IRS Acquiring Nvidia Supercomputer, ChatGPT Energy Use Overestimated, and Complex Brains Evolved Independently

We're experimenting and would love to hear from you!

In this episode of 'Discover Daily', the IRS is making waves in government tech adoption with its upcoming acquisition of an Nvidia SuperPod AI supercomputer. This sophisticated system, featuring thirty-one servers with Blackwell processors, will be installed at their Martinsburg, West Virginia computing center. The system aims to revolutionize the agency's machine learning capabilities, particularly in fraud detection and taxpayer behavior analysis, marking a significant step forward in AI integration within government operations.

In a recent study on AI energy consumption, ChatGPT's latest model GPT-4 has been found to use significantly less power than previously thought. Research by Epoch AI reveals that each query consumes approximately 0.3 watt-hours, ten times lower than earlier estimates. This efficiency improvement is attributed to advanced hardware implementation, system optimization, and more accurate calculation methods, though the cumulative energy impact remains substantial given millions of daily queries.

Lastly, research published in Science has unveiled how birds and mammals have independently evolved complex brains through distinct pathways while achieving similar cognitive abilities. The study, led by Dr. Fernando García-Moreno, used advanced spatial transcriptomics and mathematical modeling to reveal fundamental differences in brain development between species. Despite using different genetic tools and developmental processes, both groups have achieved remarkable cognitive capabilities, challenging our traditional understanding of brain evolution and intelligence.

From Perplexity's Discover Feed: 

https://www.perplexity.ai/page/irs-acquiring-nvidia-supercomp-bQ9BUXJuSLOUpHZdCJ.vEA 

https://www.perplexity.ai/page/chatgpt-energy-use-overestimat-cn02azRBR2._eM_sH2n_Pw 

https://www.perplexity.ai/page/complex-brains-evolved-indepen-uPS546tuRJaWwidjGvrNCQ

**Introducing Perplexity Deep Research:**

https://www.perplexity.ai/hub/blog/introducing-perplexity-deep-research 

Perplexity is the fastest and most powerful way to search the web. Perplexity crawls the web and curates the most relevant and up-to-date sources (from academic papers to Reddit threads) to create the perfect response to any question or topic you’re interested in.

Take the world's knowledge with you anywhere. Available on iOS and Android

Join our growing Discord community for the latest updates and exclusive content.

Follow us on:

• Instagram
• Threads
• X (Twitter)
• YouTube
• Linkedin

Episode Transcript

Speaker 1 (00:01):
Welcome to Discover Daily by Perplexity, an
AI-generated show on tech,science and culture.
I'm Alex.
Today we're exploring how birdsand mammals have independently
evolved complex brains,challenging our understanding of
cognitive evolution.
But first let's look at whatelse is happening across the

(00:24):
world of technology and AI.
Our first story today, theInternal Revenue Service is set
to acquire an NVIDIA SuperPod AIsupercomputer.
This powerful system combines31 servers with advanced
Blackwell processors, eachcontaining eight processing

(00:45):
units.
It's designed to enhance theIRS's machine learning
capabilities for tasks such asfraud detection and taxpayer
behavior analysis.
The SuperPod will be installedat the IRS Computing Center in
Martinsburg, west Virginia.
While the exact cost isn'tpublic yet, these systems

(01:07):
typically start at $7 million.
The acquisition is part of abroader AI transformation within
the agency, which currently has68 different AI-related
projects in development.
While this system may besmaller than the massive AI
training centers used by techgiants like OpenAI and Meta, it

(01:27):
represents a significantcomputing resource for a
government agency.
This acquisition marks anotable technological
advancement for the IRS.
The Research, applied Analyticsand Statistics Division will
operate the SuperPOD as part oftheir Compliance Data Warehouse
project.
The system will come with asubstantial memory upgrade from

(01:51):
NVIDIA enabling moresophisticated data analysis.
Previous machine learningapplications by this division
have already shown promise inautomated fraud detection and
identity theft prevention.
While the hardware hasn't beenpurchased or installed yet, the
SuperPOD's advanced AIcapabilities are expected to

(02:13):
enhance the agency's ability toidentify complex patterns
indicative of fraudulentactivities.
This technology investmentreflects a growing trend of AI
adoption in governmentoperations, particularly in
areas requiring sophisticatedpattern recognition and data

(02:33):
analysis.
Moving on to our second story, anew study has revealed that
ChatGPT's energy consumption ismuch lower than previously
estimated.
The non-profit researchinstitute EPIC-AI found that
ChatGPT's latest model, gpt-4-0,consumes approximately 0.3

(02:57):
watt-hours per query ten timesless than earlier estimates.
But it's worth noting thatthese figures are approximations
, as OpenAI hasn't publiclydisclosed detailed energy
consumption data.
This significant reduction inenergy use can be attributed to
advancements in AI technologyand more accurate calculation

(03:18):
methods.
While some have comparedChatGPT's energy use to other
digital services, it's importantto understand that different
measurement methodologies makedirect comparisons challenging.
To put this into perspective,typical household activities
consume significantly moreenergy, with many common

(03:39):
appliances using hundreds ofwatt-hours per hour of operation
.
These comparisons helpillustrate the scale, though
exact figures can vary based onspecific devices and usage
patterns.
Several factors contribute tothis improved energy efficiency,
including the implementation ofmore advanced hardware like

(03:59):
NVIDIA H100 chips, optimizationof AI systems and updated
calculation methods that betterreflect current technology.
These measurements specificallyrefer to standard text-based
queries and don't includeenergy-intensive features like
image generation or processingextremely long conversations,

(04:21):
but the cumulative effect ofmillions of daily queries still
results in substantial energyconsumption.
The AI industry is investing inenergy-efficient hardware and
renewable energy sources tomitigate environmental impacts.
As models become moresophisticated and handle more

(04:45):
complex tasks, their energyrequirements may increase,
making these efficiencyimprovements even more crucial.
Now let's dive into our mainstory of the day.
Recent research has revealedthat birds and mammals have

(05:08):
independently evolved complexbrains through distinct
developmental, structural andgenetic pathways.
This showcases convergentevolution, achieving similar

(05:32):
cognitive abilities despiteprofound differences in brain
architecture and molecularmechanisms, and the University
of the Basque Country usedadvanced techniques in spatial
transcriptomics and mathematicalmodelling to uncover
fundamental differences in braindevelopment between birds and
mammals.
Published in Science, thisresearch actually comprised two

(05:53):
complementary studies thattogether paint a fascinating
picture of brain evolution.
One key finding is that thepallium, a critical brain region
responsible for cognitivefunctions, develops differently
across species, despite sharingsimilar functional outcomes.
In birds and mammals, while thecircuits perform comparable

(06:16):
tasks, the neurons form duringembryonic development at
distinct locations and times.
This divergence extends to thegenetic tools employed for
establishing cellular identity,resulting in unique cell types
across species.
What's particularly interestingis that inhibitory neurons have

(06:38):
remained remarkably similaracross species for hundreds of
millions of years, whileexcitatory neurons, responsible
for information transmission,have evolved along dramatically
different pathways.
In fact, only a small number ofexcitatory neurons in birds
share genetic similarities withtheir mammalian counterparts,

(06:59):
specifically in brain regionscalled the clostrum and
hippocampus.
The evolutionary paths of birdsand mammals diverged millions
of years ago, resulting indistinct sets of genes being
used for sensory processingneurons in birds and mammals.
Yet nature has shown us thatthere isn't just one blueprint

(07:22):
for building an intelligentbrain.
Despite these differences, bothbirds and mammals have achieved
remarkably similar cognitiveabilities through convergent
evolution.
Many bird species demonstrateimpressive problem-solving
abilities and complex socialbehaviors, showing us that
advanced cognition can emergefrom very different neural

(07:45):
architectures.
The study challenges thetraditional understanding of
brain evolution, demonstratingthat complex neural circuits can
arise through diversedevelopmental pathways.
It provides crucial insightsinto the genetic and cellular
mechanisms underlying theindependent evolution of complex

(08:07):
brains.
That's it for today.
Thanks for tuning in.
Don't forget to subscribe onyour favorite platform For more
info on anything we coveredtoday, check out the links in

(08:30):
our episode description anddon't forget you can now access
Perplexity's AI-poweredknowledge base on the go with
the mobile app, available forboth Android and iOS.
There's also the Perplexitydesktop app for macOS.
In other Perplexity news, thecompany just launched Deep

(08:51):
Research, a powerful new featurethat digs deep into hundreds of
sources to create thoroughresearch reports.
The tool, which is free to tryand unlimited for pro users, can
tackle complex researchquestions in just minutes.
Early tests show it'sremarkably accurate, making it a

(09:11):
promising tool for anyoneneeding in-depth research
assistance.
We'll be back with more storiesthat matter.
Until then, stay curious.