This week, DeepSeek released an updated version of its R1 model on HuggingFace, reigniting the open-source versus closed-source competition. The updated version, called DeekSeek-R1-0528, has 685 billion parameters, an upgrade from January’s version, which had 671 billion.
Unlike OpenAI and Google’s models, which are famously closed-source, DeepSeek’s model weights are publicly available. According to the benchmarks, the R1-0528 update has improved reasoning and inference capabilities and is closing the gap with OpenAI’s o3 and Google’s Gemini 2.5 Pro.
DeepSeek also introduced a distilled version of R1-0528 using Alibaba’s Qwen3 8B model. This is an example of a lightweight model that is less capable but also requires less computing power. DeepSeek-R1-0528-Qwen3-8B outperforms both Google’s latest lightweight model Gemini-2.5-Flash-Thinking-0520 and OpenAI’s o3-mini in certain benchmarks. But the bigger deal is that DeekSeek’s distilled model can reportedly run on a single GPU, according to TechCrunch.
To… distill all this information, the Chinese rival is catching up to its U.S. competitors with an open-weight approach that’s cheaper and more accessible. Plus, DeepSeek continues to prove that AI models may not require as much computing power as OpenAI, Google, and other AI heavyweights currently use. Suffice to say, watch this space.
That said, DeepSeek’s models also have their drawbacks. According to one AI developer (via TechCrunch), the new DeepSeek update is even more censored than its previous version when it comes to criticism of the Chinese government.
Of course, a lot more happened in the AI world over the past few days. After last week’s parade of AI events from Google, Anthropic, and Microsoft, this week was lighter on product and feature news.
That’s one reason DeepSeek’s R1 update captured the AI world’s attention this week. In other AI news, Anthropic finally gets voice mode, AI influencers go viral, Anthropic’s CEO warns of mass layoffs, and an AI-generated kangaroo.
Google’s Veo 3 takes the internet by storm
On virtually every social media platform, users are freaking out about the new Veo 3, Google’s new AI video model. The results are impressive, and we’re already seeing short films made entirely with Veo 3. Not bad for a product that came out 11 days ago.
Not to be outdone by AI video artists, a reporter from The Wall Street Journal made a short film about herself and a robot using Veo 3.
Mashable’s Tech Editor Timothy Werth recapped Veo’s big week and had a simple conclusion: We’re so cooked.
More AI product news: Claude’s new voice mode and the beginning of the agentic browser era
After last week’s barrage, this week was lighter on the volume of AI news. But what was announced this week is no less significant.
Anthropic finally introduced its own voice mode for Claude to compete with ChatGPT, Grok, and Gemini. The feature is currently in beta on mobile for the Claude app and will even be available to free plans with a limit of 20 to 30 voice conversations per day. Anthropic says you can ask Claude to summarize your calendar or read documents out loud. Paying subscribers can connect to Google Workspace for Calendar, Gmail, and Docs access.
OpenAI is exploring the ability to sign into third-party apps with ChatGPT. We don’t know much yet, but the company posted an interest form on its site for developers using Codex, its engineering agent, to add this capability to their own apps. It may not sound like a big deal, but it basically means users could easily link their personalized ChatGPT memories and settings to third-party apps, much like the way it works when you sign into a new app with your Google account.
Opera announced a new agentic AI browser called Neon. “Much more than a place to view web pages, Neon can browse with you or for you, take action, and help you get things done,” the announcement read. That includes a chatbot interface within the browser and the ability to fill in web forms for tasks like booking trips and shopping. The announcement, which included a promo video of a humanoid robot browsing the robot, which is scant on details but says Neon will be a “premium subscription product” and has a waitlist to sign up.
The browser has suddenly become a new frontier for agentic AI, now that it’s capable of automating web search tasks. Perplexity is working on a similar tool called Comet, and The Browser Company pivoted from its Arc browser to a more AI-centric browser called Dia. All of this is happening while Google might be forced to sell off Chrome, which OpenAI has kindly offered to take off its hands.
Dario Amodei’s prediction about AI replacing entry-level jobs is already starting to happen
Anthropic CEO Dario Amodei warned in an interview with Axios that AI could “wipe out half of all entry-level white-collar jobs.”
Amodei’s predictions might be spot on because a new study from VC firm SignalFire found that hiring for entry-level jobs is down to 7 percent from 25 percent in the previous year. Some of that is due to changes in the economic climate, but AI is definitely a factor since firms are opting to automate the less-technical aspects of work that would’ve been taken on by new hires.
The latest in AI culture: That AI-generated kangaroo, Judge Judy, and everything else
Google wants you to know its AI overviews reach 1.5 billion people a month. They probably don’t want you to know AI Overviews still struggles to count, spell, and know what year it is. As Mashable’s Tim Marcin put it, would AI Overviews pass concussion protocol?
The proposal of a 10-year ban on states regulating AI is pretty unpopular, according to a poll from Common Sense Media. The survey found that 57 percent of respondents opposed the moratorium, including half of the Republican respondents. As Mashable’s Rebecca Ruiz reported, “the vast majority of respondents, regardless of their political affiliation, agreed that Congress shouldn’t ban states from enacting or enforcing their own youth online safety and privacy laws.”
In the private sector, The New York Times signed a licensing deal with Amazon to allow their editorial content to be used for Amazon’s AI models. The details are unclear, but from the outside, this seems like a change of tune from the Times, which is currently suing OpenAI for copyright infringement for allegedly using its content to train its models.
That viral video of an emotional support kangaroo holding a plane ticket and being denied boarding? It’s AI-generated, of course. Slightly more obvious, but no less creepy is another viral trend of using AI to turn public figures like Emmanuel Macron and Judge Judy into babies. These are strange AI-slop-infested times we’re living in.
AI has some positive uses too. This week, we learned about a new humanoid robot from HuggingFace called HopeJr (with engineering by The Robot Studio), which could be available for sale later this year for just $3,000.
And to end this recap on a high note, the nonprofit Colossal Foundation has developed an AI algorithm to detect the bird calls of the near-extinct tooth-billed pigeon. Also known as the “little dodo,” the tooth-billed pigeon is Samoa’s national bird, and scientists are using the bioacoustic algorithm to locate and protect them.
Source: https://sea.mashable.com/
