This week, Depseek launched an updated version of its R1 model in HuggingFace, the open source competition reab. The updated version, called DEEKSEEK-R1-0528“ It has 685 billion parameters, an update of January version“ That had 671 billion.
Unlike Openai and Google models, which are famous code closed, Depseek’s model weights are publicly notable. According to the reference points, the R1-0528 update has improved the reasons and inference capabilities and is closing the gap with O3 and Gemini 2.5 Pro of Google.
Deepseek also introduced a distilled version of R1-0528 using the QWEN3 8b model of Alibaba. This is an example of a light model that is less capable but also requires less computer power. Deepseek-R1-0528-QWen3-8B exceeds the last light model of Google Gemini-2.5-Flash-Thinking-0520 and O3-MINI of OpenAI at certain reference points. But the biggest problem is that Deekseek’s distilled model can Run in a single GPUAccording to Techcrunch.
To … distill all this information, the Chinese rival is up to date with its US competitors with an open weight approach that is cheaper and more accessible. In addition, Depseek continues to demonstrate that AI models may not require as much computer power such as Operai, Google and other heavyweights that currently use. It is enough to say that looks at this space.
That said, Depseek models also have their inconveniences. In accordance with an AI developer (through Techcrunch), Depseek’s new update is Even more censored that your Previous version When it comes to criticism of the Chinese government.
Of course, it happened much more in the world of AI in recent days. After the AI event parade last week by Google, Anthrope and Microsoft, this week was the light in the product and the news of functions.
That is a reason why Deepseek’s update captured AI’s attention this week. In other news of AI, Anthrope finally obtains the voice mode, the influencers of AI become viral, the CEO of Anthrope warns about mass dismissals and a kangaroo generated by ia.
I see Google 3 takes by Internet assault
In Virtumy, each social networks platform, users are driving crazy about the new Video Video Model from Google. The results are impressions, and we are already watching short films made completely with I see 3. It is not bad for a product that came out 11 days ago.
It should not be surpassed by AI Video Artists, a reporter from The Wall Street Journal He made a short film about her and a robot using I see 3.
Mashable technology editor, Timothy Werth recapitulated I see the great week and had a simple conclusion: we are very cooked.
After last week, this week was lighter in the volume of AI News. But what wax Announced this week is no less significant.
Anthrope finally introduced his Voice mode for Claude To compete with Chatgpt, Grock and Gemini. The function is currently in beta on mobile devices for the Claude application and you can even see free plans with a limit of 20 to 30 voice conversations per day. Anthrope says that he can ask Claude to summarize his calendar or read documents aloud. Subscribers who pay can connect to Google’s work space for access to the calendar, Gmail and Docs
Operai is exploring the ability to Log in third -party applications with chatgpt. We still do not know much, but the company published an interest form on its site for developers to use Codex, their engineering agent, to add this capacity to their own applications. It may not seem like a big problem, but basically it means that users could easily link their memories and custom chatpt configurations with third -party applications, as well as the way it works in a new application with its Google account.
Opera announced a new Agent browser called neon. “Much more than a place to see web pages, Neon can navigate with you or you, take action and help you do things,” said the announcement. That includes a chatbot interface within the browser and the ability to complete web forms for tasks such as the Reserve of Travel and Purchases. The announcement, which included a promotional video of a humanoid robot that navigates the robot, which scan in details, but says that Neon will be a “premium subscription product” and has a waiting list to register.
The browser has suddenly become a new border for the Agent AI, now that is capable or automation of web search tasks. Perplexity is working on a similar tool called Kiteand the browser company rotary From its arc browser to a more centered browser called day. All this is happening while Google could be Forced to sell Chromewhich Operai has kindly offered To take off your hands.
Dario Amodei’s prediction on the AI that replaces the entry level works is already beginning to happen
Anthropic CEO Dario Amodei warned In an interview with Axios that AI could “eliminate half or all white -collar input work.”
Amodei’s predictions can be perfect because a new study by the Risk Capital firm, Signalfire, found that Hiring for entry level works is inactive to 7 percent of 25 percent in the previous year. Some of that is due to changes in the economic climate, but the AI is definitely a factor since companies are opting for automatic aspects of the less technical aspects of work than the tasks of the new and legs.
The latest in the culture of AI: that kangaroo generated by AI, Judge Judy and everything else
Google wants you to know that its descriptions of AI reach 1.5 billion people per month. They probably do not do it because you know that you know the descriptions of the AI still fight to tell, spell and know what year it is. As Tim Marcin said as a mashable, would the AI summarize the Boil Shock Protocol?
The proposal of a 10 -year prohibition of the states that regulate the AI is Quite unpopularAccording to a media survey. The survey found that 57 percent of respondents opposed the moratorium, including half of the Republican respondents. As Rebecca Ruiz de Mashable reported, “the vast majority of respondents, regardless of their political affiliation, agreed that Congress should not prohibit States from enforcing their own youth security laws.”
In the private sector, The New York Times signed a License Agreement with Amazon To allow its editorial content to be used for AiAzon AI models. The details are not clear, but from the outside, this seems like a change of melody of the Times, That is currently Demanding Openai For copyright violation for supposedly using their content to train their models.
That viral video or a kangaroo of emotional support that has a plane ticket and is denied the approach? Is generated by theor course. A little more obvious, but no less spooky is another viral tendency to use AI to convert public figures such as Emmanuel Macron and Judge Judy In babies. These are strange infested times of AI-Slop in which we live.
AI also has some positive uses. This week, we learned about a new Huggingface humanoid robot called Hopejr (with Robot Studio engineering), which could be Avia for sale at the end of this year for only $ 3000.
And to finish this summary in a high note, the non -profit colossal foundation has developed a AI algorithm To detect the calls of the birds of the almost extinct pigeon with a tooth peak. Also known as “Little Dodo”, La Paloma with Dental Pico is the National Bird of Samoa, and scientists are using the bioacoustic algorithm to locate and protect them.