Good morning. There’s been a lot of talk about scaling laws, lately.

The industry is increasingly — if a little quietly — accepting the fact that scaling laws aren’t quite working anymore. The effort now is shifting beyond simply scaling compute and data and moving instead toward scaling time; time spent training, time spent post-training and time developers give a model to respond to a query.

As Dr. Gary Marcus wrote, “scaling can only take us so far; the time has come for fresh invention.”

— Ian Krietzberg, Editor-in-Chief, The Deep View

In today’s newsletter:

AI for Good: Brain mapping

We’ve talked before about just how little researchers know or understand about the human brain, human cognition and human consciousness. Even as some labs are attempting to design artificially intelligent systems that better emulate the many complexities and intricacies of human cognition, other researchers are leveraging the data-parsing capabilities of current models in an attempt to better understand the human brain. 

I’m talking about brain mapping. 

The details: Earlier this year, a group of neuroscientists at Harvard University teamed up with researchers at Google to construct an AI-generated map of part of the human brain. 

This one small sample contained 50,000 cells and about 150 million synapses, a demonstration of just how complex the human mind is.

How micro1 is Changing the Way We Find Top Engineering Talent

micro1 is the best platform to find top software engineers fast. They use AI to screen 50,000 engineers every month, train the top candidates on AI tools to 100x their output, then make the top 1% of talent available to hire with the click of a button.

We can personally recommend them at The Deep View as we've actually hired incredible engineering talent through them recently.

They're currently offering 240 free engineering hours exclusively for The Deep View subscribers. Click here to claim it.

Amazon to invest additional $4 billion in Anthropic 

Amazon on Friday said that it would invest an additional $4 billion in heavyweight AI startup Anthropic, bringing the tech giant’s total investment in the developer up to $8 billion. 

Anthropic said that Amazon will retain a minority stake in the company, though the exact percentage is unknown. 

The details: As part of the partnership, Amazon Web Services — Amazon’s cloud arm — has become Anthropic’s “primary training partner,” in addition to serving as the startup’s primary cloud provider. This means that Anthropic will use Amazon’s in-house chips to train and deploy its future models, instead of Nvidia. 

These Big Tech partnerships have attracted regulatory scrutiny around the world, with the FTC opening investigations into Anthropic’s relationship with Amazon and Google, as well as Microsoft’s partnership with OpenAI (Google has invested $2 billion in Anthropic, for a 10% stake in the company). 

The U.K. Competition and Markets Authority recently cleared Google’s partnership with Anthropic, several months after clearing the startup’s partnership with Amazon. 

Anthropic’s market share doubled to 24% this year, according to a recent report, while OpenAI’s fell significantly to 34% — the gap between the two is steadily shrinking. 

Sales Teams are Selling More with AI Role-Plays!

AI role-plays are becoming the crucial tool to unlocking sales success! Studies show that sales teams that use AI are already seeing a substantial increase in revenue vs. those that don’t.

Second Nature’s AI role-plays are built for sales teams:

Second Nature is one of Selling Power’s top AI coaching solutions and is already helping customers like Zoom, SAP, Oracle, TitleMax and Adobe see amazing results!

Book a quick demo or try a FREE AI simulation here

If you want to get in front of an audience of 200,000+ developers, business leaders and tech enthusiasts, get in touch with us here.

Judge partially dismisses another copyright lawsuit against OpenAI

Last week, a judge completely — and with prejudice — granted Microsoft’s motion to dismiss the copyright infringement lawsuit brought against the firm by media outlet The Intercept. The judge also partially granted OpenAI’s motion for a dismissal. 

The details: The initial lawsuit accused both OpenAI and Microsoft of violating two sections — b(1) and b(3) — of U.S. copyright code 1202. 

Let’s take a look at the law: Section b(3) of the copyright code holds that no one, without the express permission of the copyright holder, shall “distribute … copies of works … knowing that copyright management information has been removed or altered without authority of the copyright owner or the law.”

Section b(1) — the last remaining vestige of The Intercept’s argument — holds that no person shall, without the express permission of the copyright holder, “intentionally remove or alter any copyright management information.”

Similar to the recently dismissed case brought by RawStory and AlterNet, the central pillar of this case had to do with the removal of copyright-identifying information, such as headlines and bylines; it did not aim to address whether the act of training commercialized generative AI models on copyrighted content constitutes copyright infringement itself. 

To that end, while it will be interesting to see how this case turns out, this case will not decide the copyright question of generative AI. That burden falls instead on the New York Times and the Authors Guild, each of which has brought sweeping copyright infringement cases against OpenAI. 

Report: The hidden market of body-centric data

We hear often that, when it comes to artificial intelligence, data is king. 

This is perhaps best exemplified by the gradual adjustments in social media terms’ of service over the past few years; now, all of our data, beyond being used to fuel algorithms and hyper-targeted advertisements, is additionally being used to train generative AI models. 

While this has become relatively common knowledge, the scope of it isn’t. And according to new research out of the Mozilla Foundation, one category of this environment of data collection has been rapidly growing since the onset of the pandemic: body-centric data. 

What it means: Here, body-centric data refers to “information about people’s biological and psychological characteristics — from fingerprints used to unlock phones and face scans for security to intimate data from fitness and fertility trackers, mental health apps and digital medical records.” 

The findings: The report found that the rapid advancement and subsequent introduction of generative AI tools facilitates “continuous data processing without clear user awareness, thereby increasing the risk of data misuse.”

The report recommended the reconstruction of data privacy legislation that specifically classifies and defends against the collection and use of sensitive data, including bodily data. 

It also recommends an expansion of HIPAA in the U.S. to cover all health-related information — for instance, the data collected by fitness watches and health apps, which is not currently subject to the same restrictions and protections as information shared with your doctor — and an increase in transparency around data collection, data usage and opt-outs. 

“We need a new approach to our digital interactions that recognizes the fundamental rights of individuals to safeguard their bodily data, an issue that speaks directly to human autonomy and dignity,” Mozilla fellow Julia Keserű said. “As technology continues to advance, it is critical that our laws and practices evolve to meet the unique challenges of this era.”

The concerns laid out in this report echo and mirror concerns that researchers have been sharing for years. Several recent reports have highlighted risks of algorithmic discrimination and hallucination resulting in negative outcomes for patients, even as some hospitals push for integration. 

This is the dual nature of generative AI. The same data and the same systems that could be used to help people, can also be weaponized against people, which makes specific regulation extremely necessary. 

We need to try to prevent misuse before these things become more widespread. And we need to start caring about how our data is being used before this regime of data consumption and application — without our consent — becomes even more established as the fully acceptable norm.

🤔 Your thought process:

Selected Image 2 (Left):

Selected Image 2 (Left):

💭 A poll before you go

Thanks for reading today’s edition of The Deep View!

We’ll see you in the next one.

Here’s your view on safety testing as a part of AI infrastructure build-outs:

Nearly 40% of you said you would absolutely prioritize safety testing over speedy build-outs of AI infrastructure. 14% said they wouldn’t.

Yes, absolutely:

Yes, absolutely: