AI Unfiltered by Xenoss: Issue #2, March 2025
We created AI Unfiltered by Xenoss to add more context to headlines and media coverage.
In this second edition, we continue gathering insights from AI industry leaders to cut through the noise, helping you understand not just what's happening—but why it matters.
As always, in this newsletter, we collect thoughts of top minds in AI unpack the significance of recent breakthroughs, discuss meaningful industry shifts, and forecast the impact of new technologies.
In today's edition:
Now, let’s dive right into the latest AI buzz.
Apple AI enhancements on Siri are delayed
What happened
Last June, at WWDC 2024, 苹果 unveiled Apple Intelligence and teased personalization updates to Siri. An AI-powered Siri would be able to answer more complex questions and consider a user’s context.?
In an ad dedicated to the iPhone 16 release, Apple showcased Siri’s improved personalization, leading everyone to believe that the new features will be part of the release.?
However, on March 7th, the company announced that it is struggling to deliver the new features in an incredibly vague statement.?
“It’s going to take us longer than we thought to deliver on these features, and we anticipate rolling them out in the coming year.”
Most likely, by referring to the “coming year”, Apple implies we won’t see a personalized Siri until 2026.?
Why it matters: When all big tech companies are fighting for AI superiority, Apple has failed to make any substantial progress. Its establishment of Apple Intelligence has already been long overdue, but underdelivering on highly anticipated features proved how disorganized Apple’s AI operations are.?
Community response
Tech journalists, industry insiders, and influencers all weighed in on the matter, with takes that had few to no words of praise for Apple’s recent decisions and leadership.?
Mark Gurman , 彭博资讯 Managing Editor & Chief Correspondent shared a series of tweets highlighting that the challenges inside Apple Intelligence were well-known to insiders.?
He shared that new Siri features had significant quality issues and, according to a leaked Siri team meeting, only worked for about one-third of the time.?
David Heinemeier Hansson , the creator of Ruby on Rails and Shopify board member stated that Siri delays are a sign that Apple is losing its innovative spirit and needs a “leadership reboot”.?
Bojan Tunguz, Ph.D. , ex-ML at 英伟达 called out Apple for “missing the boat” on AI and failing to catch up to its big tech rivals.?
Tim Urban He also expressed his surprise at Apple's struggle to build a high-performance LLM, given that multiple AI startups have successfully pulled it off.?
Google releases Gemini 2.0. Flash with image generation
Google’s Gemini 2.0. Flash is a powerhouse on multiple levels: It has memory, it can reference a user’s search history for better personalization, and users believe that its image-generation tool can make Photoshop obsolete.?
Also, unlike Google’s previous models, Gemini 2.0. Flash is a chain-of-thought model that allows users to track the algorithm’s reasoning step by step.?
Community response: After having tried Gemini 2.0. Flash, the AI community took to X to share promising use cases.?
Deedy Das , former Google and Facebook employee and now an investor in AI technology, pointed out that Google’s image-generating model can instantly remove furniture from an apartment and fill the empty space with a new design, which would normally cost up to $10k.?
Halim Alrasihi , AI educator, showed that the image generator can help choose an outfit for a model in a picture, which has a wide range of applications in fashion and e-commerce.?
George Arrowsmith , an Elixir engineer, discovered a viral use case for Gemini 2.0. Flash: removing watermarks from images.
Immediately, the AI community erupted in concerns about how ethical such an application is to content creators trying to protect their intellectual property.?
Sakana.ai AI-generated a machine learning paper that successfully passed peer review?
The Japanese startup used its internal AI model, AI Scientisti-v2, to generate a paper that passed peer review at ICLR, a well-known machine learning workshop in the industry. According to the company’s statement, the AI-generated paper was written in collaboration with the University of Oxford and the University of British Columbia.?
Ultimately, Sakana.ai withdrew its paper from the conference to respect the transparency and conventions of the workshop.?
Community response
The AI community quickly picked up on nuance news stories that might have been missed, such as the fact that the published paper was a negative-results paper, so it did not contain any innovative proposals.?
When Damien Teney pointed out that Sakana.ai’s milestone should be taken with a grain of salt, the startup responded emphasizing the importance of negative results research.?
Commenting on?Sakana.ai’s achievement, Hugo Larochelle, a researcher at DeepMind and a machine learning professor, points out that AI could drive a commonly disregarded area of research: Reproducibility studies.?
Telekom and Perplexity partner to release an 'AI Phone'
Deutsche Telekom IT Solutions HU said that it is building a low-cost “AI Phone”, set to cost under $1,000 and hit the European market in 2026.?
If all goes well, the deal is a win-win decision for both parties: DT gets a shot at becoming an AI company, while Perplexity gets to penetrate the mainstream customer segment.?
Community response
Although it seems promising on paper, the announcement provoked reactions that were skeptical, to say the least.??
AI enthusiasts believe the high price for the AI Phone to be unreasonable and doubt that the product will manage to take hold successfully.?
领英推荐
Anthropic’s $3.5B raise at $61.5B valuation
The $3.5 billion Series E round, led by Lightspeed Venture Partners, with the participation from Bessemer Venture Partners, Cisco Investments, and other notable VCs in the AI space, follows the success of Sonnet 3.7, which we covered in our last newsletter.?
What will Anthropic do with all that money? The company’s official statement focuses on ensuring productive human-computer interactions and continued efforts in AI alignment (which Dario Amodei , the CEO of Anthropic, has long advocated for).?
Community response
The AI community believes that the new funding round, coupled with the excellent performance of new Claude across all benchmarks, puts Anthropic ahead of the competition.?
China’s heavy involvement in the AI race fuels US-China tensions. As the impact of AI increases, machine learning technologies become a geopolitical weapon, and leading startups in the field start playing an increasingly important role in national security matters.?
OpenAI is attempting to ban DeepSeek in the US, calling the model “state-owned”. The company also urges the government to ban the use of all models built in the People’s Republic of China.?
This move led to a backlash directed towards OpenAI and Sam Altman.?
Ashutosh Shirastava, AI influencer on X, calls OpenAI out for being “pathetic” and “hypocrites”, considering that OpenAI is involved in the government-supported Stargate Project.?
Kai-Fu Lee , AI expert, former President of Google China, and the author of AI Superpowers, wonders if it is possible to ban an open-source model like DeepSeek.?
On the other hand, DeepSeek has indeed been making headlines for increased state involvement in the control of key employees.?
Amir Efrati , Executive Editor at The Information, shared an excerpt of the article,?“DeepSeek, a National Treasure in China, is Now Being Closely Guarded,”?in an X post. The article claims that DeepSeek’s ML engineers are now forced to “hand in their passports” as they are now privy to “confidential information” that could constitute “state secrets”.?
This news caused a similar shockwave in the AI community.?
While some may be wondering what there’s left to protect at already open-source DeepSeek, the author of the tweet argues that this strategy is part of the “talent war” China is waging with the West.?
Why AI should not discourage programmers from learning coding
(by Andrew Ng )
In a to-the-point write-up for DeepLearning.ai, a machine learning newsletter, Andrew Ng, co-founder of Coursera and Adjunct Faculty at Stanford, gives the alarmism around computers replacing programmers a reality check.?
It is an insightful read from someone deeply involved with the field of machine learning and a glimmer of hope for AI engineers who could understandably grow worried seeing how good Claude 3.7 is. and other LLMs are becoming at coding.?
Why 2042 will be a big year for AI
(by Jürgen Schmidhuber )
TED recently published a TED talk given by Juergen Schmidhuber, one of the pioneers in machine learning, on the evolution of progress in AI. Schmidhuber uses his deep understanding of biology, physics, and technology to make an educated prediction of what the near future for AI will look like.?
Today's Quick Laugh might not be terribly funny. At least, not if you are a social media manager worried if AI is coming for your memes. We have bad news for you: It might be.
A group of European researcher released a paper "One Does Not Simply Meme Alone", analyzing how AI-generated memes compare to human content across three criteria: Humor, creativity, shareability.
On average, AI blew humans out of the park across the board. However, top-performing memes were still human-generated.
The best solution, as expected, was AI-human collaboration. The blend of the two helped generated more ideas (which performed better as well).
So, if you are a meme-maker yourself, don't worry about AI beating you at the meme game, embrace it.
Here's the breakdown of the AI vs human meme showdown.
In the last three years, the landscape of open-source AI tools has been growing by leaps and bounds.?
This Open-Source AI Stack chart summarizes the key (though not all) players of the ecosystem and gives AI builders a starting point for kicking off their machine-learning projects.?
Why Scope3’s agentic media platform could transform AdTech
The story: Scope3 recently unveiled a new agentic media platform that promises to reinvent programmatic advertising. Announced on March 13, the platform enables partners across the AdTech ecosystem to create AI-powered media products by replacing legacy infrastructure with intelligent automation.
Rather than using AI as a mere enhancement tool, Scope3’s approach optimizes media transactions at the impression level, potentially allowing advertisers to achieve better targeting, reduce unnecessary spending, and improve sustainability. The platform includes a centralized hub for agentic media products, scaled curation capabilities, and built-in brand safety features.
Xenoss perspective: Scope3’s agentic media platform isn’t just another shiny object in the AdTech showcase – it’s potentially a fundamental rethinking of how digital advertising functions. But whether it’s truly revolutionary or merely evolutionary remains an open question.
At its core, Scope3 is attempting to thread a difficult needle. Brian Kelley , the company’s founder who previously built AppNexus into an AdTech powerhouse, appears to be constructing what he describes as “an operating system for agents” – essentially positioning Scope3 as the infrastructure layer for AI-driven advertising. This is a considerably more ambitious play than simply creating another intermediary in an already crowded landscape.
What truly sets Scope3’s approach apart is the structural shift from using AI as a bolt-on enhancement to making it the fundamental architecture of media transactions. While traditional AdTech uses AI to incrementally improve existing systems – like upgrading components of a car while keeping the same basic vehicle – Scope3 is rebuilding the engine from scratch. By optimizing at the impression level, their system makes independent, real-time decisions for each ad opportunity rather than applying broad rules across entire campaigns. This granular approach potentially allows for more intelligent, nuanced advertising decisions that simultaneously consider performance, brand safety, and sustainability factors.
Bottom line
The stories covered in this edition of AI Unfilered by Xenoss show how quickly the multimodal LLM landscape is evolving.
With Gemini 2.0 Flash, Google is taking the lead but the race is still tight, especially now that Chinese gen AI advancements are essentially not inferior to American counterparts.
From Apple's Siri meltodwn, team leaders could learn how much sitting on the sidelines of the AI race costs even big tech companies. Failing to keep up with the innovation pressure cooker can put market leaders in a tough spot.
This is it for today's AI Unfiltered by Xenoss. To not miss a beat on AI news, subscribe to the newsletter and follow Xenoss on LinkedIn.
See you next time!