AI PULSE - DeepSeek Unveils Enhanced AI Model R1 And Brain-Computer Interface Breakthrough

AI's Impact on Talent Acquisition: Challenges and Adaptations for Recruiters and Job Seekers DeepSeek Unveils Enhanced AI Model R1, Rivals Meta and OpenAI on Hugging Face Perplexity Labs Empowers Pro Users with AI-Driven Reports and Interactive Web Apps 1000x Increase in AI Demand Meta's AI Assistant Reaches 1 Billion Users, Plans for Growth and Monetization Short Reasoning Chains Boost AI Accuracy and Efficiency, Reducing Costs Telegram Partners with Elon Musk's xAI to Launch Grok Chatbot with Revenue Sharing Brain-Computer Interface Breakthrough #AI, #TalentAcquisition, #DeepSeek, #BrainComputerInterface, #MetaAI, #xAI, #GrokChatbot

Episode Transcript

(00:00):
Welcome to Innovation Pulse, your quick no-nonsense update on the latest in AI.

(00:09):
First, we will cover the latest news.
AI is reshaping talent acquisition.
Chinese AI models are rising amid tech barriers.
Perplexity labs is expanding.
NVIDIA is scaling AI reasoning,
and Meta's AI assistant reaches a billion users.
After this, we'll dive deep into how Telegram's partnership
with XAI is transforming platform integration.

(00:33):
AI is revolutionizing talent acquisition,
with both companies and job seekers increasingly embracing its use.
Bonnie Dilber from Zapier notes the challenge of keeping up with AI's rapid changes in recruiting.
She predicts that within six months, companies will provide more transparency about AI and
hiring, including formal statements on its use and verification processes for candidates.

(00:58):
Employers plan to implement tests like on-site interviews and video verifications to detect AI use.
The interview process will become more rigorous, incorporating skills assessments and open-ended
questions to filter applicants.
AI will also play a role in initial screenings and candidate evaluations.

(01:21):
While AI aids in identifying top talent,
some job seekers are wary of its use, preferring human interaction to assess soft skills.
Companies aim to address document authenticity and identity verification through AI screening tools.
Chinese startup DeepSeq has quietly released an upgraded version of its AI reasoning model,

(01:47):
DeepSeq R1, on the AI model repository Hugging Face.
The original R1 model gained attention earlier this year for outperforming competitors like Meta
and OpenAI, raising concerns about US tech giants overspending on infrastructure.
Though markets have since stabilized, the upgraded R1 model is now just behind OpenAI's

(02:11):
latest models on the live code bench rankings.
According to Adina Yakifu from Hugging Face, the update enhances reasoning,
math and code capabilities, and reduces hallucinations, showing that DeepSeq is not only
catching up but also competing with top-tier models.
This development highlights the ongoing growth of Chinese AI,

(02:35):
despite US efforts to limit China's access to essential technology.
Meanwhile, Chinese tech giants Baidu and Tencent are optimizing their AI models to counter
US semiconductor export restrictions.
Join us as we discover the innovative power of Perplexity Labs.

(02:57):
Perplexity, an AI-driven search engine, recently launched Perplexity Labs for subscribers of its
pro plan, priced at $20 per month. This tool is designed for creating reports,
spreadsheets, dashboards, and more accessible via web, iOS, and Android,
with plans to expand to Mac and Windows apps.

(03:20):
Perplexity Labs aims to help users complete various projects by utilizing advanced tools like
file generation and mini-app creation. It can perform research and analysis,
crafting reports and visualizations in about 10 minutes, using web search,
code execution, and chart creation. The platform also supports interactive

(03:43):
web app creation and coding tasks. Files produced during these workflows are neatly
organized for easy access. Perplexity's expansion into corporate functions includes an enterprise
plan with user management and internal search features, fueled by investor interest and potential
capital, raises valuing the company at $18 billion.

(04:06):
Nvidia's recent earnings highlight a shift from one-shot AI to reasoning, enhancing accuracy for
tasks like financial projections. Reasoning demands significantly more computational power,
using thousands more tokens per task than simpler inference. This shift is evident in companies like

(04:30):
OpenAI, Microsoft, and Google, which are experiencing surges in token generation.
Microsoft's first quarter saw over 100 trillion tokens processed, a five-fold increase year over
year. Hyperscalers are deploying around 72,000 Nvidia GPUs weekly, fueling AI factories. The

(04:53):
demand surge has led to over $300 billion in capital expenditure for data centers, termed
AI factories by Nvidia. This pace is doubling the number of Nvidia-powered AI factories and GPUs
per factory. Despite algorithmic improvements, reducing model sizes, the demand for AI and

(05:14):
sophisticated reasoning continues to outstrip these advances, marking a significant growth
phase in the AI industry. Meta's AI assistant has reached one billion monthly users across its apps,
as announced by CEO Mark Zuckerberg during the annual shareholder meeting. The focus for this

(05:36):
year is enhancing personalization, voice interactions, and entertainment. Meta plans to keep expanding AI
before monetizing it through paid recommendations or subscription services. In February, plans
emerged for a standalone AI app, similar to ChatGPT. Despite the impressive user base,

(05:59):
Zuckerberg noted it doesn't feel fully scaled yet. At the shareholder meeting, 14 business-related
proposals were voted on. A notable one from JLens, affiliated with the Anti-Defamation League,
urged Meta to report on hate content following relaxed moderation policies. Proposals not
backed by Meta's board, like ending the dual-class share structure, were unlikely to pass, while

(06:24):
board-supported items were expected to succeed. Final results will be published soon.
Join us as we discuss the efficiency revolution in AI. Researchers from Meta's fair team and the
Hebrew University of Jerusalem found that shorter reasoning processes in AI improve performance and

(06:46):
reduce computational costs. Their study showed that AI accuracy increases by 34% with shorter
reasoning chains, challenging the belief that longer chains are better. The team introduced
ShortMac, a method that reduces computing costs by 40% while boosting performance by executing

(07:07):
multiple reasoning attempts in parallel and using majority voting. This approach could save tech
companies millions as it emphasizes efficiency over extensive computation. The findings suggest that
more concise reasoning enhances AI intelligence, contradicting existing trends in AI development
that favor scaling up computing resources. This research offers a new perspective on

(07:32):
optimizing AI systems, demonstrating that less thinking can lead to smarter outcomes.
Telegram has partnered with Elon Musk's AI company, XAI, to distribute the grok chatbot
through its platform for a year. In exchange, XAI will pay Telegram $300 million in cash and equity,

(07:56):
and Telegram will receive 50% of the revenue from XAI subscriptions purchased via the app.
Initially available to Telegram's premium users, grok is expected to be accessible to all users.
A video by Telegram's CEO, Pavel Durov, shows grok can be pinned in chats and accessed via the

(08:18):
search bar for various tasks, including writing suggestions, summarizing content, and creating
stickers. It will also assist with business inquiries and moderation. This move aligns with
Meta's integration of AI into Instagram and WhatsApp, highlighting a trend of incorporating AI

(08:38):
to enhance user experience across platforms. And now, pivot our discussion towards the main AI topic.
All right, everybody. Welcome to another deep dive here on Innovation Pulse.
I'm Alex, and as always, I've got my co-host, Yakov Lasker with me.

(09:02):
Today, we're diving into something that honestly sounds like it's straight out of a sci-fi movie,
but it's happening right now in labs across the country. Yakov, lay it on me. What are we talking
about today? Thanks, Alex. So picture this. You're sitting at your computer, and instead of typing
with your fingers, you're literally typing with your thoughts. Or maybe you're paralyzed,

(09:23):
but you can control a prosthetic arm just by thinking about moving your real arm.
This isn't some far-off fantasy anymore. We're talking about brain-computer interfaces,
and one company just made a massive leap forward that could change everything.
Wait, hold up. Brain-computer interfaces. I mean, I've heard the term thrown around,

(09:43):
but break this down for me. How does this actually work?
Great question. So brain-computer interfaces or BCIs are essentially systems that create a direct
communication pathway between your brain and external machines. Think of it like building
a bridge between your thoughts and technology. There are different approaches. Some use electrode-
covered caps that sit on your head to monitor brain activity. Kind of like a really sophisticated EEG

(10:08):
machine. Okay, that doesn't sound too invasive, but I'm guessing that's not the most effective
method. Exactly right. The real breakthrough systems actually place electrodes under your skull,
sometimes directly into brain tissue. Once those electrodes are in place, the system gets trained
to decode your specific brain activity patterns. So when you think about moving your hand, for

(10:30):
instance, the BCI translates those electrical signals into commands that can control a computer
cursor or a robotic limb. That's incredible, but also kind of terrifying from a surgical standpoint.
I'm guessing there's a company that just figured out how to make this less scary. Bingo. Enter

(10:50):
precision neuroscience, and here's where this story gets really interesting. These guys just
received FDA clearance for their brain interface system, and they're the first company in this
new wave of BCI developers to get commercial approval from the government for any part of their
technology. No kidding. That's huge. But tell me about this less invasive approach. What makes

(11:11):
their system different? This is where it gets fascinating. Their core technology is called
the Layer 7 cortical interface. And get this, it looks like a wristband you'd get at a music
festival. But don't let that fool you. This strip is thinner than a human hair, yet it contains
1,024 electrodes. That's more than many of their competitors. Wait, thinner than a hair, but with

(11:35):
over a thousand electrodes. How is that even possible? Right, and here's the brilliant part.
In the BCI world, more electrodes typically means greater accuracy in reading brain signals.
But the challenge has always been getting all those electrodes in place
without damaging brain tissue. Most BCI systems require surgeons to literally remove part of your
skull. Precision's array is so thin that it can rest on the surface of the brain just under the

(11:59):
skull, avoiding that tissue damage. So they're getting better performance with less risk. That
sounds like the holy grail of medical devices. How do they actually get it in there? This is my
favorite part. They slide it onto the motor cortex through an incision that's less than 1mm wide.
1mm. Compare that to traditional brain surgery, where they have to remove chunks of skull.

(12:23):
It's like the difference between keyhole surgery and major open surgery.
That's incredible. But I have to ask, have they actually tested this on people, or are we still
in the theoretical stage? They've been smart about this. So far, they've tested the Layer 7
interface in 39 patients. But here's the clever part. All these patients were already undergoing

(12:43):
brain surgery, for other reasons, like tumor removal. So the surgeons just used whatever access
they'd already created to place the array and test it. Smart strategy. No additional risk to
patients, but they get real world data. What kind of results are they seeing? Well, they had to
remove the device before the end of each surgery, so these were short trials. But because many patients

(13:05):
were awake during these procedures, which is actually common in brain surgery, precision
collected what their chief science officer called an incredibly interesting treasure trove of data.
And now with this FDA approval, I'm guessing they can collect a lot more data.
Exactly. The FDA cleared them for commercial use in patients for up to 30 days.
That's a game changer, because neural decoding algorithms like all AI-driven products need

(13:31):
vast amounts of data to work effectively. 30 days of continuous brain data from patients.
That's exponentially more information than they could gather from those brief surgical windows.
This sounds like a major breakthrough, but I keep hearing about other companies in this space too.
What about the competition? Isn't Elon Musk involved in this somehow?

(13:53):
Oh, absolutely. And here's a fun connection. Precision S co-founder and chief science officer,
Benjamin Rappaport, actually co-founded Neuralink with Elon Musk back in 2016.
But then he left to start precision with Michael Maeger in 2021. Talk about knowing the competition
from the inside out. Wow. So this is like the student becoming the master story.

(14:16):
How does Neuralink compare to what precision is doing?
Neuralink takes a different approach. They recently implanted their system
about the size of five stacked quarters directly into the motor cortex of their third patient,
a non-verbal man with ALS. He's actually using the system with the Grok AI chatbot to type using
just his thoughts. But their approach is more invasive, requiring that direct brain implantation.

(14:40):
And there are others in this race too, right? Definitely. There's Synchron,
which has backing from Bill Gates and Jeff Bezos investment firms. They're developing what might
be the least invasive approach of all. They reach the brain through blood vessels,
eliminating any need for skull drilling. They've tested it in at least 10 people and are
integrating chat GPT into their device. Blood vessels? That's wild. So we've got three different

(15:04):
approaches here. Precision surface level thin film, Neuralink's direct brain implant,
and Synchron's vascular route. But let's talk about the bigger picture.
Why is everyone so excited about this technology? The numbers tell the story.
About 5.4 million people in the US live with some form of paralysis.
Grandview Research estimated the total addressable market for BCIs

(15:28):
at $160 billion in 2024. And Morgan Stanley thinks that number could triple over the next decade.
We're potentially looking at a half trillion dollar market.
Half a trillion? That explains why billionaires are lining up to fund these companies.
But beyond the money, what does this mean for actual people?
Think about it. If these systems work as promised,

(15:50):
someone who's been paralyzed could control their environment again.
Type messages, control smart home devices, operate prosthetic limbs, maybe even drive
adapted vehicles. We're talking about restoring independence to millions of people.
And eventually this could extend to anyone. Imagine controlling your smartphone, computer,
or smart home just by thinking about it. That future sounds amazing, but we're not there yet,

(16:16):
right? What are the next steps for precision specifically?
Their next major milestone is clinical trials testing a fully implanted system,
which they're planning to begin in 2026. Right now they're preparing at their three US labs,
including one in Santa Clara, where they're running stress tests to make sure the ultra thin
device can handle the physical demands of surgery and implantation.

(16:39):
2026? So we're talking about less than two years for full trials.
That's moving pretty fast for medical technology. What needs to happen between now and then?
System integration is the big challenge. Their chief technology officer, Brian Otis,
explained they're putting all the components together. The flexible electrode array,
the implanted processing unit, the wireless transmission system, and the external devices.

(17:03):
Everything has to work seamlessly together, and it all has to be safe for long term implantation.
And I imagine getting through clinical trials and full FDA approval for an implanted system is going
to be a much bigger hurdle than this 30 day clearance they just received.
Absolutely. One of their executives made a great point about this. The difference between a prototype

(17:24):
and a product is like 100 fold in difficulty. Getting clearance for a 30 day monitoring device
is one thing. Proving that a permanently implanted brain computer interface is safe and effective
for long term use. That's a whole different level of regulatory scrutiny.
Makes sense. So while this FDA clearance is a huge milestone, we're still in the early stages

(17:45):
of what could be a very long journey. But if they succeed, the implications are enormous.
Exactly. And here's what I find most interesting. This isn't just about helping people with
paralysis anymore. If these systems become reliable and easy to use, they could fundamentally change
how we interact with technology. Imagine programming by thinking about code structure,

(18:06):
or creating art by visualizing it in your mind and having it appear on screen.
That opens up some fascinating possibilities, but also some big questions about privacy and
security. I mean, if these devices can read our thoughts, what happens to mental privacy?
That's a crucial point that the industry is going to have to address. Right now, these systems are

(18:26):
focused on reading specific motor intentions, like the thought of moving your hand. They're not
reading your private thoughts or emotions. But as the technology advances, those privacy concerns
are going to become more important. For sure. And I imagine there are still significant technical
hurdles too. This isn't exactly like developing a new smartphone app. Right, the brain is

(18:48):
incredibly complex and everyone's brain is different. These systems need to be trained for
each individual user. And they need to maintain accuracy over time as the brain adapts to the
implant. Plus, any surgical implant carries risks of infection, device failure, or tissue reaction.
So we're looking at a technology that could be revolutionary, but we're still in the early

(19:10):
adoption phase where the risks and benefits need to be carefully weighed for each patient.
Exactly. And that's why precision's approach is so smart. By making the implantation less invasive
and the device more biocompatible, they're trying to tip that risk-benefit equation
in favor of more patients being able to benefit from the technology.

(19:30):
Well, this has been fascinating. Before we wrap up, what's your take on where we'll be with this
technology in, say, five to ten years? I think we're going to see these systems become much
more reliable and accessible for people with paralysis and other neurological conditions.
The competition between precision, neural ink, synchron, and others is going to drive rapid

(19:53):
innovation. But for the broader consumer applications, typing with your thoughts,
controlling your smart home mentally, I think we're still looking at 10 to 15 years before that
becomes mainstream. That timeline makes sense. Medical applications first where the benefit
clearly outweighs the risk, and then gradual expansion to consumer uses as the technology

(20:15):
matures and becomes safer. Exactly. And the fact that we now have the first commercial FDA clearance
in this space, even if it's just for a 30-day monitoring system, shows that we've moved from
pure research into the realm of actual medical products. That's a huge psychological and regulatory
milestone. Absolutely. Well, folks, keep an eye on this space because brain-computer interfaces

(20:38):
are moving from science fiction to science fact faster than most people realize. Who knows?
In a few years, you might be sending texts or emails just by thinking about them,
and maybe more importantly, millions of people living with paralysis might regain some control
over their environment and independence in their daily lives. That's the real exciting part of

(20:59):
this story. Couldn't agree more. Thanks for breaking this down with me, Yakov. And thanks to
all of you for tuning in to Innovation Pulse. We'll be back next week with another deep dive
into the technologies that are shaping our future. Until then, keep innovating. Thanks for listening,
everyone. And that's a wrap for today's podcast. From AI's growing role in talent acquisition

(21:27):
and advancements in AI models like DeepSeek R1, to precision neuroscience's innovative brain-computer
interface, there are exciting developments on the horizon. Don't forget to like, subscribe,
and share this episode with your friends and colleagues so they can also stay updated on the
latest news and gain powerful insights. Stay tuned for more updates.