Hashtag Trending Sep 1- Can Google’s Gemini unseat GPT-4?; Millions of Duolingo user records published on hacking forum; Twitter comes for your biometrics

Can Google’s Gemini dethrone OpenAI’s GPT-4? Over 2.6 million Duolingo user records published on a hacking forum. And the return of the X files – the Musk is out there.

Hashtag Trending on Amazon Alexa Google Podcasts badge - 200 px wide


These stories and more as we bring you the top tech news on today’s Hashtag Trending.  

I’m your host James Roy. Happy Friday!

Google’s Gemini is expected to be released in fall 2023 but not much is known about its capabilities. Yet, promises continue to swirl that Gemini will unseat OpenAI’s GPT-4 as the most powerful, generative AI solution.

But is that really going to happen, and do we care at this point? 

Gemini, announced back in May, in collaboration with DeepMind, is a set of large language models (LLMs) that combine GPT-4 with training techniques taken from AlphaGo, such as reinforcement learning and tree search.

Analytics India Mag points out that if a company starts comparing its project to GPT-3 and GPT-4, we can tell, right off the bat, who the actual leader is.

Google is also insistent on making AI bold and responsible, which comes with many downs and a few ups. And we have seen how models start hallucinating the more we try to align them.

Most importantly in all this, Gemini is quite late to the game. Yes, the claim that Gemini will be multimodal is interesting, and it will have the data from YouTube that OpenAI does not have, but that’s it.

What could be a game changer is Reinforced Self-Training (ReST) for Language Modeling, which aims to remove humans from the process of reinforcement learning in machine learning and let LLMs build their own policy with just one initial command. Deep Mind just published a paper about this and if it’s able to integrate this in Gemini, the current capabilities of LLMs will improve.

However, at the same time, OpenAI just acquired Global Illumination, a company building creative simulations and training AI agents within it.

Google’s track record with Bard, PALM also does not inspire much optimism about Gemini, and OpenAI’s monopoly on the genAI market is just very hard to break right now.

What developers are looking for is an accessible API and not a locked out model like Bard or ChatGPT. And even for this, OpenAI is already on its way to make its models enterprise friendly by making it more data secure.

Source: Analytics India Mag

Over 2.6 million Duolingo user records obtained via data scraping were published on an underground hacking forum.

The data includes publicly available names and usernames, private email addresses, phone numbers, language, learning progress, achievements and more.

In January 2023, a threat actor was selling the same data on the Breached hacking forum for $1,500.

The language-learning platform said that its systems were not compromised during the incident.

Researchers explained that the private information was obtained by scraping an exposed application programming interface or API. The exposed API allows anybody to submit an email address and confirm if it is associated with a Duolingo account. On success, it returns personal data associated with the username. 

Subsequently, threat actors could use emails leaked in previous data breaches to confirm if the user has a Duolingo account. The previously leaked users’ data could contain additional information, such as phone numbers, allowing threat actors to execute social engineering and targeted phishing attacks on Duolingo users.

Max Gannon, a Senior Cyber Threat Intelligence Analyst at Cofense says that the scraped data doesn’t have much value outside of targeted attacks where the attacker spoofs Duolingo. Yet Duolingo users need to be wary of potentially spoofed communications.

Source: CPO Magazine

Meta has released a new tool to spot racial and gender bias within computer vision systems. It’s called FACET and stands for FAirness in Computer Vision EvaluaTion.

 Basically, the tool evaluates how well computer vision models perform across various characteristics including perceived gender and skin tone. 

The hope is to address the shortcomings in many computer vision models that have shown systematic bias against women and people of color.

Source: Axios

In more Meta news, the company has introduced a new privacy setting that lets you ask that your data is not used to train its AI models. It’s buried in Facebook’s Help Center under the entry Generative AI Data Subject Rights. There you can submit a form to make your plea.

This form, however, only relates to the “third party data” that Meta scraped, purchased, or licensed from outside sources. What sources, exactly? You don’t know. And the other data you’ve handed to Meta, that’s not covered.

And when you put your name and your email into this form, it’s hard to know what Meta does next. Presumably, the company has some kind of automated search that looks through the training data for its generative AI models to find exact matches for your name and email.

So if there’s information about you that doesn’t identify you by name, there’s nothing to worry about. 

Big Tech sucking up our data still bothers a lot of people. You can have legal rights to oppose that, depending on where you live. Which is why perhaps Meta asks for your country of residence in the form, so that it may not grant your request if you’re in a country where it doesn’t have to.

Source: Gizmodo

And on our continuing series, the X Files, here’s the latest from the artist formerly known as Twitter.

Elon Musk is racing to build “an everything app” to stay ahead of Meta’s Threads.  He announced in a post on X that video and audio calls are coming to the social media platform. He did not give a timeline for the rollout.

After rebranding Twitter as X, Musk signaled he would turn the platform into a super app offering a range of services from messaging and social networking to peer-to-peer payments.

Source: Reuters

Now let’s get to the gloomier news from X. 

Reportedly, the company has modified its privacy policy to include provisions for “Biometric Information” and “Job Applications/Recommendations.” It says that “based on your consent” the site may collect their biometric ID for “safety, security, and identification purposes.”

It gives no more details on what kind of biometric information it is looking for, but it could be an attempt to add another layer of verification on top of its $8 blue checkmark subscription. Like recently, it was testing out user verification using a government ID.

And in its crusade to become the ultimate superapp, Twitter is also bracing to become the next LinkedIn. The company says it wants to gather users’ employment history, educational history, and job search history” to “recommend potential jobs to you.”  It’s unclear whether Twitter will also be sharing users’ posting history or more information with potential employers.

Source: Gizmodo

That’s the top tech news stories for today.  Hashtag Trending goes to air 5 days a week, with a special weekend interview show we call “the Weekend Edition.”

You can get us anywhere you get audio podcasts and there is a copy of the show notes at itworldcanada.com/podcasts where you also can find links and more.

If you want to catch up on these and other news quickly, you can read these stories and more at TechNewsDay.com and ITWorldCanada.com on the home page.

It’s always great to hear from you. So please go to the article at itworldcanada.com/podcasts – and you’ll find a text edition there. Click on the x or the check mark, and tell us what you think. 

I’m your host, James Roy. Have a Fantastic Friday and a great long weekend if you are in Canada or the US! We’ll be back with a newscast next Tuesday.

Would you recommend this article?


Thanks for taking the time to let us know what you think of this article!
We'd love to hear your opinion about this or any other story you read in our publication.

Jim Love, Chief Content Officer, IT World Canada

Follow this Podcast

More #Hashtag Trending