Agregátor RSS

New AI tools could help more biologists experiment with problems once restricted to a handful of specialists.

Designing new proteins from scratch usually requires in-depth knowledge of their structures and chemistry, as well as familiarity with highly specialized software. But researchers have now developed an AI tool where users simply describe what they want a protein to do in plain language.

Since AlphaFold cracked protein folding in 2020, AI has revolutionized the field of structural biology. Since then, a wave of generative models have begun to work in the opposite direction, learning from vast collections of amino acid sequences, protein structures, and labels detailing their function to design proteins that do not exist in nature.

These tools remain difficult to use, however, as scientists have to guide them with formal technical specifications like structural constraints or instructions to mutate particular sites. That means trying to design something like a new plastic-degrading enzyme or a fluorescent imaging marker takes significant specialist expertise.

But in new research published inNature, scientists at Westlake University in China detail an AI system called Pinal that can generates protein sequences from natural-language prompts describing a desired protein function. And while the model outputs a lot of duds, lab tests showed that with enough tries, the system could reliably achieve a user’s goal.

“For decades, the field has largely required specialized expertise to translate desired functions into low-level structural or sequence constraints,” the authors write. “The importance of Pinal lies not only in its performance but also in reframing protein design as a problem that can be specified through functional language.”

Pinal works in two stages, neither of which asks the user for details about the molecular structure or sequence of amino acids. The first module takes the plain-language description of what the user wants and turns it into a structural blueprint. A second module then reads both that blueprint and the original text to decide the sequence of amino acids.

Crucially, the software doesn’t put all its eggs in one basket. The first module generates several candidate structures, all of which are fleshed out with their own amino acid sequences. Pinal then scores each of these combinations before settling on a final design. The researchers say this accounts for the fact that the best protein designs don’t always have the most obvious shapes.

Training the model took a vast amount of data. The researchers assembled roughly 135 million protein sequences and used large language models to write plain-language descriptions of their functions, generating about 1.7 billion protein-text pairs. They then used AlphaFold to predict the structures of the sequences.

The team used this corpus to train the two modules separately. The first learned to predict a structural blueprint from a text prompt, and the second learned to pick the right amino acid for each position in the sequence, given both the blueprint and the prompt.

To test the model, the team asked it to design four proteins: one that glows green, an enzyme that breaks down PET plastic, an enzyme that processes alcohols, and a shuttle protein that ferries chemicals between enzymes in a metabolic pathway.

The model was able to solve all four challenges, but its hit rate was a mixed bag.

When asked to produce a glowing protein, it created 50,000 sequences. These were filtered down to 29 for testing, and only one actually fluoresced. The system was better at producing the shuttle protein. Five of the 20 candidates worked, and the best one boosted a multi-enzyme carbon dioxide fixation pathway to 1.7 times the output of the natural version.

A different collaboration between Hugging Face and academics in China and Italy recently applied similar ideas to DNA sequences. The group released a 14-billion-parameter open model called Carbon-X that can answer questions about a gene sequence, predict what a mutation will do, or generate new DNA.

That model is based on Qwen3-14B, but was retrained on 300 billion tokens that mixed genomic sequences, biology text, and ordinary English. Getting it to learn DNA without forgetting English required a delicate balance though, the researchers write in a post, and forced them to do a long slow training run with a large model.

Across five benchmark tests, Carbon-X beat Opus 5.5 and Gemini 4 Argon and trailed GPT-6.1 Sol by less than a percent. And the model can run on a MacBook, which could be a major boon for researchers eager to avoid sharing sensitive or unpublished sequences with a commercial model provider.

Neither tool is perfect though. Pinal’s own developers concede that their screening heuristics “may restrict exploration to a limited region of the design space,” while Carbon-X struggles with long sequences and remains prone to hallucinations. Validating the sequences they produce also still requires a well-equipped lab and experienced scientists.

But these tools could nonetheless empower a far wider range of biologists to experiment with problems previously restricted to a handful of computational specialists.

The post Scientists Can Now Ask AI to Design Proteins Using Plain Language appeared first on SingularityHub.

David Goguen’s free chmod calculator shows what Linux permission numbers mean before you copy a command.

Zrušením koncesionářských poplatků Česko může ztratit víc. Podrobný přehled, jak dopadly podobné změny ve všech evropských zemích.

Evropský internet může působit jako decentralizovaná síť, jeho provoz však v praxi závisí na službách, firmách a institucích soustředěných v několika zemích. Jak velké je toto riziko a co s ním mohou udělat provozovatelé?

V sedmém článku o technologiích WebGL a OpenGL ES si popíšeme další možnosti, které nabízí texturovací subsystém Open GL ES. Popíšeme si filtraci textur při jejich zvětšování a/nebo zmenšování a využití blendingu.

Staré české moudro praví, že když se mají dva nebo tři rádi, na pohlaví nezáleží. Případ z Etiopie přináší názorný příklad ze zvířecí říše.

Mohutné sopečné erupce mohou na celé měsíce, někdy dokonce na několik let ochladit celou planetu. Nedávný výzkum ukázal, že v regulaci teploty na severní polokouli hrají překvapivě důležitou roli především sopky, které vybuchují v tropech. Jejich vliv je zřejmě výraznější a delší, než se dosud předpokládalo.

Bojoví roboti se stávají realitou. Válka na Ukrajině ukazuje, že pozemní roboti se stále více prosazují přímo na frontě. Mohli by se do bojových operací zapojit také humanoidní robotičtí vojáci? Technologický pokrok je působivý, ale jejich nasazení v boji zatím naráží na velmi přízemní problémy.

Po zprávách o více než 20% nárůstu IPC procesorů Intel Nova Lake přichází poněkud chladnější sprcha. Standardní verze procesu má přinést pouze ~8% posun jednojádrového výkonu…

Unknown criminals are exploiting a maximum-severity vulnerability in Node-RED, a flow-based programming tool that is widely used in critical infrastructure, to deploy a flow that downloads and runs a Linux bot. Node-RED is commonly used for workflow automation in industrial control systems (ICS) and operational technology (OT) environments. It’s built on top of Node.js and is used in commercial products including Siemens SIMATIC IoT2050. In August, Siemens assigned the vulnerability identifier – CVE-2026-58115 – to describe the bug and explain its perfect 10-out-of-10 CVSS rating. Affected Siemens devices do not enforce authentication on the Node-RED HTTP interface, allowing unauthenticated access to programming nodes that can execute system commands on the device. This means unauthenticated remote attackers can create malicious flows, which is what VulnCheck researchers began documenting last Friday. “On October 30, 2026, one of our canaries caught an attacker exploiting CVE-2026-58115,” VulnCheck researcher Patrick Garrity warned in a Monday advisory, adding that two Node-RED built-in nodes make this vulnerability “trivial to weaponize.” One is an inject node, which fires a message either on a schedule or once at startup. The second, an exec node, can run shell commands. “Wire one to the other and the flow executes a command as soon as it is deployed, with no further interaction,” Garrity said. Plus, these exploits come as criminals increasingly look to target internet-exposed OT and ICS devices, allowing their cyberattacks to impact the physical world and produce real-life harm. Case in point: the ongoing refrigeration controller hacks across Europe and the attacks on water systems across at least 12 US states over the summer. The activity observed by VulnCheck came from a single IP address, 4.201.105[.]116, which sits in Microsoft’s AS8075 network and geolocates to Brazil. “It didn't just scan us. It sent the complete exploit, flow and dropper, in a single POST /flows,” Garrity wrote. While there’s a lot that the researchers don’t yet know about the attacks - they didn’t retrieve a binary or staged scripts to analyze - the filename, bot-linux-amd64, indicates that the malware targets Linux running on AMD64 processors. It also suggests the malware operator may have other builds. “We did not fetch the binary, so we can't say what it does once running,” Garrity noted. Between October 30 and November 1, the researchers saw this IP address target the same Node-RED vulnerability 16 times, and log 125 hits against eight different CVEs across their wider canary network. Among the targets of the attack were a seven-year-old Apache Solr flaw tracked as CVE-2019-17558 – affecting Camunda, Gitea, MLflow, Traefik, and Joomla, alongside Node-RED. All this suggests “opportunistic mass scanning rather than a targeted attack,” according to the researcher. One of these hits delivered a different payload pointing to a separate server (85.215.73[.]250), which VulnCheck now tracks - but doesn’t have any telemetry other than the URL. “Because the download host is a different machine from the one that sent the exploit, the likely setup is a scan-and-infect arrangement: the hosts that find and exploit vulnerable instances pull the implant from a separate, dedicated server,” Garrity wrote. The mix of old and new bugs across unrelated products indicates “opportunistic mass scanning rather than a targeted attack,” Garrity wrote. “If you run Node-RED anywhere, including inside OT environments, check today whether it’s reachable from outside your network, apply updates where available and implement detections to reduce the risk of exploitation,” he added in a LinkedIn post. While the bug has been added to VulnCheck’s Known Exploited Vulnerabilities (KEV) catalog, it hasn’t appeared (as of press time) in the US Cybersecurity and Infrastructure Security Agency’s KEV.®

Chinese hackers used ARTEX, an open-source, multi-agent AI penetration-testing framework, in attacks against more than 200 organizations across 14 countries. [...]

A coalition of private-sector security firms working with Japan’s National Police Agency has disrupted XCPE, a backdoor used to build a massive botnet comprising more than 1.3 million Android devices across more than 200 countries. Google, Lumen, Human Security, Plume, Synthient, and TrendAI participated in the disruption. “We assess with high confidence that the backdoor and botnet were intentionally developed, deployed, and operated for financial gain by Videostrong Technology,” the Google Threat Intelligence Group (GTIG) said in a Monday blog. Videostrong Technology is a hardware manufacturer based in Shenzhen, China. It builds Android devices, including home TV boxes and projectors. According to Google, Videostrong deliberately embedded the XCPE backdoor in device firmware and sold access to criminal affiliates, allowing them to install additional malware modules on devices shipped to unsuspecting customers. When buyers connected these devices to their home networks, the preinstalled backdoor allowed operators to create proxy networks and hide malicious traffic using the customers' internet connections without their knowledge. The firmware’s pre-installed management agent purported to be a standard remote-management client, according to TrendAI. It pulled updates from a GitHub page, and connected to a Message Queuing Telemetry Transport (MQTT) server that gave the operator full remote control of each box, allowing them to install and remove apps, change settings, and push updates to the infected devices. Between June 3 and October 1, TrendAI documented more than 3.8 million distinct home and mobile IP addresses connected to the operation. The coalition also disrupted the command-and-control domains used to manage the affiliate modules. “Because these affiliates each have diversified distribution channels for their modules beyond XCPE, we assess that taking down their C2 domains will sever access to at least 10 million additional compromised devices,” according to GTIG. Google said it had since cut ties with Videostrong, and banned the OEM from making devices with preloaded Google apps. Videostrong did not immediately respond to The Register’s inquiries. This backdoor and botnet operation is part of the same ecosystem as BadBox 2.0, another notorious botnet that infected more than 10 million TV boxes and other devices before Human, Google, and TrendAI disrupted it in March 2025.®

Financial technology company Abrigo has restored its Sageworks risk management application, except for borrower-facing capabilities, following an outage triggered by "irregular activity" in its environment. [...]

Google and industry partners have dismantled the XCPE malware botnet consisting of more than 1.3 million Android TV streaming TV boxes that Google says its manufacturer intentionally infected with malware to generate revenue through cybercrime activities. [...]

Malá sluchátka typu True Wireless jsou neskutečně oblíbená. Nezůstal jediný výrobce, který by si je dovolil neprodávat. Vybrali jsme sluchátka, která patří na špičku. Některá poměrem ceny k výkonu, jiná svým zvukem a výbavou.

Some attacks this week didn't need much of a trick. A fake support call, a trusted extension, a shared link, or a device with no password. Small things that people barely think about. Yet each was enough to cause serious trouble. Then there were the bigger messes. Billions stolen, hidden networks of hacked devices, poisoned software, and attackers finding new ways to turn normal tools against Ravie Lakshmananhttp://www.blogger.com/profile/[email protected]

WorldTUI je experimentální LLM o 360 milionech parametrů, který napodobuje terminál. Podle stisknutých kláves se snaží předpovídat změny na obrazovce a zvládá tak vyonávat rozličné příkazy, například Vim, tmux i jednoduchou hru Snake. Nejedná se o skutečný příkazový řádek, jeho 'soubory' i příkazy jsou pouze simulované a model může samozřejmě různě chybovat. WorldTUI vychází z jazykového modelu Qwen2.5-Coder-0.5B, který autor doladil na záznamech skutečných terminálových relací. WorldTUI si lze vyzkoušet přímo v prohlížeči (pozor, stáhne se model o velikosti 208MB).

Josef Bacik navrhuje novou abstrakci cell, která by celý kontejner reprezentovala jako jeden kernelový objekt. Proces by vznikl přímo uvnitř cell a nemohl by ji opustit. Oprávnění ke správě by se odvozovala od držení cellfd. Návrh řeší také správu síťových namespaces, izolaci UID, předávání file descriptorů a životní cyklus procesů. RFC v2 obsahuje 67 patchů nad další třináctidílnou sérií. Autor výslovně upozorňuje, že jde o diskusi návrhu, nikoli o hotovou funkcionalitu kernelu. (pro úplnost ještě odkaz za paywallem)

OpenAI sjednocuje limity ChatGPT a Codexu. • Náročnější úlohy rychleji spotřebují společný fond tokenů. • Po vyčerpání limitu je nutné čekat nebo připlatit.