Candidate for dwarf planet discovered beyond Pluto's orbit

Scientists made a significant discovery by finding fossilized skeletons in the province of Guizhou, China. These fossils indicate the existence of an amphibian that inhabited the saltwater waters of that region approximately 237 million years ago.

This new species, named Gerrothorax maritimus, was about 74 centimeters long and had a flattened body protected by hard scales, resembling armor. For specialists, this finding has revolutionary potential because, if it is confirmed that this Triassic amphibian was entirely marine, it would significantly alter the current understanding of how modern amphibians adapt to terrestrial and freshwater life.

Currently, only 2% of amphibians show tolerance to seawater. Unlike oceanic fish, which have specific glands to expel excess salt, most amphibians have thin and permeable skin, which is essential for breathing and absorbing fresh water. However, this same characteristic makes them vulnerable to rapid dehydration when exposed to saltwater.

Ryosuke Motani, the study's author, explained to Reuters that when a frog swims in saltwater, it loses water through its permeable skin due to osmotic pressure, and it also cannot drink water for the same reason, resulting in weight loss and the need to recover in fresh water. This limitation prevents amphibians from becoming completely marine.

Gerrothorax maritimus managed to survive in saltwater thanks to a unique trait: its bony armor. These protective structures, composed of bone plates, served to defend against predators in freshwater environments but took on a distinct function in the sea. The armor is thought to have prevented salt absorption through the skin, forcing the animal to ingest seawater through its mouth and subsequently filter the salt through its gills. Motani described this process as exaptation, when a biological trait acquires a different function than its original one.

Comparative anatomical analysis indicates that the closest relative to Gerrothorax maritimus is Gerrothorax pulcherrimus, another extinct amphibian from the Triassic period. This relative also had armor and a flattened body but lived exclusively in freshwater. Researchers concluded that the marine lifestyle of Gerrothorax maritimus was an accidental evolutionary event.

Although the hypothesis arose that the skeleton might have been transported to the sea after death, researchers maintained their belief that it was a marine creature. This is due to the fact that its bones were found very far from any landmass of the time, suggesting they were not carried by ocean currents.

Additionally, the team analyzed the animal's isotopes to determine its diet and habitat. This analysis helped date minerals and water in marine samples. The levels found in the bones matched those of Triassic seawater, indicating that these creatures probably 'grew drinking seawater,' as reported in the study published in the journal Nature Ecology & Evolution.

Google announced that Gemini now has the capability to edit various types of files, including PDFs, presentations (slides), spreadsheets, programming code, and documents, all within its own interface, eliminating the need for external editing software.

This functionality is being made available to all users of the service. Although Gemini already allowed working with content such as text documents and spreadsheets, Google's announcement refers to a significant advancement in how the tool handles files in different formats.

Previously, to modify a piece of text created by Gemini, it was necessary to provide a specific instruction (prompt) to make the change. With this update, the user can make modifications directly in the service's interface, simulating Gemini acting as a text editor itself.

This improvement allows a file to be generated, viewed, or edited entirely through the tool's interaction screen. However, Google emphasizes that applicability varies depending on the device used: direct viewing and editing of PDFs are supported on both the web version and the Gemini mobile apps.

Other formats, such as Google Docs, Google Sheets, slides, Markdown, TXT, and programming code, are currently only available on the web version. According to Google, this new feature is being launched globally and in all languages where the service operates.

It is important to note that viewing and editing these files is accessible to both free Gemini users and those who use paid Google AI plans, which in Brazil start at R$ 24.99 per month under the name Google AI Plus.

Mistral AI has made the weights of Large 4, an artificial intelligence model with one trillion parameters, available. This model can be downloaded and operated by governments and companies within their own systems. Known as "Le Chonk," this AI was designed to assist in various functions, such as identifying software flaws, analyzing digital threats, and supporting responses to cyberattacks.

The launch occurred this Thursday (29) and consolidates the French company's strategy of promoting open-weight models, which grants greater control over the technology. The files are accessible on Hugging Face, and the tool can also be used via the Mistral API.

The initiative aims to reduce dependence on AI services managed by external providers. This is particularly important for entities that handle confidential data and need to investigate security incidents without transferring their information to third-party platforms.

Cybersecurity constitutes one of the pillars of this launch. Large 4 was created to support technical teams in locating vulnerabilities, replicating flaws to confirm their authenticity, and generating fixes for detected issues.

According to Mistral, the model achieved a score of 82% in a test conducted by the Artificial Analysis Cyber Index. This index evaluates the model's ability to replicate a real vulnerability in open-source software and subsequently correct it. Additionally, the company reports that the AI solved 93% of the Cybench challenges, a set composed of 40 exercises inspired by digital security competitions.

These results are disclosed by the developer itself and should be analyzed considering the specific circumstances of each evaluation. In addition to flaw detection, Mistral points to uses for malware analysis, vulnerability prioritization, and developing rules to identify threats.

The company also highlighted the model's performance in programming. In the Artificial Analysis Coding Agent Index, Large 4 achieved a combined score of 49.8%, according to data presented by the developer. Mistral claims that this achievement positions it ahead of other open-weight models in programming tasks executed by AI agents.

Weights refer to the numerical values acquired during the training process that enable the model to generate its responses. By releasing these files, organizations gain the ability to run Large 4 on their own infrastructure instead of solely depending on an external service. This provides greater autonomy over the data, the tool's configuration, and the adopted security guidelines. In critical scenarios, for example, a corporation can use the model to investigate an attack without being subject to the availability or limitations of a third-party API.

In a conversation with TechCrunch, Pierre Stock, Vice President of Science at Mistral, stated that making the weights available allows companies and governments to access advanced AI resources to protect themselves against attacks. According to the executive, the model can assist in identifying, coordinating, and responding to major security incidents in real time. He also advocated for expanding access to these capabilities to strengthen the digital protection ecosystem.

This strategy puts Mistral in competition with other companies, such as OpenAI and Anthropic, whose more sophisticated models are predominantly offered through services controlled by the developers themselves. However, the freedom to run a model without depending on an external provider brings a disadvantage: after the weights are released, it becomes more complex to monitor who is using the technology and for what purposes.

The same vulnerability analysis functionality that assists security teams may be of interest to malicious actors seeking to exploit weaknesses in systems. Furthermore, independent operation allows organizations to tailor the model to their needs, but it implies that restrictions imposed on a hosted service do not always apply to all modified versions.

Stock acknowledged the inherent risks of AI advancement and autonomous agents but argued that there is no need to increase fear regarding the technology, as security is a challenge that can be addressed. Mistral maintains that Large 4 was developed specifically to strengthen cybersecurity, possessing the capacity to perform technical tasks that more restrictive systems might reject, even in legitimate security investigations.

With this launch, the company expands its range of advanced models that can be implemented and customized by users themselves. The great challenge lies in finding the balance between providing access to powerful digital defense tools and managing the dangers of their misuse.