- The Current ⚡️
- Posts
- Another DeepSeek Moment? Say Hello to Manus, General AI Agent, Out Of China
Another DeepSeek Moment? Say Hello to Manus, General AI Agent, Out Of China
Also, Mistral’s new OCR release purports impressive benchmarks

⚡️ Headlines
🤖 AI
Microsoft's AI Chief Aims to Reduce Reliance on OpenAI - Microsoft is developing its own AI reasoning models, called MAI, to potentially replace OpenAI's technology in products like Microsoft 365 Copilot, aiming to lower costs and decrease dependency on OpenAI. [Reuters]
Google Co-Founder Larry Page Reportedly Launches New AI Startup - Larry Page is reportedly venturing into the AI industry with a new startup, marking his return to active involvement in tech innovation. [TechCrunch]
DuckDuckGo Enhances AI Features While Maintaining Privacy - DuckDuckGo has expanded its AI-assisted search answers beyond Wikipedia, providing users with private, useful, and optional AI features without compromising privacy. [DuckDuckGo]
DuckDuckGo Amplifies AI Search Tools with User Control - DuckDuckGo is enhancing its AI capabilities in its search engine and chatbot, allowing users to control the frequency of AI-generated answers or disable them entirely, emphasizing privacy and user choice. [The Verge]
Salesforce Introduces Agentforce 2dx for Autonomous AI in Enterprises - Salesforce has launched Agentforce 2dx, enabling autonomous AI agents to operate proactively across enterprise systems without constant human supervision, aiming to enhance productivity and efficiency. [VentureBeat]
Eric Schmidt Warns Against 'Manhattan Project' Approach to AGI - Former Google CEO Eric Schmidt cautions against an aggressive, large-scale initiative to develop artificial general intelligence, suggesting it could destabilize global relations and provoke international retaliation. [Business Insider]
Y Combinator Startups Embrace AI-Generated Codebases - Approximately 25% of startups in Y Combinator's current cohort have codebases predominantly generated by AI, indicating a significant shift towards AI-driven software development. [TechCrunch]
OpenAI and Perplexity AI Experience Surge in Search Traffic - OpenAI and Perplexity AI have reported substantial increases in search traffic, reflecting growing public interest and engagement with AI technologies. [Forbes]
🦾 Emerging Tech
President Trump Establishes Strategic Bitcoin Reserve - President Donald Trump has signed an executive order creating a Strategic Bitcoin Reserve, utilizing cryptocurrency assets forfeited in criminal cases to bolster national economic security. [Cointelegraph]
🤳 Social Media
Reddit Tests 'Rule Check' Feature to Prevent Rule Violations - Reddit is testing a "rule check" feature to help users comply with subreddit-specific rules, aiming to reduce post removals and enhance user engagement. [Reuters]
TikTok Plans Expansion into U.S. Local Services Market - TikTok is planning to expand its local services business into the U.S., indicating a strategic move to diversify its offerings and increase market presence. [Axios]
White House Opens Doors to Social Media Influencers - The Trump administration is incorporating social media influencers and podcasters into official roles, reflecting a new integration of media figures into government positions. [The Atlantic]
⚖ Legal
Anthropic Submits AI Policy Recommendations to the White House - AI safety and research company Anthropic has provided the White House with policy recommendations aimed at guiding the development and regulation of artificial intelligence technologies. [TechCrunch]
🔌 Plug-Into-This
A new Chinese-developed general AI agent, Manus, has demonstrated the capability to perform complex tasks, such as devising travel itineraries and providing investment advice. This development has drawn comparisons to previous AI breakthroughs like DeepSeek, signaling potential advancements in artificial intelligence.

Manus's capabilities include creating detailed travel plans and offering investment recommendations, showcasing its versatility.
The AI agent has attracted significant attention from investors, being hailed as a potential breakthrough in the AI field.
Comparisons to DeepSeek highlight Manus's potential impact on the AI industry, suggesting it could represent a significant advancement in general AI agents.
AI is getting out of hand 🤯
Manus, an AI agent from China, is automating approximately 50 tasks, creating a rather dystopian scenario
Reports suggest it is more accurate than DeepSeek, capable of simultaneously handling financial transactions, research, purchasing, etc
— Barsee 🐶 (@heyBarsee)
3:07 PM • Mar 7, 2025
🤖 Despite its rapid rise, there are concerns and doubts regarding Manus's abilities and the lack of public information about its development and functionalities.
Mistral AI has launched Mistral OCR, an Optical Character Recognition API designed to enhance document understanding by accurately processing complex elements like media, text, tables, and equations. This API accepts images and PDFs, extracting content in an ordered interleaved text and images format.

Mistral OCR is multilingual and multimodal, capable of parsing thousands of scripts, fonts, and languages across all continents.
Benchmark tests indicate that Mistral OCR outperforms other leading OCR models in accuracy across various aspects of document analysis.
The API is available on Mistral's developer suite, with plans to expand to cloud and inference partners, as well as on-premises deployments.
Mistral OCR is offered at a rate of 1,000 pages per dollar, with approximately double the pages per dollar available through batch inference.
PDF parsing is solved (again).
Mistral's new OCR API
— parses 1000-2000 pages for $1
— achieves state of the art results on tables, multilingual
— supports structure: images, bounding boxes, scans, equations90% of the world's organizational data is in PDFs.
— Deedy (@deedydas)
7:08 PM • Mar 6, 2025
🤔 While initially impressive, some independent testers have reported results inferior to the claimed benchmarks. Even so, these kind of unsexy AI solutions are where the real impact of this new tech can be seen in the most practical sense.
Artificial intelligence (AI) is being integrated into wildfire detection systems to identify fires more quickly and accurately than traditional methods. These AI-enhanced cameras can detect fires at their inception, often before they are reported by individuals, thereby improving response times and potentially reducing wildfire damage.

AI-powered cameras have successfully detected over 1,200 confirmed fires, outperforming human 911 callers approximately one-third of the time.
These systems are equipped with night vision capabilities, allowing continuous monitoring without breaks, enhancing their effectiveness in early fire detection.
Companies like Torch are developing outdoor sensors that utilize infrared and gas sensors to detect fires, aiming to provide rapid alerts and improve safety measures.
Data Blanket, a startup backed by Bill Gates, is employing AI-driven drones to map wildfire perimeters and deliver real-time information to firefighters, enhancing situational awareness and response strategies.
A great example of how University research saves lives. AI cameras developed and run out of UC San Diego spot wildfires before they spread.
— Caroline Freund (@CarolineFreund)
5:33 PM • Mar 2, 2025
🌍 Maybe AI isn’t just about using all the water and electricity after all! Regardless of overall balance on that equation, this kind of application is the sort of thing techno-optimists are ready to get excited for as AI pushes beyond the world of apps that aim to take money out of our pockets in exchange for doing new things on our smartphones.
🆕 Updates
Introducing our new family of state-of-the-art AI models: Phoenix-3, Raven-0, and Sparrow-0.
Together they bring Conversational Video Interfaces (CVI) to the next level, and power Charlie, our new demo persona 👋
— Tavus (@heytavus)
5:26 PM • Mar 6, 2025
Introducing #Ray2 Flash—3x faster, 3x cheaper new model. Flash brings Ray’s frontier production-ready Text-to-Video, Image-to-Video, audio, and control capabilities with high quality and speed to all subscribers—so you can create more, faster, and without limits. Available now.
— Luma AI (@LumaLabsAI)
5:02 PM • Mar 7, 2025
HunyuanVideo I2V is here! 🚀🚀🚀🚀🚀🚀
I2V training code is here! 🛩️🛩️🛩️🛩️🛩️🛩️
Always Open Source & Open Cccess!
Github: github.com/Tencent/Hunyua…
HuggingFace:
— Hunyuan (@TXhunyuan)
8:04 AM • Mar 6, 2025
📽️ Daily Demo
🗣️ Discourse
Earlier today Mistral released a new OCR api that claims to be state of the art. They didn’t release their benchmarks so we tested it against Gemini ourselves.
Mistral reported its OCR as better than Gemini. Our testing showed the opposite.
Full blog post in the replies!
— Adit 🔮 (@aditabrm)
3:38 AM • Mar 7, 2025
Fantastic 👏
Phoenix-3 from Tavus is the first and only Gaussian-diffusion model for real-time full-face rendering, generating complex microexpressions and emotions to match context.
They power an upgraded Conversational Video Interface (CVI) for human-like dialogue. This… x.com/i/web/status/1…
— Rohan Paul (@rohanpaul_ai)
2:29 PM • Mar 7, 2025
The bulk of this post is good + I applaud the folks who work on the substantive work it discusses.
But I'm pretty annoyed/concerned by the "AGI in many steps rather than one giant leap" section, which rewrites the history of GPT-2 in a concerning way.
— Miles Brundage (@Miles_Brundage)
11:17 PM • Mar 5, 2025
Today we're introducing Template Hub, the first step in offloading global work done by humans to agents.
— Convergence (@convergence_ai_)
2:00 PM • Mar 6, 2025