April 28, 2025 • 11 mins
In this episode of Hashtag Trending, hosted by Jim Love, the focus is on Anthropic's goal to demystify AI black boxes by 2027, spearheaded by CEO Dario Amodei. Elon Musk's social media platform X files a lawsuit against Minnesota's law criminalizing election-related deep fakes. Microsoft's Copilot faces disappointing adoption rates, likened to the infamous Clippy, and Google announces the end of software support for early versions of its Nest thermostats by 2025. The episode covers key developments and concerns in AI transparency, tech product longevity, and the legal landscape affecting social media and AI-generated content.
00:00 Introduction and Host Welcome
00:25 Anthropic's Mission to Demystify AI
03:37 Elon Musk's Legal Battle Over Deep Fakes
05:56 Microsoft Copilot: A Disappointing Launch
08:54 Google Ends Support for Older Nest Thermostats
10:22 Conclusion and Viewer Engagement
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
(00:00):
Anthropic aims to demystifyAI black boxes by 2027.
Elon Musk's lawsuit defends.
The use of deep fakes in politicalcampaigns has Microsoft Copilot Flatlined,
and Google Abandons earlier versionsof its Nest intelligent thermostat.
(00:20):
Welcome to hashtag Trending.
I'm your host, Jim Love.
Let's get into it.
Anthropic, CEO Dario Amodei has announceda bold plan to make the inner workings of
AI models significantly more transparent.
By 2027, the initiative targets one of thebiggest unsolved problems in artificial
(00:41):
intelligence, the black box mysteryof how models reach their conclusions.
In a recent essay, Amodei stressthat while today's AI models like
code and GPT-4 achieve remarkableresults, researchers still have only an
extremely limited understanding of whythese systems behave the way they do.
(01:03):
Anthropic is investing heavilyin new interpretability.
Research aimed at identifyinginternal neuron clusters that
correspond to understandable concepts.
Amodei who founded Anthropic todevelop safe AI is concerned about
the possibility of reaching artificialgeneral intelligence without
(01:23):
understanding how the models work.
He said, I'm very concerned aboutdeploying such systems without a
better handle on interpretability.
These systems will be absolutelycentral to the economy, technology, and
national security, and will be capableof so much autonomy that I consider it
basically unacceptable for humanity tobe totally ignorant of how they work.
(01:46):
And Anthropic has walked the talkon this developing what they've
termed mechanistic interpretability,a field that aims to open the black
box of AI models and understand whythey make the decisions they do.
There have been modestbut important successes.
Anthropic researchers were able toisolate and identify a neural unit in a
(02:08):
large language model that specificallytracked where it dealt with the concept
of what cities were in, what states.
It seems like a really tiny breakthrough,but it offers the promise that AI
behaviors might eventually be mappedto individual neurons or circuits.
Amodei says, our hope is to build aworld where we can say, with a very high
(02:29):
confidence what a model is thinking,reducing the risk of dangerous behaviors.
The push comes amid growing concerns aboutthe reliability and safety of powerful AI
systems without clear interpretability.
It's necessary and difficult to predictor control how models might act in
unexpected situations, especially as theyare increasingly deployed in healthcare
(02:53):
finance, and yes, even national security.
Anthropic is lobbying for governmentsto insist on other AI model developers
contributing to this knowledge and theoverall safety of AI models in general.
And while other companies rejectedthe recent California legislation
aimed at AI safety, Anthropic offeredways to improve the legislation.
(03:16):
So if they succeed, they couldhelp set new industry standards
for AI safety and transparency.
Better interpretability would notonly help catch risks earlier, but
could also build broader publictrust in advanced AI systems at a
time when skepticism is on the rise.
(03:37):
Elon Musk's Social Media Platform X.Formerly Twitter has filed a federal
lawsuit challenging Minnesota's 2023law that criminalizes the use of
AI generated deep fakes intended toinfluence elections or harm candidates.
The law imposes penalties includingjail time for creating and distributing
(03:58):
realistic fake media within specificelection related timeframes.
X argues that the law violates FirstAmendment rights and is preempted by
Section two 30 of the CommunicationsDecency Act, which shields social media
platforms from liability for user content.
The company contends that the statutecould criminalize harmless content,
(04:20):
including satire and would erode freespeech rather than defend democracy.
Minnesota Attorney General KeithEllison's office responsible for defending
the law is reviewing the lawsuit.
The law's author, democratic Senator AaronMay Quade criticized Musk for attempting
to manipulate political outcomes anddismissed his lawsuit as frivolous.
(04:42):
The Minnesota law has already facedlegal challenges from individuals
who publish AI generated politicalparodies, but a judge has denied
their request to suspend the law.
Proponents of the legislation argue thatit requires intent to injure a candidate
or influence an election result, andit has a very precise definition of
(05:03):
deep fakes as material so realisticthat a reasonable person would believe
it's real and generated by artificialintelligence or other technical means.
Still, some legal experts suggest the lawmay be struck down due to constitutional
concerns and the absence of legalexceptions for false political speech.
(05:23):
In Canada, we've seen some examplesof deep fakes used in the election
in fake news stories, and evenwhat we suspect is a video deep
fake of at least one candidate.
When that was circulated less than 48hours before the election, it becomes
next to impossible to absolutely prove itsfake, and to get social media companies
to take any action given past examples.
(05:46):
With things like robocalls, weknow that every technology will
be abused in elections without anyeffective penalties for doing this.
That's almost a certainty
Microsoft's co-pilot was supposed torevolutionize workplace productivity
instead, six months after launch,adoption rates are raising alarms and
(06:06):
some critics, including Salesforce, CEO,Mark Benioff are calling it Clippy 2.0.
Despite massive investment and rolloutacross Microsoft's office suite,
early adoption has been underwhelming.
Microsoft knows this.
Seeing that usage of copilot roseto about 20 million people worldwide
and then flatlined, and althoughfor some companies 20 million users
(06:29):
would be a real accomplishment inthe context of the overall numbers of
Microsoft users versus the billionsand billions of dollars invested.
This is a failure of proportionsthat does allow comparison
with the doomed Clippy app.
Clippy for those who don't know, wassupposed to be the latest in interactive
help systems, and I'm not sure itwas even bad for its time, but it
(06:50):
was despised, scorned, ridiculed, andunused, and Microsoft finally dropped it,
making it the icon of failed features.
According to the reporting fromXDA developers, Microsoft's
copilot weekly user base is only5% of the number of people who use
ChatGPT, and it's not increasing.
(07:12):
If there are approximately 1.5 billionWindows users worldwide, it means that
just over 1% of them are using copilot,and it's a tool that's now a Windows
default app adding to the criticism
When you look at how copilothas been delivered to its
customers, it's disappointing.
It just doesn't work.
It doesn't deliver any level of accuracy.
(07:34):
Copilot is more like Clippy 2.0.
And Benioff besides doing the burn, alsopointed to Gartner research that warns
copilot may be inadvertently exposingsensitive company data compounding
frustration among early adopters.
The comparison to Clippy thatmuch maligned animated paperclip
assistant from Microsoft's pastmay gloss over a deeper concern.
(07:58):
Users have reported that copilotoften struggles to understand context,
offers generic advice, sometimesgenerates inaccurate information,
and instead of boosting productivity,for some, at least, it's perceived
as just an intrusive add-on.
Microsoft has already had to back downon its copilot key and allow users to put
it to other uses, including reinstatingits original function as a menu key.
(08:24):
So Microsoft still has a chance to turnthings around by refining co-pilot's
utility and improving its integration.
But the company has placed a heavy beton co-pilot as the future of workplace,
and that hinges on broader adoption.
If businesses and users continueto see copilot as a cost without
clear benefit, Microsoft mightreally find itself reliving
(08:45):
another version of the Clippy Saga.
Only this time at a much higherfinancial and reputational cost.
Google has announced that it willend software support for the first
and second generation Nest learningthermostats, as well as the 2014 European
model, effective October 25th, 2025.
(09:06):
After this date, users will loseremote control capabilities via the
Google Home and Nest apps, as wellas Google assistant integration.
However, the thermostats willfunction locally allowing manual
adjustments directly on the device.
Google is offering discounts on newerthermostat models In the US customers
can receive $130 off on the fourthgeneration Nest Learning thermostat.
(09:30):
While Canadian users are eligible fora $160 discount, European users are
being offered a 50% discount on theTado Smart Thermostat X, as Google
has no plans to release additionalNest thermostats in Europe due to
the region's diverse heating systems.
The move underscores the challengesassociated with the longevity of
(09:50):
smart home devices as companiesphase out support for older hardware.
And while it's a minor thing, earlier,experiences have kept me at least from
moving ahead with integrating my home withGoogle Smart Home devices, as I'm always
worried about them dropping a product.
So I'm kind of glad that I kept thethermostat, which was included with
my furnace upgrade, which now hasall the features of the Nest model.
(10:12):
I would've paid more to install.
I wonder how many others have held backon new developments from Google because of
its reputation for killing off programs.
Love to hear from you,and that's our show.
You can reach me ateditorial@technewsday.ca, or on LinkedIn.
Many of you have, or onYouTube just under the video.
(10:33):
And if you wanna support the show,why not go to buy me a coffee.com
tech podcast and help keep us afloat.
I'm your host, Jim Love.
Have a marvelous Monday.
And if you live in Canada and youhaven't voted yet, whoever you wanna
support, why not get out there and vote?
Anthropic aims to demystifyAI black boxes by 2027.
Elon Musk's lawsuit defends.
The use of deep fakes in politicalcampaigns has Microsoft Copilot Flatlined,
and Google Abandons earlier versionsof its Nest intelligent thermostat.
(00:20):
Welcome to hashtag Trending.
I'm your host, Jim Love.
Let's get into it.
Anthropic, CEO Dario Amodei has announceda bold plan to make the inner workings of
AI models significantly more transparent.
By 2027, the initiative targets one of thebiggest unsolved problems in artificial
(00:41):
intelligence, the black box mysteryof how models reach their conclusions.
In a recent essay, Amodei stressthat while today's AI models like
code and GPT-4 achieve remarkableresults, researchers still have only an
extremely limited understanding of whythese systems behave the way they do.
(01:03):
Anthropic is investing heavilyin new interpretability.
Research aimed at identifyinginternal neuron clusters that
correspond to understandable concepts.
Amodei who founded Anthropic todevelop safe AI is concerned about
the possibility of reaching artificialgeneral intelligence without
(01:23):
understanding how the models work.
He said, I'm very concerned aboutdeploying such systems without a
better handle on interpretability.
These systems will be absolutelycentral to the economy, technology, and
national security, and will be capableof so much autonomy that I consider it
basically unacceptable for humanity tobe totally ignorant of how they work.
(01:46):
And Anthropic has walked the talkon this developing what they've
termed mechanistic interpretability,a field that aims to open the black
box of AI models and understand whythey make the decisions they do.
There have been modestbut important successes.
Anthropic researchers were able toisolate and identify a neural unit in a
(02:08):
large language model that specificallytracked where it dealt with the concept
of what cities were in, what states.
It seems like a really tiny breakthrough,but it offers the promise that AI
behaviors might eventually be mappedto individual neurons or circuits.
Amodei says, our hope is to build aworld where we can say, with a very high
(02:29):
confidence what a model is thinking,reducing the risk of dangerous behaviors.
The push comes amid growing concerns aboutthe reliability and safety of powerful AI
systems without clear interpretability.
It's necessary and difficult to predictor control how models might act in
unexpected situations, especially as theyare increasingly deployed in healthcare
(02:53):
finance, and yes, even national security.
Anthropic is lobbying for governmentsto insist on other AI model developers
contributing to this knowledge and theoverall safety of AI models in general.
And while other companies rejectedthe recent California legislation
aimed at AI safety, Anthropic offeredways to improve the legislation.
(03:16):
So if they succeed, they couldhelp set new industry standards
for AI safety and transparency.
Better interpretability would notonly help catch risks earlier, but
could also build broader publictrust in advanced AI systems at a
time when skepticism is on the rise.
(03:37):
Elon Musk's Social Media Platform X.Formerly Twitter has filed a federal
lawsuit challenging Minnesota's 2023law that criminalizes the use of
AI generated deep fakes intended toinfluence elections or harm candidates.
The law imposes penalties includingjail time for creating and distributing
(03:58):
realistic fake media within specificelection related timeframes.
X argues that the law violates FirstAmendment rights and is preempted by
Section two 30 of the CommunicationsDecency Act, which shields social media
platforms from liability for user content.
The company contends that the statutecould criminalize harmless content,
(04:20):
including satire and would erode freespeech rather than defend democracy.
Minnesota Attorney General KeithEllison's office responsible for defending
the law is reviewing the lawsuit.
The law's author, democratic Senator AaronMay Quade criticized Musk for attempting
to manipulate political outcomes anddismissed his lawsuit as frivolous.
(04:42):
The Minnesota law has already facedlegal challenges from individuals
who publish AI generated politicalparodies, but a judge has denied
their request to suspend the law.
Proponents of the legislation argue thatit requires intent to injure a candidate
or influence an election result, andit has a very precise definition of
(05:03):
deep fakes as material so realisticthat a reasonable person would believe
it's real and generated by artificialintelligence or other technical means.
Still, some legal experts suggest the lawmay be struck down due to constitutional
concerns and the absence of legalexceptions for false political speech.
(05:23):
In Canada, we've seen some examplesof deep fakes used in the election
in fake news stories, and evenwhat we suspect is a video deep
fake of at least one candidate.
When that was circulated less than 48hours before the election, it becomes
next to impossible to absolutely prove itsfake, and to get social media companies
to take any action given past examples.
(05:46):
With things like robocalls, weknow that every technology will
be abused in elections without anyeffective penalties for doing this.
That's almost a certainty
Microsoft's co-pilot was supposed torevolutionize workplace productivity
instead, six months after launch,adoption rates are raising alarms and
(06:06):
some critics, including Salesforce, CEO,Mark Benioff are calling it Clippy 2.0.
Despite massive investment and rolloutacross Microsoft's office suite,
early adoption has been underwhelming.
Microsoft knows this.
Seeing that usage of copilot roseto about 20 million people worldwide
and then flatlined, and althoughfor some companies 20 million users
(06:29):
would be a real accomplishment inthe context of the overall numbers of
Microsoft users versus the billionsand billions of dollars invested.
This is a failure of proportionsthat does allow comparison
with the doomed Clippy app.
Clippy for those who don't know, wassupposed to be the latest in interactive
help systems, and I'm not sure itwas even bad for its time, but it
(06:50):
was despised, scorned, ridiculed, andunused, and Microsoft finally dropped it,
making it the icon of failed features.
According to the reporting fromXDA developers, Microsoft's
copilot weekly user base is only5% of the number of people who use
ChatGPT, and it's not increasing.
(07:12):
If there are approximately 1.5 billionWindows users worldwide, it means that
just over 1% of them are using copilot,and it's a tool that's now a Windows
default app adding to the criticism
When you look at how copilothas been delivered to its
customers, it's disappointing.
It just doesn't work.
It doesn't deliver any level of accuracy.
(07:34):
Copilot is more like Clippy 2.0.
And Benioff besides doing the burn, alsopointed to Gartner research that warns
copilot may be inadvertently exposingsensitive company data compounding
frustration among early adopters.
The comparison to Clippy thatmuch maligned animated paperclip
assistant from Microsoft's pastmay gloss over a deeper concern.
(07:58):
Users have reported that copilotoften struggles to understand context,
offers generic advice, sometimesgenerates inaccurate information,
and instead of boosting productivity,for some, at least, it's perceived
as just an intrusive add-on.
Microsoft has already had to back downon its copilot key and allow users to put
it to other uses, including reinstatingits original function as a menu key.
(08:24):
So Microsoft still has a chance to turnthings around by refining co-pilot's
utility and improving its integration.
But the company has placed a heavy beton co-pilot as the future of workplace,
and that hinges on broader adoption.
If businesses and users continueto see copilot as a cost without
clear benefit, Microsoft mightreally find itself reliving
(08:45):
another version of the Clippy Saga.
Only this time at a much higherfinancial and reputational cost.
Google has announced that it willend software support for the first
and second generation Nest learningthermostats, as well as the 2014 European
model, effective October 25th, 2025.
(09:06):
After this date, users will loseremote control capabilities via the
Google Home and Nest apps, as wellas Google assistant integration.
However, the thermostats willfunction locally allowing manual
adjustments directly on the device.
Google is offering discounts on newerthermostat models In the US customers
can receive $130 off on the fourthgeneration Nest Learning thermostat.
(09:30):
While Canadian users are eligible fora $160 discount, European users are
being offered a 50% discount on theTado Smart Thermostat X, as Google
has no plans to release additionalNest thermostats in Europe due to
the region's diverse heating systems.
The move underscores the challengesassociated with the longevity of
(09:50):
smart home devices as companiesphase out support for older hardware.
And while it's a minor thing, earlier,experiences have kept me at least from
moving ahead with integrating my home withGoogle Smart Home devices, as I'm always
worried about them dropping a product.
So I'm kind of glad that I kept thethermostat, which was included with
my furnace upgrade, which now hasall the features of the Nest model.
(10:12):
I would've paid more to install.
I wonder how many others have held backon new developments from Google because of
its reputation for killing off programs.
Love to hear from you,and that's our show.
You can reach me ateditorial@technewsday.ca, or on LinkedIn.
Many of you have, or onYouTube just under the video.
(10:33):
And if you wanna support the show,why not go to buy me a coffee.com
tech podcast and help keep us afloat.
I'm your host, Jim Love.
Have a marvelous Monday.
And if you live in Canada and youhaven't voted yet, whoever you wanna
support, why not get out there and vote?
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.
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com
On Purpose with Jay Shetty
I’m Jay Shetty host of On Purpose the worlds #1 Mental Health podcast and I’m so grateful you found us. I started this podcast 5 years ago to invite you into conversations and workshops that are designed to help make you happier, healthier and more healed. I believe that when you (yes you) feel seen, heard and understood you’re able to deal with relationship struggles, work challenges and life’s ups and downs with more ease and grace. I interview experts, celebrities, thought leaders and athletes so that we can grow our mindset, build better habits and uncover a side of them we’ve never seen before. New episodes every Monday and Friday. Your support means the world to me and I don’t take it for granted — click the follow button and leave a review to help us spread the love with On Purpose. I can’t wait for you to listen to your first or 500th episode!