AI Voices Are Used To Steal Millions, Slack Integrates ChatGPT, and StabilityAI Acquires App With 15 Million Users

The next essay is in the making and will dive into the question of whether the world will run out of text data before we reach AGI.

The amount of time the newsletter takes up is already hard to justify and I am currently on skiing vacation. However, the conditions are far from optimal. So, we decided to take a day off. Despite the weather, machine learning progresses at a mind-bending speed that would dwarf any sumo wrestler sliding down a black slope on a plate slide.

Hence, I had to interrupt a long sequence of card games, grab a glass of single malt, and serve you some of the most noteworthy developments I came across in the last days.

We will start with Slack’s integration of ChatGPT. Next we will touch on StabilityAI acquisition of an image editing app with 15 million users. Last but not least, we will look at the uneasy topic of AI-enhanced fraud.

Let’s jump in!

​

Slack Integrates ChatGPT (Link)

Every slack user knows the feeling.

You come back from a holiday and you get swamped with notifications. It can take hours to read through near-infinite threads until you know what is going on in that important project your team is working on.

On Tuesday, the Salesforce-owned company Slack announced its integration with ChatGPT. The promise is that ChatGPT will allow us to summarize long threads and meeting notes and even draft responses to colleagues. I am sure it will not be long before nifty workers will extend their coffee breaks because they know that their replies on slack are taken care of.

What Should We Pay Attention To, Here?

Slack is another popular app that integrates language generation technology. These models particularly shine when it comes to creating drafts and overcoming the problem of getting started writing. How big the productivity gain from this will be remains to be shown. However, if every slack user globally can save a few minutes every day the impact of this could be huge.

​

Stability AI Acquires Image Editing App With 15 Million Users (Link)

The news broke on March 7th. StabilityAI acquired InitML the makers of the popular photo editing software ClipDrop. In the accompanying press release, StabilityAI’s CEO Emad Mostaque stated that the acquisition marks a milestone on the path to democratizing the power of foundational models.

Sure, that’s great! Democratization is awesome! *cough* buzzword *cough*

However, this shows again foundation models need distribution!

I have written about this extensively here but the problem basically boils down to the following: Companies that make foundation models such as GPT or Stable Diffusion are more like steel plants than they are like software firms. This can be seen through the following three points:

• Their product has high fixed and marginal costs
• Other companies replicate their product and the competition essentially drives down prices to the costs of providing the model. Hence, profits approach zero.
• They have little or no access to the end user and therefore cannot create defensibility through branding or UI

As a result, it is hard for foundational model companies to capture some of the value they create. If they would attempt to capture a vertical, they would need to build out internal sales teams and acquire specific domain knowledge. In doing so, they run the risk of changing their culture and scaring away their AI talent.

When OpenAI and Anthropic.ai partnered with Microsoft and Google respectively, they got access to vast networks of distribution. StabilityAI went the other way here. Instead of partnering with a large company, they acquired a smaller player that has already captured a significant market share. As a result they get:

• Direct access to consumers
• They can compete on branding and usability while generating sticky subscription revenue.
• They do not need to change their organization and preserve their strong focus on model development

​

AI Voices Are Used To Scam People Out Of Millions (Link)

A couple sent $15,000 in bitcoin to scammers after believing they had spoken to their son on the phone. So-called impostor scams are extremely common and can essentially. In 2022, a total of $11 million were stolen in the US alone by scammers impersonating loved ones in need.

Why Is This A Trend To Watch?

Generative AI essentially supercharges these types of scams. Modern systems require only a very small amount of data to create a believable replica of a person’s voice. These AI-enhanced scams allow for attacks on more and more sophisticated targets. There are documented cases of scammers impersonating senior executives to trick a bank manager into transferring $35 million to “complete an acquisition”. Generative models do not just enhance a scammer’s ability to perform social engineering. It was also shown that a deep fake hologram of Binance’s chief communication officer could be used to bypass biometric authentication solutions.

These are challenging but amazing times to be alive! I am excited for the future! Till next week!

Thank you for reading! As always, I really enjoyed making this for you and sincerely hope you found it useful!

At The Decoding ⭕, I send out a thoughtful 5-minute email every week that keeps you in the loop about machine learning research and the data economy. Click here to subscribe!

​

113 Cherry St , #92768, Seattle WA
​Unsubscribe · Preferences​