the health benefits of eating rocks), was based on Gemini 1.5 Flash until March when it was updated to Gemini 2.0.

In its current state, Google’s AI Mode is truly great. It’s better than most comparable chatbots in several respects, especially two very important ones.

Why attribution matters

The best feature is that Google surfaces attribution. Just last week, I co-hosted a TWiT podcast about AI called “Intelligent Machines,” where I advocated for RAG-based chatbots (Retrieval-Augmented Generation systems like Perplexity that use data from searches in their results, rather than just the old data in their training datasets) showing their links conspicuously on the right side of the search. I’m in favor of highlighting and offering the easiest access to the actual sources of information rather than stolen knowledge being genericized, commoditized, and buried by AI chatbots.

What I described and advocated for in that podcast is exactly what Google Search’s AI Mode does.

Like Perplexity, Google AI Mode is a RAG system that performs a Google Search (Perplexity uses Google’s PageRank, too), then gets its information for the result from the pages on the other side of those links. 

But Google AI Mode does a better job showing the main links used for information provided in the answer, complete with an image for each link drawn from the source, where available. This is important for the reasons I’ve already detailed. AI chatbots are threatening to bury and replace the very sources they’re built on. By highlighting the sources with links, users can explore those original sources from the chatbot response — you know, like a search engine.

Still, Google understands that even with highlighted sourcing, people are far less likely to click through to linked sources than with Google Search.

Less is more

Another surprising AI Mode virtue is brevity. Unlike most AI chatbots, which often provide long-winded explanations that go off into tangents and background content, Google’s AI Mode gets right to the point with very concise responses.

Results tend to be very high quality. AI Mode uses what’s called a “query fan-out” technique to handle complex searches. It turns a prompt into multiple related queries targeting different subtopics and data sources, and the results are then combined into a single, detailed response that covers more ground than a traditional search.

It also uses a method called agentic reinforcement learning, developed with Google’s DeepMind group, to improve the accuracy of its AI-generated responses. This technique rewards the model for producing statements that are more likely to be correct and supported by reliable sources, such as Google’s Knowledge Graph or live web data, while also encouraging the retrieval of additional factual information.

Despite all this, Google warns that AI Mode can hallucinate, lie, give opinions instead of facts, and generally misbehave in the ways AI chatbots sometimes do. 

AI Mode offers a surprisingly appealing user interface with a responsive design and fluid layout. And it’s fast. Besides these qualities, AI Mode has other, more standard benefits found elsewhere, including multimodal queries support, meaning it accepts voice, text, or image inputs.

If the subject of the prompt is deemed controversial or problematic in some way, AI Mode throws up its conceptual hands and delivers a list of URLs like old-fashioned Google Search rather than a typed-out response. 

When AI Mode becomes the leading search option offered by Google — and I think that’s likely — the book on SEO will have to be torn up and re-written. At this point, it’s not clear how, exactly. Beyond that, the rules for advertising will also need to be re-written. 

AI Mode is currently a fantastic search replacement. But in the future, how Google handles monetization will make or break the service. 

Advertising and the future of search

Google isn’t exactly sure how it will integrate ads into AI Mode searches. If AI Overviews suggests a direction, then Google’s AI Mode will likely show ads under the AI answer and will be labeled “Sponsored.” The ads will come from the same places as normal Google ads, like Search and Shopping, and will try to match the content searched for. The ads should link to things that are helpful for your search. For example, if you search for how-to information like “how to get wine stains out of a shirt,” ads for Wine Away and other such products will appear below or within the results.

Ultimately, it’s advertising that’s likely to lead AI Mode astray, given how Google Search itself has evolved over time.

Ads now often blend seamlessly with organic search results, making it harder for users to distinguish between paid content and genuine information. The prominence of ads pushes organic content further down the page, especially on mobile devices, reducing visibility for non-sponsored links and forcing users to scroll more to find unbiased results. Google’s strategy of embedding ads within organic listings also disrupts the traditional flow of search results, subtly prioritizing monetization over clarity and usability.

So, while AI Mode is a refreshing change from plain old Google Search and clearly one of the best AI chatbots out there, I’m pessimistic that Google will be able to monetize the feature without making the user experience much worse.

It’s a real challenge, because the company really depends on search-based ad revenue. How Google can keep making money on search ads in the age of AI, smart glasses and other trends isn’t yet clear. 

But changes are coming. And it will be interesting to see how Google navigates the future of AI search and advertising.

AI holds the promise to revolutionize all sectors of enterpriseーfrom fraud detection and content personalization to customer service and security operations. Yet, despite its potential, implementation often stalls behind a wall of security, legal, and compliance hurdles. Imagine this all-too-familiar scenario: A CISO wants to deploy an AI-driven SOC to handle the overwhelming volume of security [email protected]

V našem výběru najdete deskovky napříč cenovými kategoriemi, žánry i náročností. Vybírat můžete od epických kampaňovek až po jednoduché karetní hry.

Kalifornský startup Aptera Motors vyrazil se svým tříkolovým solárním vozidlem na legendární dálnici Route 66. Futuristický vůz, který může většinu dní jezdit bez potřeby dobíjení, zvládl urazit úsek dlouhý 480 kilometrů napříč jihozápadem USA. Díky integrovaným solárním panelům dokáže vozidlo ...

Bylo oznámeno, že vývojáři e-mailového klienta Thunderbird pracují na open source webových službách Thunderbird Pro a Thundermail, kterými chtějí konkurovat Gmailu a Office365. Thunderbird Pro nabídne Thunderbird Appointment, Thunderbird Send a Thunderbird Assist. Webová e-mailová služba Thundermail je již pro vývojáře dostupná na thundermail.com a tb.pro. Zapsat se lze na čekací listinu. Thundermail vychází z projektu Stalwart.

Recovery's never been harder in today's tangled, outsourced infrastructure

Comment  Disaster recovery is getting tougher as IT estates sprawl across on-prem gear, public cloud, SaaS, and third-party ITaaS providers. And it's not floods or fires causing most outages anymore - ransomware now leads the pack, taking down systems faster than any natural disaster.…

Ačkoli ekonomové nepředpokládají, že nastavení cel zůstane na stávající úrovni, svět – včetně USA – podle nich zcela zbytečně v následujících letech zchudne.

V Česku od listopadu 2018 funguje pod patronací ČNB systém okamžitých plateb, kdy odchozí korunové transakce dorazí na účet v jiné bance za pár sekund. Z počátku tuto novinku podporovalo jen pár bank a účtovaly si za to poplatek. V průběhu let se situace normalizovala tak, že okamžité platby ...

Cybersecurity researchers have disclosed details of a new vulnerability impacting Google's Quick Share data transfer utility for Windows that could be exploited to achieve a denial-of-service (DoS) or send arbitrary files to a target's device without their approval. The flaw, tracked as CVE-2024-10668 (CVSS score: 5.9), is a bypass for two of the 10 shortcomings that were originally disclosed byRavie Lakshmananhttp://www.blogger.com/profile/[email protected]

Lidé mají tendenci hledat nejrychleji akcelerující produkční škodovku mezi modely s odznáčkem RS, ale doteď to byl docela normální Superb 2.0 TSI 4x4, který se z 0 na 100 km/h dostane za 5,6 sekundy. Právě představená Škoda Elroq RS je ještě o dvě desetiny sekundy rychlejší. [********************] ...