DeepSeek Tops App Store after R1 Release Last Week

⚡️ Headlines

🤖 AI

Now Wanted in Silicon Valley: Ho-Hum Businesses With Thin Profit Margins – Venture-capital firms are bringing AI and dealmaking to accounting and other low-margin industries. [The Wall Street Journal].

Meta Scrambles After Chinese AI Equals Its Own, Upending Silicon Valley – The Chinese fund shocked the AI industry in late December by releasing a version of DeepSeek that matched or outperformed models from Meta. [The Information].

Google pushes global agenda to educate workers, lawmakers on AI – Google is ramping up efforts to shape public perception and policies in the field of artificial intelligence (AI) amidst looming global AI regulations. [Reuters].

ElevenLabs has raised a new round at $3B+ valuation led by ICONIQ Growth, sources say – The New York startup has closed a Series C of $250 million at a valuation between $3 billion and $3.3 billion, multiple sources tell TechCrunch. [TechCrunch].

🦾 Emerging Tech

Musk Exploring Blockchain Use in US Government Efficiency Effort – Elon Musk has initiated conversations about using blockchain technology at the new Department of Government Efficiency, according to people familiar with the matter. [Bloomberg].

The Electric: Humanoid Robots Are Cool—but How Will Their Battery Problem Be Solved? – Content not accessible - This article is behind a paywall or inaccessible. [The Information].

🤳 Social Media

Perplexity AI proposes to merge with TikTok, with US government getting half, source says – Perplexity AI has revised its merger proposal to ByteDance, suggesting a new entity that would merge Perplexity and TikTok U.S., with the U.S. government potentially owning up to 50% following an IPO valued at a minimum of $300 billion. [Reuters].

Wall Street Banks Prepare to Sell Billions of Dollars of X Loans – Banks are hoping to sell the X debt at around 90 to 95 cents on the dollar. [The Wall Street Journal].

🔬 Research

Qwen2.5-1M: Deploy Your Own Qwen with Context Length up to 1M Tokens – The latest Qwen2.5-Turbo supporting 1M tokens is used in the same way as the standard Qwen API and is compatible with the OpenAI API. [Qwen].

⚖ Legal

DOGE Will Allow Elon Musk to Surveil the US Government From the Inside – A former USDS employee called the repurposing of the Digital Service into DOGE an “A+ bureaucratic jiujitsu move.” It will give Musk and his associates access to unclassified data in every government agency. [Wired].

President Trump Signs Executive Order To Ban Central Bank Digital Currencies (CBDC) – The President has also assigned a working group to assess the creation of a strategic bitcoin stockpile. [Nasdaq].

US securities regulator opens door for Wall Street banks to hold crypto – The US Securities and Exchange Commission has overturned a guidance that treated digital tokens as liabilities on bank balance sheets, making it expensive for banks to hold cryptocurrencies. [Financial Times].

🎱 Random

PhysicsForums and the Dead Internet Theory – LLMs are writing posts that are being back dated with human authors on PhysicsForums, a scientific community founded in 2001. [Hall of Impossible Dreams].

🔌 Plug-Into-This

DeepSeek's AI Breakthrough Captivates Silicon Valley

Last week Chinese open-source AI model DeepSeek-R1 broke the internet with a reasoning model that rivals OpenAI's o1 in performance while operating at a fraction of the computational cost. The app has since surged to the top of global app store rankings, further amplifying its influence and solidifying its reputation among users and industry leaders.

• DeepSeek-R1 matches OpenAI's o1 in tasks such as mathematical problem-solving, programming, and natural language inference, achieving a 97% accuracy rate in math problems and outperforming 96% of humans in programming tests.

• The DeepSeek app has climbed to the #1 spot in both the Apple App Store and Google Play Store across multiple regions, with millions of downloads recorded within the first week of release.

• The model's development cost was only approximately $6 million, significantly lower than the investments made by competitors like OpenAI and Meta, resulting in more affordable pricing for users.

• Industry leaders have expressed admiration for the model, with Marc Andreessen describing it as "one of the most amazing and impressive breakthroughs I've ever seen."

🇨🇳 With the amount of capital tech giants dumped into building foundation models, having one pop out of nowhere with such a small budget has shaken the tech industry, and also financial markets.

Meta to Boost Capex 60% This Year to Develop AI Data Centers

Meta Platforms plans to increase its capital expenditures by approximately 60% in 2025, allocating between $60 billion and $65 billion to enhance its AI infrastructure. This significant investment underscores Meta's commitment to advancing artificial intelligence capabilities.

• The increased spending will focus on developing AI data centers, expanding computing power, and supporting AI-driven products and services.

• This move aligns with industry trends, as other tech giants like Microsoft and Google are also ramping up investments in AI infrastructure.

• Meta's CEO, Mark Zuckerberg, emphasized the importance of AI in driving innovation and maintaining competitive advantage in the tech industry.

• The company aims to integrate advanced AI into its platforms to improve user experiences and operational efficiency.

• Investors are closely monitoring these developments, weighing the potential for long-term gains against the substantial upfront costs.

📈 Maybe Zuck was a little bummed at not being included in the Stargate deal, and this is his way of coping? Jokes aside, Meta continues to shovel money into developing AI and concerns of falling behind have never been higher with DeepSeek shaking the whole industry last week.

Siri's Declining Performance in Answering Basic Sports Queries

In a recent analysis, Siri's ability to answer straightforward sports questions, such as identifying Super Bowl winners, has been found lacking. This apparent decline in performance raises concerns about the effectiveness of Apple's much touted AI assistant.

• Siri correctly identified the winner in only 20 out of 58 instances when asked about Super Bowls I through LX, resulting in a 34% accuracy rate.

• In contrast, other AI-powered answer engines like ChatGPT, Kagi, DuckDuckGo, and Google successfully provided correct answers for the same set of questions, highlighting Siri's comparative shortcomings.

• Siri's errors included attributing multiple Super Bowl victories to the Philadelphia Eagles inaccurately and providing inconsistent responses to repeated queries.

• The analysis revealed that Siri's performance has not improved with recent updates, and in some cases, has regressed, especially when compared to its competitors.

🤖 Looks like AI hasn’t quite bailed Siri out of our collective disappointment (and sometimes anger) yet…beyond the anecdotes, there’s a mounting body of evidence that the features tech companies have been lauding like “writes your emails for you” don’t quite meet the level of quality needed to really drive widespread adoption.

 🆕 Updates

📽️ Daily Demo

🗣️ Discourse