June 25, 2025 • 12 mins
In this episode, Bob provides an introduction and overview before diving into Google DeepMind's latest advancement with the Gemini Robotics AI model and its accompanying development kit, exploring its potential applications and impact. The discussion then moves to Amazon Web Services' Project Rainier, highlighting the capabilities of its new AI supercomputer and Trainium2 chips. Bob also examines innovations from Hewlett Packard Enterprise and NVIDIA's AI factory stack, considering their implications for industry advancements. The episode continues with an exploration of the United States Patent and Trademark Office's efforts to improve efficiency through AI. The episode wraps up with a summary of the key topics and closing remarks.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
(00:00):
Ever wondered what it would take for a robot tofunction seamlessly without needing to connect
to the internet?
Welcome to The AI News Daily Brief, your go-tofor the latest artificial intelligence updates.
Today is Wednesday, June 25, 2025.
Here’s what you need to know about GoogleDeepMind’s latest breakthrough in robotics.
(00:20):
Let’s dive in.
Google DeepMind is making waves with theirlatest rollout of an on-device version of the
Gemini Robotics artificial intelligence model.
This new development allows robots to operateindependently of an internet connection, a
significant step forward in the field ofrobotics.
(00:41):
Imagine a robot performing complex tasks withthe dexterity and understanding of a human, all
without needing to "phone home" forinstructions.
That’s the promise of this optimized artificialintelligence model.
Previously, the flagship Gemini Robotics modelrequired a hybrid approach, using both
on-device capabilities and cloud support toperform advanced tasks.
(01:04):
Now, with this new device-only model, users canaccess offline features that come close to
matching the capabilities of the flagshipversion.
Carolina Parada, head of robotics at GoogleDeepMind, highlighted the model’s efficiency,
saying it’s "small and efficient enough to rundirectly on a robot."
The on-device model is designed to perform avariety of tasks straight out of the box and
(01:29):
can adapt to new situations with just 50 to 100demonstrations.
This adaptability is crucial for real-worldapplications, where internet connectivity can
be unreliable or security requirements arestringent.
It’s a starter model that opens uppossibilities for companies with specific
needs, especially those in areas with poorconnectivity or strict security protocols.
(01:54):
To support developers and further enhance themodel’s capabilities, Google is also launching
a software development kit for this on-devicemodel.
This kit allows developers to evaluate andfine-tune the model, marking a first for one of
Google DeepMind’s vision-language-actionmodels.
The on-device Gemini Robotics model, along withits software development kit, will initially be
(02:17):
available to a select group of trusted testersas Google continues to refine its safety
protocols.
Project Rainier is set to redefine thelandscape of artificial intelligence computing.
Imagine a machine so powerful, it dwarfs allother attempts at creating a supercomputer for
training AI models.
This is exactly what Amazon Web Services hasembarked on with Project Rainier.
(02:41):
The project's scale is so massive, it's spreadacross multiple data centers in the United
States, a testament to Amazon's ambition tolead the next generation of AI.
Named after the iconic Mount Rainier, thisproject isn't just about power—it's about
unprecedented speed and agility.
AWS's AI safety and research partner,Anthropic, is among the first to harness this
(03:06):
enormous computational power.
According to Gadi Hutt, director of product andcustomer engineering at Annapurna Labs, Rainier
will provide Anthropic with five times morecomputing power than their current largest
training cluster.
The more compute power you have, the smarterand more accurate your AI models become.
(03:27):
It's like giving AI the keys to a much faster,more efficient brain.
At the heart of Project Rainier are theTrainium2 chips, custom-designed by AWS
specifically for AI training.
These aren't your average chips—they're capableof completing trillions of calculations per
second.
(03:47):
If you think about it, it would take a humanover 31,700 years just to count to one
trillion.
With Trainium2, that task is done in the blinkof an eye.
And Project Rainier doesn’t stop at one chip—ituses tens of thousands of these in
UltraServers, each housing 64 chips.
(04:07):
What makes these UltraServers special is theirability to communicate at lightning speed.
AWS has introduced specialized high-speedconnections called NeuronLinks, which allow
data to flow rapidly across the system,minimizing delays.
This setup is what Gadi Hutt affectionatelycalls a 'friendly giant,' a powerhouse of
(04:29):
computation ready to tackle the most complex AIchallenges.
But it's not just about raw power.
AWS is committed to sustainability, ensuringthat Project Rainier aligns with their
environmental goals.
They're using 100% renewable energy and havemade significant strides in water efficiency,
(04:51):
achieving rates more than twice as good as theindustry average.
This commitment ensures that even as they pushthe boundaries of what's possible with AI, they
do so responsibly.
In essence, Project Rainier is not just atechnical marvel; it's a beacon for the future
of AI.
By providing a template for deploying vastcomputational resources, it's paving the way
(05:14):
for breakthroughs in fields like medicine andclimate science.
Just as Mount Rainier stands tall and proud,Project Rainier marks a before-and-after moment
in the history of computing, promising totransform the technological landscape, one chip
at a time.
Imagine walking into a data center and seeing acomplex web of interconnected systems all
(05:36):
working in harmony to power the next big thingin artificial intelligence.
That’s exactly what Hewlett Packard Enterpriseand NVIDIA are offering with their new AI
factory stack, debuted at the HPE Discoverevent in Las Vegas.
This new offering is a game changer forindustries eager to embrace artificial
(05:56):
intelligence.
It’s not just about having powerful hardware;it’s about having a full-stack solution that
can be tailored for various AI applications.
Think of it like a one-stop shop for everythingyou need to get AI up and running smoothly.
The lineup includes modular AI factoryinfrastructure, HPE’s AI-ready RTX Pro Servers,
(06:19):
and the next generation of HPE’s turnkey AIplatform, known as HPE Private Cloud AI.
This combination gives enterprises a robustframework to build and scale generative,
agentic, and industrial AI solutions.
One of the key components is the NVIDIA AIComputing by HPE portfolio, which is among the
(06:41):
broadest in the market.
It combines NVIDIA’s accelerated computingtechnologies, such as the Blackwell chips,
Spectrum-X Ethernet, and BlueField-3networking, with HPE’s comprehensive suite of
servers, storage, services, and software.
This setup isn’t just about raw computingpower.
(07:01):
It’s about making AI deployment faster and moreefficient.
The pre-integrated, modular infrastructurestack helps teams get AI into production
without the usual hiccups.
It’s like having a streamlined assembly linefor AI innovation.
And there’s more.
HPE’s Private Cloud AI now comes with NVIDIA’slatest AI Blueprints, which include the AI-Q
(07:26):
Blueprint for creating AI agents and workflows.
These blueprints are like detailed guides thathelp companies design and implement AI
solutions effectively.
But it’s not just about the technology.
HPE is also focusing on security, governance,and sustainability.
They’ve added features like air-gappedmanagement, multi-tenancy, and post-quantum
(07:50):
cryptography to ensure that AI deployments aresecure and future-proof.
HPE is also expanding its ecosystem with 26 newpartners, offering over 70 packaged AI
workloads.
These range from fraud detection and videoanalytics to cybersecurity and sovereign AI,
providing a wide array of options for differentindustries.
(08:14):
To top it off, HPE is offering atry-before-you-buy program in collaboration
with Equinix data centers.
This allows customers to test the systems in areal-world environment before committing,
ensuring they meet the specific needs of theirbusiness.
In essence, this partnership between HPE andNVIDIA is setting a new standard for AI
(08:35):
deployment.
It’s not just about building powerful systems;it’s about creating an ecosystem that supports
rapid innovation and growth in the AIlandscape.
Have you ever thought about how artificialintelligence could revolutionize the way
patents and trademarks are examined?
Well, the United States Patent and TrademarkOffice is on the brink of doing just that.
(08:57):
They’re diving deeper into artificialintelligence to streamline their processes, and
trust me, it’s a fascinating journey.
The United States Patent and Trademark Officehas already made significant investments in
artificial intelligence, equipping theiremployees with tools to handle documents and
lessen the workload on examiners for thosetedious tasks.
(09:19):
But now, they’re looking to take things up anotch.
They’ve issued a request for information,asking industry experts for feedback on how
artificial intelligence can further enhance theefficiency of patent and trademark
examinations.
Greg Vidovich, the acting deputy commissionerfor patents, explained that they’re exploring
artificial intelligence for comprehensivesearch reports and drafting office actions.
(09:43):
It’s about using artificial intelligence tosimplify the complexities of patent
applications.
The United States Patent and Trademark Officewants to hear from vendors willing to offer
solutions at little to no cost, which is quitea unique approach in government contracting.
The United States Patent and Trademark Officeis not new to artificial intelligence, though.
(10:06):
Jamie Holcombe, the chief information officer,shared that their artificial intelligence
journey began four years ago.
They’ve been crafting machine learning pilotsthat have evolved into full-fledged
applications for classification, citation, anddetection of fraud.
Holcombe emphasizes that understanding yourdata is critical for successful artificial
intelligence implementation.
(10:29):
And it’s not just about the data.
The United States Patent and Trademark Officeis utilizing artificial intelligence tools like
the "More Like This Document" feature, whichhelps examiners find related documents more
efficiently.
Since its launch in March 2024, it’s been usedalmost 850,000 times, and its popularity is
(10:49):
still growing this year.
Another impressive tool is the Doc Code qualitycontrol, which automates the review of incoming
documents, ensuring accuracy and reducingcosts.
With over 18 million documents filed annually,this tool is a game changer for the United
States Patent and Trademark Office.
(11:11):
Deborah Stephens, the deputy chief informationofficer, highlighted their SCOUT tool, which
leverages generative artificial intelligencefor code development and cybersecurity threat
detection.
It’s already making waves internally, andthey’re gearing up to expand its use even
further.
So, what’s the takeaway from all this?
(11:32):
The United States Patent and Trademark Officeis not just dipping its toes in artificial
intelligence; it’s diving in headfirst, settinga precedent for how government agencies can
harness cutting-edge technology to improveefficiency and reduce costs.
That’s it for today’s AI News Daily Brief.
From Google DeepMind’s on-device robotics tothe United States Patent and Trademark Office’s
(11:56):
innovative use of artificial intelligence,today’s stories highlight how artificial
intelligence is reshaping industries across theboard.
Thanks for tuning in—subscribe to stay updated.
This is Bob, signing off.
Until next time.
Ever wondered what it would take for a robot tofunction seamlessly without needing to connect
to the internet?
Welcome to The AI News Daily Brief, your go-tofor the latest artificial intelligence updates.
Today is Wednesday, June 25, 2025.
Here’s what you need to know about GoogleDeepMind’s latest breakthrough in robotics.
(00:20):
Let’s dive in.
Google DeepMind is making waves with theirlatest rollout of an on-device version of the
Gemini Robotics artificial intelligence model.
This new development allows robots to operateindependently of an internet connection, a
significant step forward in the field ofrobotics.
(00:41):
Imagine a robot performing complex tasks withthe dexterity and understanding of a human, all
without needing to "phone home" forinstructions.
That’s the promise of this optimized artificialintelligence model.
Previously, the flagship Gemini Robotics modelrequired a hybrid approach, using both
on-device capabilities and cloud support toperform advanced tasks.
(01:04):
Now, with this new device-only model, users canaccess offline features that come close to
matching the capabilities of the flagshipversion.
Carolina Parada, head of robotics at GoogleDeepMind, highlighted the model’s efficiency,
saying it’s "small and efficient enough to rundirectly on a robot."
The on-device model is designed to perform avariety of tasks straight out of the box and
(01:29):
can adapt to new situations with just 50 to 100demonstrations.
This adaptability is crucial for real-worldapplications, where internet connectivity can
be unreliable or security requirements arestringent.
It’s a starter model that opens uppossibilities for companies with specific
needs, especially those in areas with poorconnectivity or strict security protocols.
(01:54):
To support developers and further enhance themodel’s capabilities, Google is also launching
a software development kit for this on-devicemodel.
This kit allows developers to evaluate andfine-tune the model, marking a first for one of
Google DeepMind’s vision-language-actionmodels.
The on-device Gemini Robotics model, along withits software development kit, will initially be
(02:17):
available to a select group of trusted testersas Google continues to refine its safety
protocols.
Project Rainier is set to redefine thelandscape of artificial intelligence computing.
Imagine a machine so powerful, it dwarfs allother attempts at creating a supercomputer for
training AI models.
This is exactly what Amazon Web Services hasembarked on with Project Rainier.
(02:41):
The project's scale is so massive, it's spreadacross multiple data centers in the United
States, a testament to Amazon's ambition tolead the next generation of AI.
Named after the iconic Mount Rainier, thisproject isn't just about power—it's about
unprecedented speed and agility.
AWS's AI safety and research partner,Anthropic, is among the first to harness this
(03:06):
enormous computational power.
According to Gadi Hutt, director of product andcustomer engineering at Annapurna Labs, Rainier
will provide Anthropic with five times morecomputing power than their current largest
training cluster.
The more compute power you have, the smarterand more accurate your AI models become.
(03:27):
It's like giving AI the keys to a much faster,more efficient brain.
At the heart of Project Rainier are theTrainium2 chips, custom-designed by AWS
specifically for AI training.
These aren't your average chips—they're capableof completing trillions of calculations per
second.
(03:47):
If you think about it, it would take a humanover 31,700 years just to count to one
trillion.
With Trainium2, that task is done in the blinkof an eye.
And Project Rainier doesn’t stop at one chip—ituses tens of thousands of these in
UltraServers, each housing 64 chips.
(04:07):
What makes these UltraServers special is theirability to communicate at lightning speed.
AWS has introduced specialized high-speedconnections called NeuronLinks, which allow
data to flow rapidly across the system,minimizing delays.
This setup is what Gadi Hutt affectionatelycalls a 'friendly giant,' a powerhouse of
(04:29):
computation ready to tackle the most complex AIchallenges.
But it's not just about raw power.
AWS is committed to sustainability, ensuringthat Project Rainier aligns with their
environmental goals.
They're using 100% renewable energy and havemade significant strides in water efficiency,
(04:51):
achieving rates more than twice as good as theindustry average.
This commitment ensures that even as they pushthe boundaries of what's possible with AI, they
do so responsibly.
In essence, Project Rainier is not just atechnical marvel; it's a beacon for the future
of AI.
By providing a template for deploying vastcomputational resources, it's paving the way
(05:14):
for breakthroughs in fields like medicine andclimate science.
Just as Mount Rainier stands tall and proud,Project Rainier marks a before-and-after moment
in the history of computing, promising totransform the technological landscape, one chip
at a time.
Imagine walking into a data center and seeing acomplex web of interconnected systems all
(05:36):
working in harmony to power the next big thingin artificial intelligence.
That’s exactly what Hewlett Packard Enterpriseand NVIDIA are offering with their new AI
factory stack, debuted at the HPE Discoverevent in Las Vegas.
This new offering is a game changer forindustries eager to embrace artificial
(05:56):
intelligence.
It’s not just about having powerful hardware;it’s about having a full-stack solution that
can be tailored for various AI applications.
Think of it like a one-stop shop for everythingyou need to get AI up and running smoothly.
The lineup includes modular AI factoryinfrastructure, HPE’s AI-ready RTX Pro Servers,
(06:19):
and the next generation of HPE’s turnkey AIplatform, known as HPE Private Cloud AI.
This combination gives enterprises a robustframework to build and scale generative,
agentic, and industrial AI solutions.
One of the key components is the NVIDIA AIComputing by HPE portfolio, which is among the
(06:41):
broadest in the market.
It combines NVIDIA’s accelerated computingtechnologies, such as the Blackwell chips,
Spectrum-X Ethernet, and BlueField-3networking, with HPE’s comprehensive suite of
servers, storage, services, and software.
This setup isn’t just about raw computingpower.
(07:01):
It’s about making AI deployment faster and moreefficient.
The pre-integrated, modular infrastructurestack helps teams get AI into production
without the usual hiccups.
It’s like having a streamlined assembly linefor AI innovation.
And there’s more.
HPE’s Private Cloud AI now comes with NVIDIA’slatest AI Blueprints, which include the AI-Q
(07:26):
Blueprint for creating AI agents and workflows.
These blueprints are like detailed guides thathelp companies design and implement AI
solutions effectively.
But it’s not just about the technology.
HPE is also focusing on security, governance,and sustainability.
They’ve added features like air-gappedmanagement, multi-tenancy, and post-quantum
(07:50):
cryptography to ensure that AI deployments aresecure and future-proof.
HPE is also expanding its ecosystem with 26 newpartners, offering over 70 packaged AI
workloads.
These range from fraud detection and videoanalytics to cybersecurity and sovereign AI,
providing a wide array of options for differentindustries.
(08:14):
To top it off, HPE is offering atry-before-you-buy program in collaboration
with Equinix data centers.
This allows customers to test the systems in areal-world environment before committing,
ensuring they meet the specific needs of theirbusiness.
In essence, this partnership between HPE andNVIDIA is setting a new standard for AI
(08:35):
deployment.
It’s not just about building powerful systems;it’s about creating an ecosystem that supports
rapid innovation and growth in the AIlandscape.
Have you ever thought about how artificialintelligence could revolutionize the way
patents and trademarks are examined?
Well, the United States Patent and TrademarkOffice is on the brink of doing just that.
(08:57):
They’re diving deeper into artificialintelligence to streamline their processes, and
trust me, it’s a fascinating journey.
The United States Patent and Trademark Officehas already made significant investments in
artificial intelligence, equipping theiremployees with tools to handle documents and
lessen the workload on examiners for thosetedious tasks.
(09:19):
But now, they’re looking to take things up anotch.
They’ve issued a request for information,asking industry experts for feedback on how
artificial intelligence can further enhance theefficiency of patent and trademark
examinations.
Greg Vidovich, the acting deputy commissionerfor patents, explained that they’re exploring
artificial intelligence for comprehensivesearch reports and drafting office actions.
(09:43):
It’s about using artificial intelligence tosimplify the complexities of patent
applications.
The United States Patent and Trademark Officewants to hear from vendors willing to offer
solutions at little to no cost, which is quitea unique approach in government contracting.
The United States Patent and Trademark Officeis not new to artificial intelligence, though.
(10:06):
Jamie Holcombe, the chief information officer,shared that their artificial intelligence
journey began four years ago.
They’ve been crafting machine learning pilotsthat have evolved into full-fledged
applications for classification, citation, anddetection of fraud.
Holcombe emphasizes that understanding yourdata is critical for successful artificial
intelligence implementation.
(10:29):
And it’s not just about the data.
The United States Patent and Trademark Officeis utilizing artificial intelligence tools like
the "More Like This Document" feature, whichhelps examiners find related documents more
efficiently.
Since its launch in March 2024, it’s been usedalmost 850,000 times, and its popularity is
(10:49):
still growing this year.
Another impressive tool is the Doc Code qualitycontrol, which automates the review of incoming
documents, ensuring accuracy and reducingcosts.
With over 18 million documents filed annually,this tool is a game changer for the United
States Patent and Trademark Office.
(11:11):
Deborah Stephens, the deputy chief informationofficer, highlighted their SCOUT tool, which
leverages generative artificial intelligencefor code development and cybersecurity threat
detection.
It’s already making waves internally, andthey’re gearing up to expand its use even
further.
So, what’s the takeaway from all this?
(11:32):
The United States Patent and Trademark Officeis not just dipping its toes in artificial
intelligence; it’s diving in headfirst, settinga precedent for how government agencies can
harness cutting-edge technology to improveefficiency and reduce costs.
That’s it for today’s AI News Daily Brief.
From Google DeepMind’s on-device robotics tothe United States Patent and Trademark Office’s
(11:56):
innovative use of artificial intelligence,today’s stories highlight how artificial
intelligence is reshaping industries across theboard.
Thanks for tuning in—subscribe to stay updated.
This is Bob, signing off.
Until next time.
Popular Podcasts
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
24/7 News: The Latest
The latest news in 4 minutes updated every hour, every day.
The Joe Rogan Experience
The official podcast of comedian Joe Rogan.