Welcome back to In the knowThe new Time Time newsletter twice a week about artificial intelligence. If you read this in your browser, why not subscribe to the next one to be delivered directly to your mailbox?
What to know: the dangers of browsers AI
Last week, “embarrassment” announced that its browser with AI, called Comet, would be made free of all users after a paid subscription was previously required. A comet is a new type of browser containing an integrated chat -Bot AI, which can view the Internet on your behalf, as well as perform autonomous actions, such as making purchases, sending e -letters or creating the events of a calendar.
There is only one problem. The internal and comet could be captured by evil links that forced the Siphon browser personal information from connected services, such as Gmail, and send it to the mockering, according to the new research Layerx cybersecurity company. This is a sign that even despite the fact that browsers with AI can make users more productive, they can also represent new vulnerabilities.
Attack – Layerx discovered a vulnerability called “Cometjacking”, in which a malicious hint for the artificial intelligence of the browser is hidden in the URL. When the user presses this link, the browser is mistaken in a malicious clue for instructions from the user – and begins to release it. In the example of Layerx, an attacker, an attacker, it is possible to force COMET to extract data from the user and calendars e -mail accounts. While the comet has guarantees from the theft of data, the attacker was able to bypass them by indicating AI to encode the stolen information in Base64 (in fact, scribbled to look like a harmless text) before sending it to a remote server under their control.
A sign of the future—Thoday, Google Chrome, of course, is the most popular browser. But some believe that a new “war on the browser” may soon appear, fed by new participants, such as the comet of “immaturity”. (Openai is also considered working According to Layerx CEO, he can also add new vulnerability classes, but since the creators of the browser, the creators leading to the addition of artificial intelligence function, they can also add new vulnerability classes. Perhaps we are going to enter the “world in which the viewing becomes more risky,” says Azhtes. “We will see the old types of attacks that have almost died out, or even new types of attacks, similar to those that we have just discovered.”
The answer of embarrassment “When Layerx announced the bewilderment of vulnerability last month, the company“ replied that it could not determine any impact on safety, ”Layerx wrote in the blog post. But in his statement of time, a representative of bewilderment said that the Layerx error report was poorly formulated, that he did not respond to requests for explanations, and that bewilderment “later revealed the problem independently and corrected.” The representative said that vulnerability was never operated, and continued: “We are grateful to the security community that participates in our prosperous generosity program, and we are working to ensure that these types of misunderstanding do not occur in the future.”
If you have a minute, please take ours quickly survey To help us better understand who you are, and what topics of AI interest you the most.
Who know: Lisa Su, general director AMD
On Monday, Chipmaker Advanced Micro Devices (AMD) announced that she had concluded a deal with multimillions of dollars with Openai, which would acquire a ChatGPT 6 Gigavatt-Brass manufacturer from its last chip AI for several years and the equivalent of about 4 million houses in the USA. This is just the last megadeal for Openai, which last month received an agreement on $ 100 billion with NVIDIA, since it is looking for the computer power necessary for the training and launch of its AI models, including last week last week. Sora 2Field
When I am Sitting with Lisa Su last year, she was at the height of her company to build a new type of accelerator chip of artificial intelligence. Her efforts seem to have paid off. AMD remains in a remote second place for NVIDIA in a wider semiconductor industry, but the Openai deal is confident that AMD chips are suitable for the workload of Topline AI. This is also another sign that Openai seeks to reduce its dependence on NVIDIA, even if it brings the company of Jensen Huang.
Openai and AMD did not disclose the contract value in dollars, although they said that this gives Openai the right to purchase up to 10% of the company. AMD shares jumped by 25% in the news.
AI in action
About a week after the launch of their application for generating the Sora, Openai video, users allows users more options to limit how their similarities appear in the video. Previously, users had the opportunity to either allow or prohibit their similarity (called “Camoo” in the open language) in the Sora video. Now users can give instructions such as “do not put me in the video that include a political commentary” or “do not let me say this word.” in accordance with Head of Sora Openai, Bill Pibls.
What we read
DizzinessFor money
This time is not an article, but a very readable book that was engaged in an integral rise in China. Dan Vang's main argument is that China is Engineering state, focused on construction at any cost, while the USA Law Societywhere it is easier to stop the construction of things than to build them. The result, as Van sets out in convincing details, is that in China there is now brilliant public infrastructure and a reliable production economy, while the United States has not created large public works in decades and forgets about its production know-how. Dizziness Raises the sobering question about AI, since the US economy doubles the technology: what use is a plentiful digital intelligence if it comes to the economy that forgot how to apply it?
