Stay Ahead on AI.

Good morning, robotics enthusiasts. Robotics major Figure is bringing its humanoids into the home—with plans to start testing *as soon as this year*.

Figure is charging toward a $39.5B valuation and evolving faster than ever. But are humanoids ready to leave controlled factory floors and step into the messy chaos of our homes?

In today’s robotics rundown:

Figure fast-tracking home humanoids
Beijing pouring billions on robot ambitions
The shape-shifting ‘GOAT’ robot
Harvard’s walking-jumping bug bot
Quick hits on other robotics news

LATEST DEVELOPMENTS

FIGURE

🤖 Figure fast-tracking home humanoids

Image source: Figure

The Rundown: Just days after launching the Helix AI system for humanoids, California-based robotics company Figure revealed plans to bring its robots into homes this year—two years ahead of the original schedule.

The details:

CEO Brett Adcock posted on X that he intends to start carrying out alpha tests using the latest iteration of its robot, Figure 02.
He said that the accelerated timeline is a direct result of integrating Helix, the company’s proprietary “generalist” Vision-Language-Action (VLA) model.
The model integrates visual data and language commands, allowing robots to understand speech and handle unfamiliar objects end-to-end.
These advances unlocked Figure’s second customer use case in just 30 days while also leading the company to end its OpenAI partnership.

Why it matters: Figure is pushing its humanoid robot business forward at breakneck speed, but bringing these machines from industrial settings into homes won’t be easy. While still in its early stages, Figure appears to be pulling ahead of rivals like Apptronik and Tesla, many of whom still rely on remote operation rather than true autonomy.

CHINA

🔥 Beijing pouring billions on robot ambitions

Image source: Ideogram/The Rundown

The Rundown: In response to the U.S. ramping up its AI and robotics efforts, Beijing committed up to CNY 100B ($13.7B) to fast-track key research and industrialization projects—strengthening its bid for leadership in the two sectors.

The details:

Beijing aims to achieve over 100 technological breakthroughs and develop at least 10 world-leading software and hardware products by 2027.
As part of this, the city will focus on deploying 10,000 robots with embodied intelligence, which is AI capable of learning and making decisions in real time.
These robots will be initially utilized in domains like scientific research, auto manufacturing, retail, home services, and elderly care.
Beijing also plans to “nurture 50 core enterprises along the industrial chain,” according to Chinese media outlet Yicai Global.

Why it matters: Beijing is home to around 2,400 AI-focused enterprises, with its core AI industry surpassing $41.2B in value last year. This initiative, along with the development of its AI city, E-Town, underscores the city's commitment to funding its ambitions of becoming a global leader in AI and robotics.

EPFL

⛰️ The shape-shifting GOAT robot

Image source: CREATE EPFL

The Rundown: Researchers from the École Polytechnique Fédérale de Lausanne (EPFL) School of Engineering developed Good Over All Terrains (GOAT), a robot that can change its shape on the fly to move faster over different terrains.

The details:

Inspired by how animals move in nature, GOAT can switch between a flat 'rover' and a sphere configuration to drive, roll, and even swim across different paths.
The robot's frame is made from low-cost materials—two crossed elastic fiberglass rods with four motorized rimless wheels.
Most interestingly, it uses only a satellite navigation system and an orientation-measuring device to determine the optimal path. No cameras/sensors onboard.
In field testing, GOAT navigated a 2.8-mile path through mountains, water, and urban areas, outperforming multimodal robots, the team said.

Why it matters: The GOAT robot’s bio-inspired design can navigate unpredictable terrain efficiently with limited environmental knowledge—and by adapting its shape, it can conserve energy. While more work needs to be done, the team envisions it could prove useful for environmental monitoring and disaster response.

HARVARD

🪳 Harvard’s walking-jumping bug bot

Image source: Harvard Microrobotics Laboratory

The Rundown: Inspired by the remarkable jumping abilities of springtail arthropods, Harvard University researchers developed a 1.7-inch micro-robot that can jump a whopping 23 times its own body length.

The details:

The bot has a robotic version of the springtail's furcula, a tail-like, spring-loaded appendage that is folded in beneath its body, held under tension.
When the furcula’s latch-mediated spring is released, the robot flings into the air, executing powerful jumps.
Based on the Harvard Ambulatory MicroRobot platform, the tiny bot achieved a maximum jumping distance of 1.4m, far more than any same-sized insect.
The bot can also walk swiftly, using its four ceramic actuator-powered legs to achieve speeds of up to 8.4 body lengths per second.

Why it matters: The robot's jumping mechanism activates in just 14 milliseconds, mimicking the speed of a natural springtail, and it can even manipulate objects and scuttle across water. What sets this micro-bot apart is its versatility, making it a good candidate for future search and rescue missions and even space exploration.

QUICK HITS

📰 Everything else in robotics today

NASA sent an ice-hunting drill to the Moon’s south pole to search for frozen water reservoirs—an essential step in its plan to send humans there in 2027.

The viral video of a Unitree humanoid nearly attacking a bystander at the Spring Festival Gala in China sparked fresh debate about robot safety.

China’s Wuhan Glory Road Intelligent Technology Co. recently unveiled its heavy-lifting Laborer-1 humanoid, designed to carry loads ranging from 88 to 132 lbs.

Chinese researchers developed a new self-powered biosensor that can detect E. coli bacteria in drinking water and autonomously destroy it.

UK robotics company ULC Technologies signed a deal to design and develop a new robotic system for solar panel installation, potentially revolutionizing the field.

Vancouver-based Nexera Robotics secured $4.5M to further improve its tech aimed at enabling robots to interact with diverse objects more reliably.

German researchers found that humanoids with “eyes” are more likely perceived by humans as having a mind and being capable of expressing emotions.

COMMUNITY

🎥 Join our next live workshop

We just did a full workshop on how to take full advantage of Claude 3.7 Sonnet’s frontier reasoning capabilities, along with advanced prompting strategies with Dr. Alvaro Cintas, The Rundown’s AI professor.

Watch it here. Not a member yet? Join The Rundown University on a 14-day free trial.

See you soon,

Rowan, Jennifer, and Joey—The Rundown’s editorial team

OpenAI's GPT-4.5 is here

Install today and build with Augment’s free plan.

Good morning, AI enthusiasts. OpenAI just dropped its largest model yet — but instead of crushing benchmarks, GPT-4.5 is doubling down on *feeling more human*.

With Sam Altman calling it "the first model that feels like talking to a thoughtful person," will users embrace this emotional evolution or scoff at the lack of raw performance gains they've come to expect?

In today’s AI rundown:

OpenAI’s GPT-4.5 with emotional intelligence
Tencent’s new ‘fast-thinking’ model
Building a smart calendar management system
Ideogram eyeing speed boost with new model
4 new AI tools & 4 job opportunities

LATEST DEVELOPMENTS

OPENAI

🔥 OpenAI’s GPT-4.5 with emotional intelligence

Image source: OpenAI

The Rundown: OpenAI just released GPT-4.5 (code-named Orion), the company’s largest model to date — which uses unsupervised learning instead of reasoning to achieve deeper world knowledge and improved emotional intelligence.

The details:

OpenAI says GPT 4.5 delivers a more natural conversational experience, with an improved understanding of human intent and greater emotional intelligence.
The model hallucinates less and delivers more accurate answers than previous versions, with testers liking it for pro tasks, creative work, and everyday queries.
It isn't a step up from previous models on math or science but does surpass o3-mini and o1 on SWE-Lancer, OpenAI’s new freelance coding task benchmark.
Only Pro users and developers on paid plans can access GPT-4.5 immediately, with Plus and Team users gaining access next week.
Notably, the API price of the model has been kept shockingly high at $75/$150 per million input/output tokens. For reference, GPT-4o costs just $2.50/$10.

Why it matters: While the benchmarks and pricing might leave some disappointed, 4.5 seems like more of a ‘vibe’ personality upgrade than a major step up. With high costs and fewer improvements than users have come to expect, this might also be the last stop both practically and acceleration-wise in non-reasoning model development.

TOGETHER WITH TURING

🎤 NeurIPS 2024, decoded

The Rundown: What happened at NeurIPS 2024? This recap breaks down the biggest AI advancements from leaders at Oracle, IBM, and Turing so you can stay ahead without the noise.

Gain insights on:

Multi-agent reasoning: How AI systems are learning to collaborate
Real-time adaptation: Smarter, more responsive models in action
Frontier scaling: Pushing the limits of training and optimization

Get the full recap here.

TENCENT

💨 Tencent’s new ‘fast-thinking’ model

Image source: Tencent

The Rundown: Chinese giant Tencent just released Hunyuan Turbo S, a new ‘fast-thinking’ AI designed for instant responses rather than deep reasoning — achieving 2x the speed while matching the performance of leading models on key benchmarks.

The details:

Turbo S matches models like DeepSeek V3, GPT-4o, and 3.5 Sonnet across knowledge, mathematics, and reasoning despite a focus on speed.
Tencent has significantly lowered the price of the new model, making it a fraction of the cost of the previous generation.
The company is also preparing to launch a complementary T1 reasoning model with "deep thinking," positioning the two models for different use cases.
The release comes amid increasing AI competition from China, with DeepSeek nearing a new launch and Alibaba debuting QwQ-Max for reasoning this week.

Why it matters: It wasn’t long ago that reasoning models were the new shiny toy, and now we have a ‘fast-thinking’ vs. ‘slow-thinking’ divide. With DeepSeek’s R1 shining a massive global spotlight on Chinese AI, rival labs are quickly rushing to one-up the industry darling — and U.S. chip restrictions don’t seem to be slowing anything down.

AI TRAINING

📅 Building a smart calendar management system

The Rundown: In this tutorial, you will learn how to create a personalized AI calendar assistant that can check your schedule and create events using n8n's workflow automation and AI capabilities.

Visit n8n and create a new workflow.
Add a chat trigger as your entry point.
Set up an AI Agent with calendar tools (Get Events and Create Event) using AI expressions, such as {{ $fromAI('placeholder_name') }}.
Add system instructions for the AI agent node to guide the AI's behavior.

Pro tip: Test your assistant with natural language prompts like "Find a one-hour slot tomorrow in my calendar and schedule a research meeting with Alex."

Check out our recent workshop showing how to create AI agents with n8n here.

PRESENTED BY AUGMENT CODE

🤖 Where pro engineers meet real AI

The Rundown: Augment Code isn’t your average AI assistant — it’s purpose-built for serious software engineers handling massive codebases. Backed by Eric Schmidt and trusted by top teams like Webflow, Lemonade, and DDN, Augment helps take productivity to the next level.

With Augment Code, you can:

Instantly integrate into VS Code, JetBrains, or Vim to stay in the zone
Leverage full-codebase context in under 200ms to keep your workflow fast
Handle tens of thousands of files or millions of lines effortlessly

IDEOGRAM

🖼️ Ideogram eyeing speed boost with new model

Image source: Ideogram / The Rundown

The Rundown: Ideogram launched its 2a model, a major update to the text-to-image platform that significantly reduces generation time and cost while maintaining high-quality outputs—with optimizations for graphic design and photorealistic generations.

The details:

2a generates image outputs in just 10 seconds, with an even faster ‘2a Turbo’ option delivering results at twice the speed.
The new model excels at graphic design and text generation, with the ability to create content like homepages, movie posters, and advertisements.
It is also optimized for photorealism and is priced at 50% less than Ideogram 2.0 for both API and web use.
Users can access it now via Ideogram’s web platform, API, or through applications like Freepik, Poe, and Gamma.

Why it matters: While many image-generation models still struggle with basic spelling, Ideogram’s latest release is taking graphic design and text generation to absolutely mindblowing new levels. With these new abilities, another complete step change in creative workflows is about to be unlocked.

QUICK HITS

🛠️ Trending AI Tools

💬 Scribe - ElevenLabs’ new SOTA speech-to-text model
🪨 Granite 3.2 - IBM’s compact open models for enterprise use
🗣️ Octave TTS - Generate AI voices with emotional delivery
🧑‍🔬 Deep Review - AI co-scientist for literature reviews

💼 AI Job Opportunities

🏗️ Grammarly - Software Engineer, Android
🔧 Scale AI - Field Engineering Tech Lead
📊 Dataiku - Senior Renewals Manager
🤖 Meta - AI Research Scientist (Robotics)

📰 Everything else in AI today

Pika Labs released its new 2.2 model, featuring upgraded quality, 10-second, 1080p resolution generations, and new transition and transformation capabilities.

Meta is reportedly planning to release a standalone Meta AI app in Q2 with potential paid subscription options similar to OpenAI's model.

Figure is pushing up its timeline to bring its humanoid robots into the home, beginning Alpha testing this year thanks to improvements from its recently revealed Helix AI.

Microsoft rolled out new updates to Copilot, including a dedicated MacOS app, the ability to upload PDF or text files, and UI improvements.

Meta introduced Aria Gen 2 glasses with advanced sensors, on-device AI processing, and an all-day battery for research in machine perception, contextual AI, and robotics.

You Labs unveiled ARI, a research agent capable of analyzing up to 400 sources and generating professional reports with charts, citations, and visuals in under 5 minutes.

COMMUNITY

🎥 Join our next live workshop

Join our workshop today at 4 PM EST to learn how to take full advantage of Claude 3.7 Sonnet’s frontier reasoning capabilities, along with advanced prompting strategies with Dr. Alvaro Cintas, The Rundown’s AI professor.

RSVP here. Not a member yet? Join The Rundown University on a 14-day free trial.

🤝 Share The Rundown, get rewards

We’ll always keep this newsletter 100% free. To support our work, consider sharing The Rundown with your friends, and we’ll send you more free goodies.

See you soon,

Rowan, Joey, Zach, Alvaro, and Jason—The Rundown’s editorial team

Tech

Instagram Reels may get a new home

Good morning, tech enthusiasts. As TikTok faces a potential ban in the U.S., tech companies are racing to fill the void with their own alternatives, including Meta’s Instagram—which may soon spin off Reels as a standalone app.

It’s no secret that Meta has long hoped to dethrone its Chinese rival, but the question remains: will TikTok’s vast Gen Z fanbase embrace what it has to offer?

In today’s tech rundown:

Instagram’s standalone Reels app
Waymo robotaxis racing ahead
Amazon’s first quantum chip
Biotech’s DeepSeek moment
Quick hits on other major news

LATEST DEVELOPMENTS

INSTAGRAM

👀 Instagram’s standalone Reels app

Image source: Ideogram/The Rundown

The Rundown: Meta’s Instagram is reportedly considering launching its Reels offering as a standalone short-form video app to take advantage of ByteDance-owned TikTok’s uncertain future in the U.S.

The details:

Instagram’s CEO Adam Mosseri was reportedly heard discussing the plan, according to an anonymous source cited by The Information.
Code-named “Project Ray,” the proposed app will likely support longer reels, going up to three minutes, and improved content recommendations.
The move comes after Meta launched a video-editing app called Edits, tailored for Instagram, to compete with TikTok-centric CapCut.
In 2018, Meta launched a standalone video app called Lasso to take on TikTok, but it was shut down two years later due to low user adoption.

Why it matters: If the move takes effect, Meta could strengthen Instagram’s user base and engagement by drawing a big chunk of TikTok’s 170 million U.S. users—should the platform go dark after April 5 due to a Trump ban. However, it won’t be an easy win, as rivals like X, YouTube, and Bluesky are also ramping up their vertical video game.

WAYMO

🚘 Waymo robotaxis racing ahead

Image source: Ideogram/The Rundown

The Rundown: Alphabet-owned Waymo announced that its fleet of self-driving taxis is now serving over 200,000 paid rides every week in the U.S.— marking a 20-fold jump from two years ago and pushing the company far ahead of its competitors.

The details:

Waymo’s driverless taxis operate in Los Angeles, San Francisco, and Phoenix, with plans to test in 10 new cities this year, including Austin, Atlanta, and Miami.
The company says its driverless taxis have captured 22% of the taxi market in San Francisco, surpassing Lyft and in tight competition with Uber.
Alphabet also invested an additional $5B in Waymo last year, supporting its rapid expansion and technological development plans.
As of October 2024, the company’s self-driving taxis were covering 1 million+ miles a week—a number that will only grow as it launches in new markets.

Why it matters: Waymo remains far ahead of the competition, especially with Cruise shutting down its robotaxi program and Zoox testing in just two cities. However, challenges do persist—Waymo's service is confined to geofenced areas, and regulatory and safety roadblocks can easily complicate expansion, putting operations at risk.

AMAZON

💻 Amazon’s first quantum chip

Image source: Amazon Web Services

The Rundown: Amazon has stepped into the quantum race with its first quantum computing chip, Ocelot, which it hopes will shave off as much as five years in the effort to build the world’s first viable quantum computer.

The details:

Developed in partnership with the California Institute of Technology, the Ocelot prototype integrates two small silicon microchips stacked atop each other.
It uses cat qubits, a type of qubit technology that suppresses certain quantum errors, which is essential for building fault-tolerant quantum computers.
Amazon says that the chip’s design could cut down resource costs associated with quantum error correction by as much as 90%.
Just recently, Microsoft unveiled its first quantum chip, Majorana. Google, meanwhile, announced its first quantum chip, Willow, in December.

Why it matters: One of the biggest challenges in quantum computing is qubits' sensitivity to vibrations, heat, and electromagnetic disturbances, leading to computational errors. Amazon’s Ocelot chip tackles this with its “cat qubit” design, which enhances stability. Its simpler architecture could also make scaling easier.

BIOTECH

🧪 Biotech’s DeepSeek moment

Image source: Ideogram/The Rundown

The Rundown: The biotech industry is having a DeepSeek moment as China celebrates a small biotech company’s drug for outperforming Merck’s cancer drug Keytruda—a $30B-a-year immunotherapy juggernaut and pharma’s bestselling drug.

The details:

The new drug from Akeso, Ivonescimab, outperformed Keytruda in treating lung cancer during Chinese trials last year.
Patients treated with the drug went 11.1 months before their tumors reappeared, compared to 5.8 months for Keytruda.
After the findings were revealed, shares in Akeso’s US partner, California-based Summit Therapeutics, more than doubled to a record high.
Over the past decade, Chinese firms have been innovating with advanced drugs, signing billions of dollars in licensing deals with Western companies.

Why it matters: China approved 37 new drugs in 2024, but concerns over trial data quality have led regulators to impose strict legal restrictions on its biotech firms. Now, with a global trial for Akeso’s new drug underway, the results could signal China’s rising strength in cutting-edge drug development.

QUICK HITS

📰 Everything else in tech today

Amazon unveiled Alexa+, its highly anticipated next-generation digital AI assistant, priced at $19.99 but free to 100M+ Amazon Prime members.

OpenAI is finally making its video generation model, Sora, available to users in the EU and UK via subscriptions to ChatGPT’s Plus and Pro tiers.

Meta is reportedly planning to release a standalone app for its AI assistant, Meta AI, as soon as this spring—in a move to take on ChatGPT.

Meta also revealed its second-gen experimental smart glasses—Aria Gen 2—intended to bolster research into AI, robotics, and machine perception.

Fintech company Stripe announced a tender offer for employees and shareholders that valued the company at $91.5 billion—41% higher than last year.

Nvidia plans to launch its next-gen AI computing chip, Blackwell Ultra, next month at the GTC conference.

Jeff Bezos’ rocket company Blue Origin is launching pop star Katy Perry and broadcast journalist Gayle King into space for a high-priced 10-minute joy ride.

Canadian biotech startup Afynia Laboratories raised $5 million in seed funding to commercialize a blood test for endometriosis.

COMMUNITY

🎥 Join our next live workshop

RSVP here. Not a member yet? Join The Rundown University on a 14-day free trial.

See you soon,

Rowan, Jennifer, and Joey—The Rundown’s editorial team

Amazon's new AI-powered Alexa

Start integrating Radar today.

Good morning, AI enthusiasts. Alexa’s long-awaited AI revamp is here — and it may be Amazon’s biggest AI move yet.

With a massive intelligence upgrade and new agentic power set to land in the hands of over 100M Prime members, is this the ‘ChatGPT moment’ for voice assistants?

In today’s AI rundown:

Amazon’s gen AI-powered Alexa+
ElevenLabs’s new speech-to-text AI
Personalize your AI coding assistants
Inception Labs’ ultra-fast diffusion model
4 new AI tools & 4 job opportunities

LATEST DEVELOPMENTS

AMAZON

🔊 Amazon’s gen AI-powered Alexa+

Image source: Amazon

The Rundown: Amazon just unveiled Alexa+, its highly-anticipated next-generation digital assistant completely rebuilt with AI — promising more conversational interactions, personalization, and agentic capabilities for everyday tasks.

The details:

Alexa+ can connect and leverage multiple LLMs, including Amazon's Nova and Anthropic's Claude, choosing the best model for each task at hand.
The revamped assistant can perform complex agentic tasks like booking reservations, ordering groceries, purchasing concert tickets, and more.
Other features include document analysis, remembering user preferences, maintaining conversation context, and integration with hundreds of services.
It will cost $19.99 monthly but comes free with Amazon Prime membership, with early access rolling out in the U.S. next month.

Why it matters: Legacy voice assistants like Alexa and Siri have lagged massively behind the AI boom, but this release will finally put advanced voice agents in the homes of 100M+ Prime members — potentially triggering another ‘ChatGPT moment’ for consumers outside the tech bubble (assuming it goes better than Apple Intelligence).

TOGETHER WITH WORKOS

🔐 Protect your AI apps from bad actors

The Rundown: WorkOS Radar is a security solution that shields your AI platform from fake signups, throwaway emails, and brute force attempts — all powered by advanced device fingerprinting and real-time detection.

With WorkOS Radar, you can:

Rapidly detect and challenge unfamiliar and suspicious devices in real time
Stop free-tier abuse and fraudulent behavior with advanced detection
Customize threat responses to fit your app’s exact security needs

ELEVENLABS

🗣️ ElevenLabs’s new speech-to-text AI

Image source: ElevenLabs

The Rundown: ElevenLabs released Scribe, a new speech-to-text model that claims to be the most accurate in the world, outperforming industry leaders like Google's Gemini 2.0 Flash and OpenAI's Whisper v3 across dozens of languages.

The details:

Scribe supports 99 languages, with claimed accuracy rates exceeding 95% for over 25 languages, including English, Italian, and Spanish.
The model raises the bar in a variety of languages that traditionally lack speech recognition and transcription options, like Serbian, Cantonese, and Malayalam.
Its other features include multi-speaker labeling, word-level timestamps, and the ability to detect non-verbal audio markers like laughter or music.
Scribe is priced at $0.40 per hour of transcribed audio for pre-recorded audio, with a low-latency version for real-time applications coming soon.

Why it matters: With Scribe’s accuracy and focus on the unpredictability of real-world audio, people can expect flawless subtitles, searchable podcast archives, and more. It also opens up high-level transcriptions to a more global audience — particularly for low-resource languages that have previously been neglected by other models.

AI TRAINING

🤖 Personalize your AI coding assistants

The Rundown: In this tutorial, you’ll learn how to add custom instructions to Cursor and Windsurf AI coding tools to personalize and enhance your AI coding development workflow.

Here’s your step-by-step guide:

Download Cursor or Windsurf on your device from their official websites.
To set up Windsurf rules, create a .windsurfrules file in your project root or choose between global rules or project rules via “Edit Rules” in Settings.
Similarly, to configure rules in Cursor, head over to its Settings and set up global AI rules or project-specific rules.

Pro tip: Global rules set universal preferences, while project rules define instructions for specific project requirements. You can also browse this GitHub repo for ready-to-use templates of Cursor rules.

PRESENTED BY INNOVATING WITH AI

💼 Start your career as an AI Consultant

The Rundown: Innovating with AI’s new program, AI Consultancy Project, equips AI enthusiasts with all the resources they need to capitalize on the booming AI consulting market — set to grow 8x to $54.7B by 2032.

The program offers:

Tools and framework to find clients and deliver top-notch services
A 6-month roadmap to build a 6-figure AI consulting business
Student landing their first AI client in as little as 3 days

Request early access to The AI Consultancy Project.

INCEPTION LABS

⚡ Inception Labs’ ultra-fast diffusion model

Image source: Inception Labs / Artificial Analysis

The Rundown: Inception Labs just emerged from stealth with Mercury, a new ‘diffusion’ LLM that generates text up to 10x faster than traditional LLMs while still matching their quality — with speeds over 1000 tokens/sec on standard H100 chips.

The details:

LLMs generate text one token at a time, but Mercury’s diffusion approach generates entire blocks in parallel for increased speed, efficiency, and control.
Their first model, Mercury Coder, matches or beats the coding performance of models like GPT-4o Mini and Claude 3.5 Haiku at 5-10x the speed.
Inception was founded by Stanford professor Stefano Ermon, who researched how to apply diffusion (commonly used for image and video generation) to text.
Mercury models can serve as drop-in replacements for traditional models in areas like code generation, customer support, and enterprise automation.

Why it matters: By bringing "Sora-like" diffusion to text, Inception is going against the grain on fundamental assumptions about how AI should generate language. Its technique could potentially enable more powerful agents, better and more efficient reasoning, and AI experiences that feel truly instantaneous.

QUICK HITS

🛠️ Trending AI Tools

🎥 Wan 2.1 - Alibaba’s new open-source SOTA AI video suite
🧠 Advanced Voice - ChatGPT’s conversational voice feature for free users
⚙️ Gemini Code Assist - Free AI coder with 180K code completions per month
📹 Project Starlight by Topaz Labs - Bring old videos back to life with AI

💼 AI Job Opportunities

💼 UiPath - Enterprise Sales Executive, Public Sector
📈 Cresta - Analytics Engineer
🌍 Luma AI - Community Marketer
🔒 Curai - Privacy Lead

📰 Everything else in AI today

Hume AI released Octave, a text-to-speech LLM that understands emotional context, allowing creators to design custom voices with control over emotion and delivery.

Perplexity introduced a redesigned voice mode in its latest iOS update, featuring six different voice options, direct search result navigation, and more.

Poe launched Poe Apps, enabling users to create apps and visual UI interfaces using a combination of reasoning, multimodal, image, video, and audio models on the platform.

Vevo Therapeutics launched the Arc Virtual Cell Atlas featuring Tahoe-100M, an open-source dataset mapping 60,000 drug-cell interactions across 100M cells.

Exa launched Websets, a search product that deploys agents for better results, beating Google by over 20x and OpenAI Deep Research by 10x on complex queries.

IBM unveiled its new Granite 3.2 model family, featuring compact reasoning, vision-language, and specialized time series models for enterprise use.

Microsoft launched Phi-4 multimodal and Phi-4 mini SLMs, matching or exceeding the performance of models twice their size on certain tasks.

COMMUNITY

🎥 Join our next live workshop

Join our next workshop today, Thursday, February 27th, at 1 PM EST and learn how to transform your creative concepts into professional videos with LTX Studio's Tom Factor, Director of Creative Strategy and AI, and Alon Yaar, VP of Product.

RSVP here. Not a member yet? Join The Rundown University on a 14-day free trial.

🤝 Share The Rundown, get rewards

We’ll always keep this newsletter 100% free. To support our work, consider sharing The Rundown with your friends, and we’ll send you more free goodies.

See you soon,

Rowan, Joey, Zach, Alvaro, and Jason—The Rundown’s editorial team

Robotics

Self-replicating robot revolution begins

Good morning, robotics enthusiasts. Texas-based Apptronik just signed a landmark new deal to put its Apollo humanoids to work building more of themselves.

While most robotics startups struggle with manufacturing bottlenecks, this partnership could scale production faster than anyone predicted. Are you ready to witness the birth of a self-replicating robotic workforce?

In today’s robotics rundown:

Apptronik humanoid building itself
Humanoids learning to get themselves up
A drone made with moth antennae
MIT’s thumb-sized rover headed to moon
Quick hits on other major news

LATEST DEVELOPMENTS

APPTRONIK

🤖 Apptronik humanoid building itself

Image source: Apptronik

The Rundown: Texas-based robotics company Apptronik announced a pilot deal with U.S. manufacturing giant Jabil, which could eventually see its Apollo humanoid working on the factory floor to assemble more units of itself.

The details:

Jabil will build Apollo robots and begin to integrate them into specific manufacturing operations, including production lines building Apollo.
Once integrated, Apollo will perform various tasks, including inspection, sorting, and transporting parts—before being upgraded to more complex tasks.
Apptronik also has pilot agreements with Mercedes-Benz and GXO Logistics and has teamed up with NASA to advance its use of humanoids.
The “real-world validation” deal comes mere weeks after Apptronik’s $350M fund-raise aimed at scaling up Apollo production.

Why it matters: With this deal, Apptronik aims to create a “flywheel” effect—generating real-world training data for robot AI models. This will not only improve their performance in factory settings but also make humanoids smarter and faster at building more of themselves, removing long-standing production bottlenecks.

HUMANUP

🦿Humanoids learning to get themselves up

Image source: University of Illinois Urbana-Champaign

The Rundown: Unitree’s G1 humanoid boasts some impressive feats, even kung fu. Now, a group of U.S. and Canadian researchers is using it to test a new machine learning tech that allows humanoids to get up on their own after falling.

The details:

The team has designed a framework—HumanUP—that relies on reinforcement learning rather than programmed responses to ensure autonomous recovery.
Essentially, it’s a two-stage RL approach that identifies limb trajectories for getting up and refines those motions into smooth real-world movements.
During the testing on G1, HumanUP achieved a 78.3% success rate in getting up and 98.3% in rolling over different terrains, including concrete and mud.
The framework's efficient movements may also reduce the wear and tear of robots’ motors and equipment, potentially extending their lifespan.

Why it matters: While other humanoids like Sigmaban and Valkyrie can rise after falls using deep RL, HUMANUP’s success rate is a notable distinction, at least compared to traditional controllers. The tech is new but could potentially transform how robots operate in unpredictable environments such as disaster sites.

DRONE RESEARCH

🪰 A drone made with moth antennae

Image source: Ideogram/The Rundown

The Rundown: Japanese researchers developed a novel bio-hybrid drone with live silkworm moth antennae, allowing it to navigate by detecting odors — and paving the way for a new generation of search-and-rescue drones.

The details:

Researchers from Shinshu University and Chiba University created the bio-hybrid drone to detect and track airborne chemical signatures.
They extracted the antennae from male silkworms and connected them to an EAG sensor measuring the antennae’s electrical responses to odor stimuli.
When mounted onto a drone, the system sensed odor sources from up to 5 meters away, an improvement from the 2-meter range of previous models.
The work integrates the precision and agility of robots with biosensory mechanisms but is limited by issues like the limited lifespan of the antennae.

Why it matters: If and when scaled, this tech can overcome the limitations of visual sensors, allowing drones to work in conditions hit by moisture, smoke, or low light and helping with disaster response, detecting gas leaks, explosives, or early fires. However, the use of biological parts could raise ethical concerns.

MIT ROBOTICS

🌗 MIT’s thumb-sized rover headed to moon

Image source: MIT

The Rundown: MIT scientists just sent a thumb-sized rover on board a SpaceX Falcon 9 to the moon’s south pole—where it will wheel around the roof of the main mission rover, taking temperature readings and performing tasks too dangerous for humans.

The details:

The mini-rover, AstroAnt, is a tiny swarm bot designed for inspections and diagnostic tasks on external surfaces of spacecraft, rovers, and landers.
Weighing 28 grams, the rover will use magnetic wheels to stick to the main rover’s surface and take temperature readings to monitor its operations.
Its space voyage is part of Intuitive Machines’ IM-2 mission and marks the first time MIT has sent active technology to the Moon since the Apollo era.
The modular design of the rover also allows it to perform specialized tasks in high-radiation or dangerous areas, reducing risks to human astronauts.

Why it matters: MIT’s tiny bot will play a key role in collecting data for NASA as it prepares to send humans back to the Moon as part of its Artemis III mission in 2027. The information about the frozen terrain of the lunar south pole (among other things) will help astronauts build settlements and launch missions to Mars and beyond.

QUICK HITS

📰 Everything else in robotics today

Polish startup Nomagic raised $44M in funding for its robotic arm designed for picking, packing, and moving in logistics operations.

Sanctuary AI just announced the integration of new tactile sensors into its Phoenix humanoid, enhancing its dexterity to perform complex, touch-driven tasks.

Amazon expects to spend $100B on AI initiatives in 2025, with roughly a quarter earmarked for robotic automation in its warehouses, according to the Financial Times.

Researchers at Columbia Engineering created an autonomous robot arm that learns its physical form through self-observation in a mirror.

NASA astronauts are testing Astrobee, a robotic system with tentacle-like grippers, on the International Space Station to safely remove orbital debris and protect satellites.

Japan’s Rakuten Group expanded its delivery service in the country with the inclusion of autonomous robots from Avride.

New consumer data shows that 2.1 million floor-cleaning robots were sold globally in 2023, accounting for 57% of domestic service robots.

U.S. researchers are using origami as inspiration to build a lightweight, flexible robotic arm to offer wheelchair users a way to grasp objects otherwise out of reach.

COMMUNITY

🎥 Join our next live workshop

Join our next workshop today, Thursday, February 27th at 1 PM EST and learn how to transform your creative concepts into professional videos with LTX Studio's Tom Factor, Director of Creative Strategy and AI, and Alon Yaar, VP Product.

RSVP here. Not a member yet? Join The Rundown University on a 14-day free trial.

See you soon,

Rowan, Jennifer, and Joey—The Rundown’s editorial team

Alibaba's AI video powerhouse

Click here to download the report.

Good morning, AI enthusiasts. China's Alibaba just released a powerful AI video generation suite that ANYONE can download, run, and modify — and early tests suggest it outperforms even OpenAI's Sora.

While most AI labs keep its best AI locked behind waitlists and paywalls, Chinese labs are starting to push ahead with open-source alternatives that anyone can use today.

Speaking of AI video, our next workshop is today at 3 PM EST — where you’ll learn how to create personalized AI video messages at scale with Synthesia's Kevin Alster. RSVP.

In today’s AI rundown:

Alibaba’s advanced AI video suite
Google’s free AI coding assistant
Find your next SaaS idea using AI research
Claude plays Pokemon Red live on Twitch
4 new AI tools & 4 job opportunities

LATEST DEVELOPMENTS

ALIBABA

🎬 Alibaba’s advanced AI video suite

Image source: Alibaba

The Rundown: Alibaba's Tongyi Lab just released Wan2.1, an open-source suite of powerful video generation models that outperform SOTA open-source and closed models such as Sora on key benchmarks — while generating videos at 2.5x the speed.

The details:

Wan2.1-T2V-14B tops the VBench leaderboard, excelling in areas like complex motion dynamics, real-world physics simulation, and text generation.
All models support text-to-video, image-to-video, and video-to-audio, and are the first with the ability to render text in both English and Chinese.
Wan’s editing tools include video inpainting and outpainting, multi-image referencing, and the ability to maintain existing structures and characters.
The release also includes a light 1.3B version capable of running on consumer hardware—it can generate a 5-sec 480P clip on RTX 4090 in 4 minutes.

Why it matters: Another day, another wild open-source release out of China. Wan is a continuation of the accelerating quality we’ve seen from recent launches like Google’s Veo 2 — with telltale AI signs (choppy motion, artifacts, etc.) all but completely eliminated. Between Qwen and Wan, Alibaba is bringing the open-source heat in 2025.

TOGETHER WITH GOOGLE CLOUD

🔮 AI predictions every startup needs to know

The Rundown: Google Cloud’s Future of AI: ‘Perspectives for Startups’ report gathers predictions, real-world applications, and expert strategies from 23 AI leaders like Amin Vahdat (Google Cloud) and Chamath Palihapitiya (Social Capital) to help you position your business for success in 2025 and beyond.

What’s inside:

Expert-backed predictions on AI’s role in business and tech
Lessons from real-world AI adoption at startups
Practical strategies to leverage AI for a competitive advantage
Areas investors prioritize for AI startups

GOOGLE

🔧 Google’s free AI coding assistant

Image source: Google

The Rundown: Google just launched a free version of Gemini Code Assist for individual developers, offering access to advanced AI-powered coding help with usage limits that dwarf competitors like GitHub Copilot.

The details:

Gemini Code Assist is powered by a fine-tuned version of Google's Gemini 2.0 model optimized specifically for programming tasks.
The new tool provides up to 180,000 monthly code completions — 90 times more than GitHub Copilot's free tier limit of 2,000.
The assistant features a 128,000 token context window, allowing it to process and understand much larger codebases than competitors.
The free version also integrates with dev environments like Visual Studio Code, GitHub, and JetBrains, with just a personal Google account needed.

Why it matters: AI has changed programming forever, with powerful free tools driving the biggest shift. Google's latest push with Gemini Code Assist could further disrupt this market dominated by GitHub Copilot—unlocking new possibilities for developers worldwide.

AI TRAINING

🧠 Find your next SaaS idea using AI research

The Rundown: In this tutorial, you’ll learn how to use Perplexity’s Deep Research to build a validated software product, complete with development specifications and launch strategy.

Step-by-step:

Visit Perplexity and select "Deep Research".
Use the prompt: "You are a Market Research Specialist. Analyze [INDUSTRY] by identifying top 10 problems, severity levels, current solutions, and market size."
Transform your chosen problem into a product by prompting: "Design a SaaS solution for [problem], including essential features, tech stack, timeline, and revenue streams."
Get implementation details with: "Provide step-by-step coding instructions, including API specifications and deployment configurations."

Pro tip: If you don’t know how to code, AI coding assistants like Cursor, Windsurf, or Replit Agent can massively help you build your new idea.

PRESENTED BY UNSTRUCTURED

🙅 Ditch your old data connectors

The Rundown: Join Unstructured for a deep dive into best practices for data connectors for generative AI applications, and see where solutions like Fivetran, Boomi, and AirByte fall short with maintaining rich context or keeping your team out of the weeds updating them.

In this live session, you’ll discover:

Frameworks for evaluating ETL tools across critical features
Why traditional solutions struggle with messy data sources
Best practices to keep engineers focused on innovation — not maintenance

AI RESEARCH

🕹️ Claude plays Pokemon Red live on Twitch

Image source: Anthropic

The Rundown: Anthropic just debuted "Claude Plays Pokémon" on Twitch, a continuation of the company’s research showcasing their new AI model Claude 3.7 Sonnet attempting to navigate the classic Game Boy game Pokémon Red in real-time.

The details:

3.7 Sonnet made major progress compared to its predecessors, defeating three gym leaders — with the original Sonnet struggling to leave the starting location.
The livestream shows Claude's "thought process" on the left while real-time gameplay appears on the right, giving viewers insight into the AI's reasoning.
Claude has access to a knowledge base to store info, function calling to take actions, and vision capabilities to observe the game.
Unlike previous versions, 3.7 Sonnet’s reasoning capabilities help navigate the game more effectively — planning, adapting, and remembering objectives.

Why it matters: Watching Claude tackle Pokemon is both fascinating and endearing — especially when seeing the thought process behind every move. While we’ll be playing games alongside AI agents in the future, Claude’s experiment is also an interesting window into a future where AIs are observed passively as popular gaming streamers.

QUICK HITS

🛠️ Trending AI Tools

🤖 Claude 3.7 Sonnet and Claude Code - New reasoning AI and coding agent
🧠 QwQ-Max - Alibaba’s open-source reasoning LLM
🔊 Luma Labs Video to Audio - Generate synced audio for video generations
🔥 EasyGen - A LinkedIn post generator that creates algorithm-optimized posts

💼 AI Job Opportunities

🤖 Figure AI - FPGA Design Intern
🖥️ Anyscale - Software Engineer, ML Developer Platform
💰 Coreweave - Credit and Collections Manager
📊 UiPath - GTM Operations Planning Program Manager

📰 Everything else in AI today

OpenAI rolled out its Deep Research feature to ChatGPT Plus, Team, Edu, and Enterprise tiers, offering 10 queries per month compared to the $200 Pro tier’s 120.

Anthropic is reportedly set to raise a larger-than-planned $3.5B funding round at a $61.5B valuation, with the news coming just days after the release of Sonnet 3.7.

DeepSeek is reportedly aiming to push the release of its next AI model up from its initial May timeline, seeking to capitalize on its success in the wake of R1.

OpenAI also launched a GPT-4o mini-powered Advanced Voice to free ChatGPT users—promising the same conversation style as the GPT-4o version for Plus and Pro users.

Microsoft removed usage limits on Copilot's Voice and Think Deeper features, giving all free users unlimited access. Pro users will retain priority access during peak hours.

Over 1,000 musicians released a silent album to protest the UK’s proposed copyright changes that would allow AI companies to train on works without explicit permission.

IBM announced plans to acquire DataStax, aiming to help unlock enterprise data for AI applications while expanding its NoSQL database offerings.

COMMUNITY

🎥 Join our next live workshop

Join our next workshop today, Wednesday, February 26th at 3 PM EST and learn how to create personalized AI video messages at scale with Synthesia's Kevin Alster, Strategic Advisor and former Head of Synthesia Academy.

RSVP here. Not a member yet? Join The Rundown University on a 14-day free trial.

See you soon,

Rowan, Joey, Zach, Alvaro, and Jason—The Rundown’s editorial team

Claude enters the reasoning era

Click here to request early access to The AI Consultancy Project.

Good morning, AI enthusiasts. Reasoning models are the new battleground in AI—and Anthropic has officially entered the race with Claude 3.7 Sonnet, the world’s first-ever hybrid reasoning model.

After months of quiet, Anthropic is making a big splash on the AI leaderboards. But with DeepSeek’s aggressive open-source push and a unified GPT-5 on the horizon, how long will this dominance last?

In today’s AI rundown:

Claude 3.7 Sonnet with 'hybrid reasoning'
Qwen’s new open-source thinking model
How to talk to your apps with Concierge AI
AI agents get their own communication protocol
4 new AI tools & 4 job opportunities

LATEST DEVELOPMENTS

ANTHROPIC

🧠 Claude 3.7 Sonnet with 'hybrid reasoning'

Image source: Anthropic

The Rundown: Anthropic just released Claude 3.7 Sonnet, the world's first ‘hybrid reasoning’ AI that can combine instant responses with controllable extended thinking capabilities — alongside a new agentic coding tool called Claude Code.

The details:

Claude 3.7 Sonnet enables users to toggle between a standard and "extended thinking" mode, with the latter showing the AI’s reasoning via a scratchpad.
API users can precisely control how long Claude thinks (up to 128K tokens), allowing them to balance speed, cost, and quality based on task complexity.
The AI achieves SOTA performance on real-world coding benchmarks and agentic tool use, surpassing competitors like o1, o3-mini, and DeepSeek R1.
Anthropic also introduced Claude Code, a command-line coding agent that can edit files, read code, and write and run tests, in a limited research preview.

Why it matters: Anthropic has finally brought Claude into the reasoning era — with vastly improved coding benchmarks, precise thinking control, and a new agentic feature that points to a major push on the dev side. With OpenAI also planning a hybrid reasoner, 3.7 may be the start of the next leg up for a new generation of models.

TOGETHER WITH IMAGINE AI LIVE

🌟 Your AI future starts here

The Rundown: IMAGINE AI LIVE '25 is a transformative three-day conference at the Fontainebleau Las Vegas where business leaders can connect with AI innovators and learn practical strategies for enterprise AI implementation.

Attend on May 28-30 and experience:

Enterprise AI strategies from world-class practitioners and thought leaders
Hands-on workshops covering practical applications of AI in business operations
Networking opportunities with hundreds of executives actively transforming their organizations with AI

Apply by Friday, February 28 with code IMAGINEAILIVE25 and save 25% on VIP and General Passes!

ALIBABA

🤔 Qwen’s new open-source thinking model

Image source: Alibaba

The Rundown: Alibaba's Qwen team just released QwQ-Max-Preview, a new reasoning-focused AI that introduces thinking capabilities to their chat platform — while promising a full open-source release soon.

The details:

QwQ-Max-Preview is built on Qwen2.5-Max but significantly enhanced for deep reasoning, excelling in mathematics, coding, and agentic tasks.
The model introduces a "Thinking (QwQ)" feature to Qwen Chat that allows users to see the AI's reasoning process as it works through complex problems.
Qwen announced plans to open-source QwQ-Max and Qwen2.5-Max under an Apache 2.0 license soon, making the models freely available for developers.
The team will also release smaller variants like QwQ-32B for local deployment on devices with limited compute resources.

Why it matters: Reasoning has become the new competitive frontier in AI, and Qwen’s move to open-source their flagship reasoner could push the industry toward having these capabilities as a standard rather than a gated, premium feature. Open source is staying right on the heels of industry leaders — with Chinese labs leading the way.

AI TRAINING

⚡ How to talk to your apps with Concierge AI

The Rundown: Concierge is the first connected AI Assistant that can read and write to your favorite software tools, by finding and updating information across all your apps in real-time.

Step-by-step:

Create a Concierge account here to get started for free.
Click "Connect" and choose the tools you use — Gmail, Slack, Jira, Notion, HubSpot, or something else.
Type requests like "summarize my unread emails", “create a Jira ticket based on feedback in Slack" or “research these prospects from Salesforce.”
Watch as your AI assistant instantly finds information and takes action across your connected apps.

Pro tip: Start with one or two apps you use most frequently, then gradually add more as you get comfortable with the workflow.

PRESENTED BY INNOVATING WITH AI

🤝 Turn AI passion into a consulting career

The Rundown: Innovating with AI's new program, AI Consultancy Project, transforms AI enthusiasts into professional consultants — tapping into a market projected to reach $54.7B by 2032.

The 6-month program delivers:

Proven frameworks for client acquisition and service delivery
A step-by-step path to six-figure consulting income
Students who land their first AI client in as little as 3 days

AI & COMMUNICATION

📞 AI agents get their own communication protocol

Image source: Anton Pidkuiko on YouTube

The Rundown: Two developers just introduced Gibber Link, a sound-based communication protocol that allows AI agents to detect each other on calls and switch from human speech to direct data transmission — reducing time and compute costs.

The details:

Created by Anton Pidkuiko and Boris Starkov at ElevenLabs’ recent Hackathon, the project uses an open-source data-over-sound library called “ggwave.”
In the demo, an agent detects another AI on the phone and switches to dial-up-style ggwave audio signals with transcriptions, instead of normal voice.
Using the sound-level protocol instead of generating speech reduces compute costs by up to 90% and shortens communication time by as much as 80%.
The design also ensures clearer communication in noisy environments compared to traditional speech recognition-based systems.

Why it matters: AI voice agents are about to be everywhere, meaning the volume of AI-to-AI calls will grow exponentially (especially for businesses). This hackathon-winning project is a great look at how finding more efficient and cost-effective methods for these interactions might take AI communication down completely new paths.

QUICK HITS

🛠️ Trending AI Tools

🤖 Zapier Agents - Equip agents with internal data to work across 7,000+ apps
👀 SmolVLM2 - Small AI models to analyze videos on phones and laptops
🗣️ ElevenLabs Studio - Structure, edit, and generate long-form audio
🧬 Evo 2 - Biomolecular AI model for insights into DNA, RNA, and proteins

💼 AI Job Opportunities

💊 Tempus - Key Account Director, Pharma East
🔧 DeepL - Customer Technical Solutions Manager
🏛️ Scale AI - Communications Manager, Public Sector
📢 Cresta - Senior Customer Marketing Manager

📰 Everything else in AI today

Luma Labs released a new Video to Audio feature in Dream Machine, enabling users to easily generate synced audio for video outputs.

Perplexity posted a teaser for a new agentic search browser called Comet, with a waitlist available to sign up for early access.

Salesforce and Google expanded their partnership to (among other things) integrate Gemini into Agentforce, allowing AI agents to process images, audio, and videos.

Alibaba revealed plans to invest $52B in cloud computing and AI infrastructure over the next three years, surpassing total spending in these sectors over the past decade.

Nothing unveiled its upcoming AI-powered Nothing Phone (3a), featuring an unboxing video by 1X’s recently debuted NEO Gamma humanoid robot.

Meta AI launched in Arabic across 10 Middle Eastern and North African countries, taking text generation, image creation, and animation to millions of new users.

COMMUNITY

🎥 Join our next live workshop

Join our next workshop on Thursday, February 27, at 1 PM EST and learn how to transform your creative concepts into professional videos with LTX Studio's Tom Factor, Director of Creative Strategy and AI, and Alon Yaar, VP of Product.

RSVP here. Not a member yet? Join The Rundown University on a 14-day free trial.

🤝 Share The Rundown, get rewards

We’ll always keep this newsletter 100% free. To support our work, consider sharing The Rundown with your friends, and we’ll send you more free goodies.

See you soon,

Rowan, Joey, Zach, Alvaro, and Jason—The Rundown’s editorial team

Tech

Apple’s $500B power move