Good morning! This is your daily ☕️ Techpresso.

In today's Techpresso:

🖼️ DeepSeek is coming for Midjourney and DALL-E

🛍️ Microsoft is in talks to acquire TikTok

💥 Meta creates 'war rooms' for DeepSeek

🤖 Alibaba releases AI models that can control PCs and phones

⌚ Pebble smartwatch returns with open-source software

🧠 Meta AI will use its ‘memory’ to provide better recommendations

🎁 + 7 other news you might like

🔮 + 1 handpicked research papers and tools

🖼️ DeepSeek is coming for Midjourney and DALL-E LINK

DeepSeek's new open-source AI model, Janus Pro, surpasses OpenAI's DALL-E 3 and other top models such as PixArt-alpha and Emu3-Gen on benchmarks like GenEval and DPG-Bench.

Janus Pro, which features a novel autoregressive framework, can process and generate images at 768x768 resolution, boasting versatility in analyzing and creating visuals.

Despite its broad capabilities, Janus Pro requires careful prompting for optimal results and currently lacks compatibility with traditional UIs, necessitating command-line usage for local deployment.

🛍️ Microsoft is in talks to acquire TikTok LINK

Microsoft is in discussions to acquire TikTok due to increasing national security concerns about the app's Chinese ownership, with President Trump confirming the tech company's interest in the purchase.

The U.S. government is pressuring TikTok to sell its American operations or face a possible ban, and Trump has expressed interest in seeing competition in the bidding process.

This is not Microsoft's first attempt to buy TikTok; in 2020, they were a leading contender to acquire TikTok's U.S. operations, but those negotiations eventually collapsed after Trump left office.

💥 Meta creates 'war rooms' for DeepSeek LINK

DeepSeek, a Chinese AI startup, quickly became the top free app on the App Store, surpassing ChatGPT, and causing a significant $400 billion market cap loss for NVIDIA.

Meta has formed four "war rooms" with specialized teams to study DeepSeek's cost-effective AI model development and its competitive edge over established rivals like ChatGPT.

Two Meta teams will analyze DeepSeek's cost-cutting strategies, another will investigate its training data, and the last will explore redesigning Llama's architecture to rival Chinese AI technology.

🤖 Alibaba releases AI models that can control PCs and phones LINK

Alibaba's Qwen team introduced the Qwen2.5-VL AI models that can perform tasks such as text and image analysis, video comprehension, and even control PCs and phones.

Qwen2.5-VL surpasses competing models like OpenAI's GPT-4o and Google's Gemini 2.0 Flash in evaluations involving video understanding, mathematics, and document analysis.

The Qwen2.5-VL models have constraints on certain topics, reflecting Chinese regulatory standards, and are available for testing on Alibaba's Qwen Chat app and Hugging Face platform.

⌚ Pebble smartwatch returns with open-source software LINK

Eric Migicovsky, the original founder of Pebble, is restarting the smartwatch brand, focusing on creating a new device using open-source software from Google, despite the brand's previous closure and acquisition by Fitbit.

Google has decided to open-source Pebble's software stack, allowing for new third-party hardware development, which Migicovsky plans to leverage for his new smartwatch venture, currently self-funded with a modest scope.

The new Pebble aims to fulfill specific user needs with features like an always-on e-paper screen, long battery life, and hackability, as Migicovsky seeks to cater to a niche audience still interested in the original Pebble experience.

🧠 Meta AI will use its ‘memory’ to provide better recommendations LINK

Meta is expanding its AI chatbot's memory feature to remember user preferences and interests, using past conversations and account details from Facebook and Instagram for better recommendations.

The memory feature is now available on Facebook, Messenger, and WhatsApp on iOS and Android in the US and Canada, allowing Meta AI to adapt responses based on user interactions.

Meta's AI will only retain information from one-on-one chats, not group conversations, and users have the option to delete stored memories at any time for privacy control.

Other news you might like

• Google Maps will soon rename Gulf of Mexico to ‘Gulf of America’ LINK

• Sam Altman calls DeepSeek R1 ‘impressive’, promises to ‘deliver much better models’ LINK

• Former Intel CEO Pat Gelsinger is already using DeepSeek instead of OpenAI at his startup, Gloo LINK

• With iOS 18.3, Apple Intelligence is now on by default LINK

• Facebook flags Linux topics as 'cybersecurity threats' — posts and users being blocked LINK

• Cornell’s robot jellyfish and worm are powered by a hydraulic fluid battery LINK

Latest research and tools

GITHUB • 25 points View Source
Discussion: Reduce error handling boilerplate in Golang using '?': Discussion: a tool that helps reduce error handling boilerplate in Golang using the '?' syntax, streamlining code for more efficiency.

Want to get the latest news differently? Find us on:

See you tomorrow for a new dose of ☕️ Techpresso!