New Digit 5 Humanoid Robot AI Upgrades, Gemini 3.8 Live, Claude Docs + Slides

Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.

Original language: English

Original video: New Digit 5 Humanoid Robot AI Upgrades, Gemini 3.8 Live, Claude Docs + Slides · Channel: AI News · 07:50 · Watch on YouTube

Topics: Task and workflow automation, Multimodal models, Robotics and autonomous systems

Summary

Agility Robotics unveiled Digit 5, highlighting fleet integration, long uptime, real-world training, industrial lifting specifications, adaptable tools, and cooperative safety, while noting that the robot is still in development. The video also covers UBTE’s new humanoid robot factory in China, which it says can produce one robot every 10 minutes. Google introduced Gemini 3.8 Live models for conversational voice AI, while Anthropic added Docs and Slides to Claude. Meta announced updates to Muse Code and a vision-first bionic arm that can select grips based on what it sees.

Key points

Digit 5 is designed to integrate with existing warehouse systems through fleet management software.
Agility says Digit 5 has a 10:1 run-to-charge ratio, allowing up to 20 hours of work per 24-hour shift.
Agility says the system was trained on 65,000 hours of real operational data; the robot is rated for a 23 kg payload and 2.2 m reach.
Digit 5 uses swappable grippers and a physical AI stack to expand the workflows it can perform, with safety features intended for working near people.
Digit 5 remains in development, and its listed specifications are preliminary estimates.
UBTE says its Chinese factory has an annual capacity of 10,000 humanoid robots and produces one complete robot every 10 minutes.
Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking support near-real-time visual input, multilingual conversation, and background tool calls.
Claude Docs and Slides are in beta, allowing users to create, edit, export, and share documents and presentations from chat.
Meta’s Muse Code exited beta with new developer features, while Smart Arm introduced a vision-first bionic arm that chooses grips automatically.

Chapters

Start	Chapter	Summary
00:00	Digit 5 humanoid robot upgrades	The video counts down Digit 5’s planned features and specifications, emphasizing cooperative safety while noting that the robot remains in development.
02:37	UBTE’s humanoid robot factory	UBTE describes a Chinese smart factory producing humanoid robots at scale, with digital scheduling and inspection processes.
03:37	Gemini 3.8 Live voice AI	A demonstration of visual voice assistance leads into Google’s two Gemini 3.8 Live models and their conversational, multilingual, and reasoning capabilities.
06:17	Claude productivity tools and Meta announcements	Anthropic’s Claude Docs and Slides launch in beta, followed by updates to Meta’s Muse Code and a vision-first bionic arm.

Quotes

Agility says Digit 5 hits a 10:1 runto charge ratio — 00:31

Digit 5 is still in development. — 02:06

one complete robot rolling off the line every 10 minutes. — 02:37

Google just released two new voice AI models, Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking — 04:43

Claude Docs and Claude slides are now in beta — 06:17

Concepts and entities

Digit 5, Agility Robotics, UBTECH, Gemini 3.8 Live, Gemini 3.8 Live Extended Thinking, Claude Docs, Claude Slides, Muse Code, DINOv2, Smart Arm

Full transcript (original language)

00:00 How many humanoids are too many to control? Today on AI News, we’re counting down several features coming to Agility’s new Digit 5 robot. Plus, the first 10,000 unitoid smart factory and a new wave of AI tools from Google, Anthropic, and Meta. But first, Agility Robotics just unveiled its newest Digit 5 humanoid packed with six new upgrades. And we’re counting them down, starting with feature number six, which is fleet integration. Digit ships with fleet management software built to plug directly into a warehous’s existing

00:31 systems, all without requiring the facility to rebuild its infrastructure around one robot. But software integration only matters if the robot can actually keep working. Which brings us to number five, uptime. Agility says Digit 5 hits a 10:1 runto charge ratio, translating to up to 20 hours of work per 24-hour shift, spending far more time on the floor than at a charging dock. Number four steps it up with real world training at scale because Agility says the system behind Digit 5 was

01:02 trained on 65,000 hours of real operational data, not lab simulation alone. So the company says this foundation is reliable. Then with number three, we get the actual specs. Digit 5 is rated for 23 kg or 50 lb of payload with a 2.2 m or 7.2 two foot reach that’s aimed squarely at industrial-grade lifting rather than delicate tasks at home. But number two is where it gets genuinely adaptable because Digit 5 runs on what Agility calls a physical AI stack, letting it

01:34 continuously expand which workflows it can perform paired with swappable grippers. So the same robot can actually switch tools for different jobs rather than being built for just one task. But the number one feature and the one that the entire product is actually built around is cooperative safety. Here, Agility says Digit 5 uses proprietary human detection software and functional safety behaviors specifically designed to let it work near people without external safety barriers or cages, which is the exact constraint that’s kept most

02:06 industrial robots fenced off from human workers until now. But one important caveat is that Agility’s own page confirms Digit 5 is still in development. So every spec listed is more of a preliminary estimate from pre-production hardware and some safety features aren’t fully built out yet. So this is more of a road map than a finished shipping product at this stage. But tell us in the comments when you think this will begin rolling off the production line like this next robot because UBTE just started production at what it’s calling the world’s first

02:37 10,000 unit scale humanoid robot smart factory and it’s located in China with one complete robot rolling off the line every 10 minutes. And says it scaled from 1,000 to 10,000 units of annual capacity in under 9 months, calling it just the beginning of large-scale humanoid industrialization. And what makes it special is that this facility is using a smart brain manufacturing system for full digital scheduling. So each robot gets over four hours of

03:07 testing plus an automotive grade inspection before shipping. And the factory primarily produces two robot lines as of now. The ube Walker S series which is an industrial robot for manufacturing and logistics and the Cruiser series which is more of a serviceoriented robot. But while humanoids are learning to build themselves, streaming multimodal intelligence like Google’s new Gemini 3.8 Live could end up serving as one of their many digital brains down the road. Maybe even one of yours. Because who’s

03:37 going to be smarter, faster, and cheaper than this? I’m looking to install an automatic water shut off valve and need to know the size of the main cold water line. Which one of these is the right pipe? The cold water line is typically the one with the valve that has a blue handle. In the video, the pipe on the right side leading into the heater seems to have a valve with a blue handle, which would mark it as the cold water inlet. Got it. Now, let’s measure the pipe. Looks like it’s a little shy of 1 in.

04:11 Based on what you’re showing, measuring a bit shy of 1 in in outside diameter, it’s highly likely you have a standard nominal 3/4 in copper pipe. That’s because copper pipe sizes are based on their nominal or inside diameter. So, the actual outside measurement is always slightly larger. Oh, that’s good to know. Help me find some videos to see if I can install it myself. Of course, I can help you find videos that show the installation process. Keep in mind that it often involves cutting into a main water line and can require specialized tools or skills. So, you’ll

04:43 want to see if it’s something you’re comfortable with. When you’re ready, I can search for videos specific to the type of valve you’re considering. So, Google just released two new voice AI models, Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, and they’re built to make talking to AI feel like an actual conversation rather than a request and response exchange. And the two models split by use case. Gemini 3.8 Live is built for scale and cost efficiency. And it combines conversational intelligence with fluid

05:15 dialogue and visual grounding. Then Gemini 3.8 live extended thinking is built for harder multi-step tasks but trades some speed for deeper reasoning. And on benchmarks, extended thinking currently holds the top spot on artificial analysis speech to text quality index. It leads agentic task completion on two of the top benchmarks and posts a near-perfect score on big bench audio and the standard Gemini 3.8 Live model came in a close second in the speech agent arena while remaining the

05:46 notably cheaper option to run of the two. And when it comes to functionality, both models process visual input in near real time, automatically detecting and switching between 97 supported languages midcon conversation. Plus, they can run tool calls and API requests in the background without breaking the conversation. And extended thinking goes even further with reasoning and speaking at the same time, using natural filler phrases like, “Let me check that,” while it works and narrating progress out loud during longer multi-step tasks. But

06:17 Google may be facing mounting competition for AI work tools as Anthropic just gave Claude its own productivity suite. Claude Docs and Claude slides are now in beta, letting users create documents and presentations directly inside a chat, then export, edit, and share them, joining Claude design and anthropics tool set. And it works like Google Docs, private by default, sharable for real-time co-editing with collaborators and Claude both able to leave comments. And everything built across Docs, Slides, or design lives behind one sharable link,

06:49 openable from a phone. Plus, Anthropic is also merging chat and co-work into one interface, letting Claude pick the right tool for a task automatically. And it’s rolling out to pro and max users now with team and free plans following soon. And finally, Meta just had two more announcements this week, one for developers and one for accessibility. First, Muse Code exited beta, adding inter session messaging, a workflows engine for coordinating multiple sub aents, a rewind feature, and a developer

07:19 preview SDK, along new subscription tiers starting at $5 a month, which undercuts claude code and codecs on price. And second, the Toronto startup Smart Arm unveiled a vision first bionic arm that sees through a palm mounted camera using Meta’s open-source Dino V2 model, and it can recognize everyday objects from just a few reference photos and then automatically select the right grip with no manual mode switching required. But what’s really cool is that optional Meta AI glasses can add a

07:50 firsterson view for extra context, which lets users personalize the arm to their own daily needs. Anyways, like and subscribe and check out this video here for more of the latest in AI and robotics news. And thanks for watching.