Good morning. Sean Astin (AKA, Samwise Gamgee, from Lord of the Rings) came out alongside SAG-AFTRA in support of SB-1047, urging Gov. Gavin Newsom to sign the bill.

As someone pointed out on Twitter, o1 (OpenAI’s new model) spelled backward is One Ring…

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

In today’s newsletter:

• 🪐 AI for Good: Exoplanet discovery

• 🤖 Google DeepMind makes progress in robotic dexterity

• 🧑‍🍳 Chipotle adds robots to its staff

• 💻 Microsoft makes Bill Gates’ AI dreams come true

AI for Good: Exoplanet discovery 

For four years, NASA’s Kepler Mission observed more than 200,000 stars in a deep search for signs of distant planets in orbit around stars. 

In 2021, NASA deployed an AI algorithm — ExoMiner — to parse through the years of data gathered by the mission to identify exoplanets.

The algorithm added 301 validated exoplanets to NASA’s total exoplanet tally. 

The details: The deep neural network is able to distinguish a real (or validated) exoplanet from any number of false positives. The process of validation involves statistical measures of certain planet-specific features. 

• The 301 planets identified by ExoMiner were unable to be validated by scientists before the introduction of the algorithm. 

• Perhaps the most important quality of ExoMiner is that, according to NASA, it “isn't a black box — there is no mystery as to why it decides something is a planet or not.”

Why it matters: This level of guaranteed explainability makes the system reliable and trustworthy, allowing it be leveraged by scientists without restraint. Now that it’s validated on the Kepler mission, the same system will be used on other missions, all to aid the search for more exoplanets.

How to demonstrate secure AI practices with ISO 42001

AI is everywhere, but customers need to know that you're using it safely and responsibly. The ISO 42001 standard helps companies demonstrate their AI security practices in a verifiable way.

Join Vanta and A-LIGN for a Coffee and Compliance session on ISO 42001—what it is, what types of organizations need it, and how it works. The discussion also covers practical strategies and best practices for successfully integrating ISO 42001 into your organization.

Watch the recorded session.

Google DeepMind makes progress toward achieving robotic dexterity

Google DeepMind last week published two new papers on achieving greater levels of robotic dexterity, something that might enable robots to become actually useful and usable in daily life. 

The details: Where most advanced robots can only pick up and place objects with a single arm, DeepMind’s research — ALOHA Unleashed — was able to crack bi-arm manipulation. 

• The two-armed robotic system was trained to mimic tasks that were first remotely demonstrated by the researchers. The robot was able to tie a shoelace, hang a shirt and clean a kitchen. 

• DeepMind at the same time unveiled DemoStart, a reinforcement learning algorithm that helps robots “learn” these highly dexterous tasks in simulation. 

“One day, AI robots will help people with all kinds of tasks at home, in the workplace and more,” DeepMind said. “Dexterity research, including the efficient and general learning approaches we’ve described today, will help make that future possible.”

Despite recent advances in robotics, that end goal of developing a C-3PO-like butler robot, brimming with common sense, sarcasm and the ability to interact with the physical world, as DeepMind notes, is far away. The challenges faced by robotics go beyond those faced by AI (reasoning and reliability), and into challenges around sensors and mechanics and physics. 

Transform your hiring game with Ashby

For a hectic startup, hiring can be a messy process. Ashby makes it so much easier. (That’s why writing this particular slot was so much fun — at The Deep View we use Ashby and love it).

Their interface is intuitive, and applicants are easily organized, making reviews quicker, more targeted, and as painless as possible.

With Ashby, our scheduling is almost completely automated, and as it turns out, their new AI-assisted application review actually enables us to deliver a much more human experience. Having all our workflows in a single tool helps us deliver a better candidate experience with better data and more accuracy.

Plus, all of Ashby’s AI-backed features are reliable, secure and trustworthy (it checks all the boxes).

If you’re hiring — but want to handle that process without your usual migraines — switch to Ashby.

• Building AI systems? Whether you're a beginner or looking to build confidence in launching to production, Vellum’s new webinar series is for you. Sign up today, space is limited.*

• Tired of receiving compliance notices from state tax agencies? Automate payroll, compliance, and benefits management with Warp. Get started today and get a $1,000 gift card.*

• The Air Force is researching an AI to help drones adapt to their environments (404 Media).

• Data center emissions probably 662% higher than big tech claims (The Guardian).

• TikTok faces tough questions from court over challenge to US law (Reuters).

• How Google got away with charging publishers more than anyone else (The Verge).

• Biden administration awards Intel up to $3 billion under the Chips Act (CNBC).

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.

… Speaking of robots, Chipotle is trying some out

Chipotle on Monday announced the integration of self-described “cobots,” which the fast food chain describes as “collaborative robots,” in two California-based locations. 

The details: One store is getting an “Autocado,” a machine that cuts and scoops avocados. The other is getting an “augmented makeline,” which automates part of the bowl (and salad) production process. 

• Chipotle said that the Autocado takes 26 seconds to fully remove the flesh from a given avocado. (Chipotle uses 129.5 million pounds of avocados every year).  

• The makeline automatically dispenses rice, corn, lettuce and other ingredients into a bowl beneath the counter, which employees can then customize. 

Chipotle said it had invested in Vebu (the company behind the Autocado) and Hyphen (the company behind the makeline) through its $100 million Cultivate Next venture fund. 

The context: Fast food chains have been trying to integrate generative AI and robotics for a little while now. McDonald’s recently scrapped its AI drive-thru ordering system after it kept screwing up; at around the same time, Taco Bell said it plans to expand AI drive-thru ordering to hundreds of locations by the end of the year.

Such automation also highlights concerns of job loss. Chipotle’s cobotic testing ground is in California, where fast food chains are required to pay workers a minimum of $20 an hour. 

I asked Chipotle if it plans to use the new cobots to hire fewer workers, or otherwise trim its staff. A company spokesperson told me that “Autocado and the Augmented Makeline will not eliminate any jobs. Crew members will have a cobotic relationship with the devices in our restaurants.”

Microsoft makes Bill Gates’ AI dreams come true 

Almost exactly a year ago, Microsoft launched its first version of Copilot, a genAI-based Windows feature that the company pitched as a “digital companion for your whole life.” 

Yesterday, Microsoft unveiled a second wave of Copilot. This time, the spotlight is on something called Copilot “agents.” Microsoft said that customers will be able to custom-build AI agents that can then carry out tasks across Microsoft’s suite of apps as well as third-party websites. 

Microsoft said the agents can range from “simple, prompt-and-response agents to agents that replace repetitive tasks to more advanced, fully autonomous agents.”

• There’s been a lot of talk of AI agents, lately. Salesforce recently unveiled “Agentforce,” which offers basically the exact same thing.

• Google is working on its own AI agents, and then, of course, there’s Apple and Amazon, which are working on overhauling Siri and Alexa, respectively, with generative AI.

The idea of a legitimately reliable (and trustworthy) agent has had Microsoft co-founder Bill Gates excited for a while. In early 2023, Gates said that the first company to develop a good digital assistant would win the AI race, since an agent would negate the need for manual online searching and shopping.

The other updates: In addition to the agents, Microsoft announced Copilot Pages, intended to serve as a hub of customer-specific data that Copilot can generate content from. 

• Microsoft said Pages will be collaborative across teams, enabling “human-to-AI-to-human collaboration.” 

• Microsoft is also shipping improvements to Copilot in Excel, PowerPoint, Teams, Outlook, Word and OneDrive. The big Excel improvement involves a new ability for Copilot to generate code in Python directly in Excel. 

Microsoft said that its Copilot customer base has grown some 60% quarter-over-quarter. Copilot costs $30 per user per month in addition to the cost of an existing Microsoft 365 plan, which costs between $6 and $22 per user per month. 

The Bubble: Microsoft — and the industry on the whole — need this to work. The problem is, so far, things aren’t going as smoothly as the Big Tech players would like you to believe.

There are a few issues inherent to the large language models (LLMs) that power Microsoft’s generative AI technology, including Copilot. A major one involves hallucination, genAI’s propensity to generate confidently incorrect output; another involves explainability. 

• Current LLMs have been described as black boxes; we don’t know exactly how they work, or why they generate a certain output in response to a certain query. 

• This lack of explainability raises issues of trust, reliability, discrimination, fairness and over-reliance that, in the enterprise (and everywhere else, frankly) are cause for concern. 

Despite the rosy picture of time-savings and productivity enhancements that Microsoft painted in its post about the new wave of Copilot, The Information recently reported that some Microsoft Copilot customers are pausing their rollout of the feature due to annoying buys, high costs and unclear value. 

Right now, the AI ecosystem is sustaining itself in a strange, almost closed-loop cycle; Nvidia sells super expensive AI chips, Big Tech buys billions of dollars worth of those chips to build AI, and everyone’s stock prices soar. 

But if the companies that are using those Nvidia chips can’t generate legitimate returns, they may have to stop buying new chips. If that happens, it could signal the bursting of the AI Bubble, and everyone’s share prices will likely plummet.

This, in essence, is the $600 billion gap in the industry — $600 billion in spend, a few billion in return.

If reliability cannot be guaranteed, I’m not sure if this new round of agents will be enough to crack the enterprise at a scale that would justify the ongoing size of the investment. 

But I suppose we’ll see what happens as it rolls out.

🤔 Your thought process:

Due to technical issues with our CMS, I have no idea how you fared on this one. The voting numbers are obscured. But, if you selected Image 2 (Left), congrats, you won! If you selected Image 1 … swing and a miss.

Selected Image 1 (Right):

• “Oh man! That was such a good, natural fence compared to the stamped out thing in #2. I think your AI has started living in the woods.”

If my AI ‘lived’ anywhere, it would absolutely be the woods. Have I mentioned how much I like trees?

Selected Image 2 (Left):

• “I could only verify the shadows cast on this image, and I didn’t think AI would get the telephone/power line right.”

💭 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 AI governance:

Again, I have no idea how you voted on this one. All I have are your thoughts, which I always appreciate.

Something else:

• “Registration at least may be a good place to start with an assessment of regulatory need before wholesale jump to regulation.”

Developers should NOT be required to share models with governments:

• “No, but that said, they should make their products identifiable as well as be held responsible for any harm they might be held responsible for.”