Tech Brew Ride Home - Wed. 03/29 – Hit Pause On AI Development?
Episode Date: March 29, 2023Over 1,000 big names sign a petition urging a pause on AI development. What analysts think AI could to for or to the economy. North Korean hackers allegedly have a new trick. And did you know ByteDanc...e has an Instagram rival? Well, it looks like they’re putting the pedal to the metal on that. Sponsors: Miro.com/podcast go.tech/tm Links: Elon Musk and top AI researchers call for pause on ‘giant AI experiments’ (The Verge) Generative AI set to affect 300mn jobs across major economies (Financial Times) North Korea Is Now Mining Crypto to Launder Its Stolen Loot (Wired) Google Search is adding new ‘Perspectives’ and ‘About this author’ features to help users verify info (TechCrunch) TikTok ban backup plan? ByteDance-owned Instagram rival Lemon8 hits the US App Store’s Top 10 (TechCrunch) Cerebras releases seven large language models for generative AI, trained on its specialized hardware (SiliconAngle) Google Partners with AI Startup Replit to Take on Microsoft’s GitHub (Bloomberg) Microsoft Security Copilot is a new GPT-4 AI assistant for cybersecurity (The Verge) Learn more about your ad choices. Visit megaphone.fm/adchoices
Transcript
Discussion (0)
Welcome to the Tech meme right home for Wednesday, March 29th, 2020.
I'm Brian McCullough today. Over a thousand big names sign a petition urging a pause on AI development.
What analysts think AI could do four or two the global economy? North Korean hackers allegedly have a new trick.
And did you know that Bight Dance has an Instagram rival? Well, it looks like they're putting the pedal to the metal on that.
Here's what you miss today in the world of tech.
Over 1,000 people, including Elon Musk, have signed an open letter urging.
AI Labs to pause training of systems more powerful than GPT4 for at least six months.
Quoting the Verge.
The letter published by the nonprofit Future of Life Institute notes that AI Labs are currently
locked in an out-of-control race, their words, to develop and deploy machine learning systems,
quote, that no one, not even their creators, can understand, predict, or reliably control.
Therefore, we call on all AI labs to immediately pause for at least six months the training
of AI systems more powerful than GPT4.
says the letter. This pause should be public and verifiable and include all key actors. If such a pause
could not be enacted quickly, governments should step in and institute a moratorium, end quote.
Signatories include author Yuval Noah Harari, Apple co-founder Steve Wozniak, Skype co-founder,
Jan Tallinn, politician Andrew Yang, and a number of well-known AI researchers and CEOs, including
Stuart Russell, Yoshua Benjiyo, Gary Marcus, and Ahmad Mostak. The full list of signatories can be found
in the piece linked in the show notes, though new names should be treated with caution as there
are reports of names being added to the list as a joke. For example, OpenAI CEO Sam Altman,
an individual who is partly responsible for the current race dynamic in AI recently saw his name
included. The letter is unlikely to have any effect on the current climate in AI research,
which has seen tech companies like Google and Microsoft rush to deploy new products, often sideline
previously avowed concerns over safety and ethics. But it is a sign of the growing opposition
to this ship it now and fix it later approach, an opposition that could potentially make its way
into the political domain for consideration by actual legislators. As noted in the letter,
even Open AI itself has expressed the potential need for, quote, independent review of future
AI systems to ensure they meet safety standards. The signatories say that this time has now come,
quote, AI labs and independent experts should use this pause to jointly develop and implement
a set of shared safety protocols for advanced AI design and development that are rigorously
audited and overseen by independent outside experts, they write. These protocols should ensure that
systems adhering to them are safe beyond a reasonable doubt, end quote. Along similar lines,
lots of people were chatting overnight about this Goldman Sachs research paper that suggests
that generative AI could increase global annual GDP by 7% over a 10-year period, but at the same
time could expose 300 million full-time jobs to automation. Quoting the Financial Times.
They calculate that roughly two-thirds of jobs in the U.S. and Europe are exposed to some degree
of AI automation based on data on the tasks typically performed in thousands of occupations.
Most people would see less than half of their workload automated and would probably
continue in their jobs with some of their time freed up for more productive activities.
In the U.S., this should apply to 63% of the workforce they calculated, a further 30% working
in physical or outdoor jobs would be largely unaffected, although their work might be susceptible
to other forms of automation. But about 7% of U.S. workers are in jobs where at least half of their
tasks could be done by generative AI and are thus vulnerable to replacement. Goldman said
its research pointed to a similar impact in Europe. At a global level, since manual jobs
are a bigger share of employment in the developing world, it estimates about a fifth of work
could be done by AI or about 300 million full-time jobs across big economies.
quote. Researchers say North Korean hackers are likely laundering stolen crypto by renting cloud
compute to mine fresh coins, thereby avoiding more scrutinized crypto mixers, quoting wired.
Cybersecurity firm Mandiant published a report on a prolific North Korean state-sponsored hacking
group. It is now calling APT-43, sometimes known by the names Kemsuki and Thalium,
the group whose activity suggests its members' work in the service of North Korea's reconnaissance
General Bureau Spy Agency has been primarily focused on espionage, hacking think tanks, academics,
and private industry from the U.S. to Europe, South Korea, and Japan, since at least 2018,
mostly with fishing campaigns designed to harvest credentials from victims and plant malware
on their machines. Like many North Korean hacker groups, APT-43, also maintains a sideline
in profit-focused cybercrime, according to Mandiant, stealing any cryptocurrency that can enrich
the North Korean regime, or even just fund the hackers' own operations. And as
Regulators worldwide have tightened their grip on exchanges and laundering services that thieves and hackers use to cash out criminally tainted coins,
APT 43 appears to be trying out a new method to cash out the funds it steals while preventing them from being seized or frozen.
It pays that stolen cryptocurrency into hashing services that allow anyone to rent time on computers used to mine cryptocurrency,
thereby harvesting newly minted coins that have no apparent ties to criminal activity.
That minting trick allows APT43 to take advantage of the fact that cryptocurrency,
is relatively easy to steal while avoiding the forensic trail of evidence that it leaves on
blockchains, which can make it difficult for thieves to cash out. It breaks the chain, says Joe Dobson,
a mandiant threat intelligence analyst. This is like a bank robber stealing silver from a bank vault
and then going to a gold miner and paying the miner in stolen silver. Everyone's looking for the
silver while the bank robbers walking around with fresh newly minted gold, end quote. In theory,
the payouts from those pools should be clean with no ties to APT-43's hackers. That seems,
after all, to be the point of the group's laundering exercise. But in some cases of operational
sloppiness, Mandate said it found that the funds were nonetheless co-mingled with crypto and wallets
it had previously identified from its years-long tracking of APT-43 hacking campaigns, end quote.
Google plans to add a perspectives carousel to search, showcasing experts and others under
a new top stories label. Also, an about-this author feature is coming to U.S. English searches,
quoting TechCrunch.
The new prospectus feature is a carousel that will appear below top stories and showcase insights
from a range of journalists, experts, and other relevant voices on the topic you're searching
for.
The idea behind the feature is to give users a variety of noteworthy voices on a news topic to broaden
their understanding of the subject matter.
The carousel will launch soon in English in the U.S.
and will be available on desktop and mobile, Google says.
Google is also launching a new feature called About This Author that lets users easily learn
more about the authors behind the content they're reading.
With this new feature, users will be able to find more information about the background of the authors that Google surfaces on search.
The feature is launching on search results in English globally and on the perspectives carousel in the U.S. in English.
The About This Author feature is an expansion of Google's current About This Result feature, which first rolled out in 2021 in English.
The company announced that it's now launching the About This Result feature globally in all languages where search is available in the coming days.
Now all users will see three dots next to most results on Google search.
Tapping those three dots gives users details about where the information is coming from and how Google's system determined it would be useful for the query.
In addition, Google announced that it's making it easier for pages to access its about this page feature starting today.
When you click on the three dots next to a result, you can learn more information about the source and topic of a particular page.
The company said in a blog post, now we're making this information even easier to access.
Say you're searching for a rainforest protection organization.
Starting today, you can type in the URL of the organization in Google Search.
and information from about this page will populate at the top of search. You'll be able to quickly
see how the website describes itself, what others on the web have said about a site, and any recent
coverage of it. From there, you can evaluate whether you want to visit the website and learn more.
This feature is now available globally in English, end quote.
Lemon 8, ByteDance's Instagram rival, launched in March 2020, has hit the U.S. App Store's
top 10 for the first time ever. According to Data AI, the app,
never before even ranked in the top 200 overall charts. I was today years old when I learned that
ByteDance even had an Instagram rival, quoting TechCrunch. This is a dramatic move for the little
known app and one that points to a paid user acquisition effort powering the surge. The firm confirms
that such a fast move from being an unranked app to being number nine among top free apps in the
US, ahead of YouTube, WhatsApp, Gmail, and Facebook implies a significant and recent user acquisition push
on the app publishers part. Unfortunately, because the app is so new to the app store's top charts,
third-party app analytics firms don't have the precise data on Lemonade's U.S. installs or how those
installs have recently changed over the past few days. But given the app was launched globally back
in March 2020, what is most likely is that it was quietly released on the U.S. App Store,
but only for testing purposes. Then sometime over the past few days it was more officially launched,
meaning it was accompanied by this clearly sizable spend on paid discovery or app install ads.
According to App Intelligence provider Apptopia's data, Lemonate debuted on both iOS and Android in March 2020 and has since gained 16 million global downloads, with Japan as its top market, accounting for 38% of its total installs.
While the firm also doesn't have a figure for its U.S. installs, it was able to estimate the app currently has 4.25 million monthly active users.
Apptopia noted it didn't yet see Lemonate having spent on paid search on either the App Store or Google Play, but cautioned it may have paid install campaigns that just haven't populated in its system.
them yet or spend that's on networks it doesn't have insight into. However, we believe
ByteDance might simply be leveraging one of its own channels to drive app installs TikTok.
Over on TikTok, we noticed a number of creators recently began posting about Lemonate,
with many new videos appearing in just the past 24 hours. Concerningly, many of their reviews
are extremely positive but are not marked as sponsored content, end quote.
Finally today, it's back to the AI stuff, because look, people, the way the headlines are coming
fast and furious every day, I could do nothing but AI headlines some days. So I'm going to spare
you that by throwing a whole bunch of related headlines into this one segment. All AI and all generally
of interest to the devs out there. First up, Sarah Bras has open-sourced seven GPT-based
LLMs, ranging from 100 million to 13 billion parameters and trained using its Andromeda
supercomputer for AI. Let's release this to GitHub and Hugging Face, quoting Silicon Angle.
The new LLMs are notable as they are the first to be trained using CS2 systems in the Cerebrus
Andromeda AI Supercluster, which are powered by the Cerebrus WSE2 chip that is specifically designed
to run AI software. In other words, they're among the first LLMs to be trained without relying
on graphics processing unit-based systems. So Rebrush says it's sharing not only the models, but also
the weights and training recipe that was used via a standard Apache 2.0 license. The Sunnyvale,
California-based startup is backed by more than $720 million in venture funding. The company sells a chip
called the WSC2 that's specifically designed to run AI software. It's the WSC2 that sits at the heart
of the Cerebrus-Andromeda supercomputer, optimized to run AI applications, boasting more than 13.5 million
processor cores, end quote. Then, to compete with GitHub co-pilot, Google has partnered with
Replet, which used AI to make coding tools and now will rely on Google's LLMs for its
Ghostwriter product, quoting Bloomberg. Repplet, which has 20 million users, said its Ghostwriter
app will rely on Google's language generation AI to improve its ability to suggest blocks of
code, complete programs, and answer developer questions. Google Cloud Vice President June Yang
declined to specify which language AI products Replit will use, noting that it's a customized
combination of systems that address different tasks like chat and code generation. Previously, Replis
built the product with its own AI. Google has much better technology than most people know,
Repplet chief executive officer Amjad Massad said in an interview. The startup will also expand
its use of Google's cloud services and hopes the relationship with the tech giant will help it win
over larger corporate customers. Right now, Replet's clients are largely individual developers and
startups. Google will also distribute Replet's software as part of the partnership, end quote. And finally,
finally, it's not just coding, it's cybersecurity too. Microsoft has launched security co-pilot,
a GPT4-powered assistant to help security professionals with incident investigations, event
summaries, reporting, and more. Quoting the verge, powered by OpenAI's GPT4, generative AI,
and Microsoft's own security specific model. Security co-pilot looks like a simple prompt
box like any other chatbot. You can ask, what are all the security incidents in my enterprise,
and it will summarize them. But behind the scenes, it's making use of the 65.
trillion daily signals Microsoft collects in its threat intelligence gathering and security-specific
skills to let security professionals hunt down threats. Microsoft Security Copilot is designed to assist
a security analyst's work rather than replace it and even includes a pinboard section for
co-workers to collaborate and share information. Security professionals can use security
co-pilot to help with incident investigations or to quickly summarize events and help with reporting.
Security Copilot accepts natural language inputs so security professionals could ask for a summary
of a particular vulnerability, feed in files, URLs, or code snippets for analysis, or ask for
incident and alert information from other security tools. All prompts and responses are saved,
so there's a full audit trail for investigators. Results can be pinned and summarized into a shared
workspace so colleagues can all work on the same threat analysis and investigations. This is
like having individual workspaces for investigators and a shared notebook with the ability to promote
things you're working on, says Chang Kawaguchi, an AI security architect at Microsoft, in
an interview with The Verge, end quote. I asked this last night on Twitter, but can I play any of the
original Guitar Hero games on the PlayStation 5? I can't, right? That's a franchise that they've inexplicably let
die. If there is a way to do it, can someone please let me know because I've got a nine-year-old
that would be perfect for that game? And if there's not a way to do it, again, I want to play
it on the PS5, if possible. Dude, that's a huge mistake. You know how nostalgia cycles run in
roughly 20-year cadences.
Well, right about now is exactly the time when a bunch of folks who grew up on that game
would want to start playing it with their kids.
Anybody listening with access to that guitar hero IP, get on that.
Talk to you tomorrow.
