Good morning. Meta, Microsoft, Tesla and IBM are all reporting earnings later today (and Apple will report on Thursday).

It’s a massive Magnificant 7 lineup that very shortly follows Monday’s enormous tech sell-off. The quality of earnings reported by each of these giants (IBM isn’t part of the Mag7, but it is certainly an AI-related stock) as well as the market’s response to their earnings will be quite telling.

We’ll be watching it closely. In the meantime, OpenAI is trying to get itself in the hands and computers of government employees.

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

In today’s newsletter:

• 🧠 AI for Good: AlphaFold 3, but better

• 📊 The aftermath of the DeepSeek rout 

• 🏛️ OpenAI launches ChatGPT for the U.S. government

AI for Good: AlphaFold 3, but better

Google DeepMind’s release of AlphaFold 2 in 2020 marked a pivotal moment in the fields of computer science and molecular biology alike. The model, which DeepMind open-sourced, is highly capable of predicting protein structures, something that can accelerate all sorts of research, including the kind that’s relevant to drug creation. 

Following the success of AlphaFold 2 — which earned two DeepMind researchers a Nobel Prize in chemistry — there was a lot of excitement for AlphaFold 3, a model released last year that promises to predict the structures of proteins as they interact with other molecules. Beyond the fact that AlphaFold 3, due to a lack of relevant data, isn’t as impactful as AlphaFold 2, the model wasn’t released the same way AlphaFold 2 was; it’s available exclusively to academic researchers. 

What happened: So a team of MIT researchers spent four months building their own version of AlphaFold 3, called Boltz-1, a model that was intended from the very beginning to be openly released to the public. 

• Though they followed a similar approach initially, the team developed and incorporated new algorithmic approaches that, they say, boost the model’s predictive accuracy. 

• In testing the completed model, the researchers found that Boltz-1 performs at parity with AlphaFold 3 when it comes to predicting complex protein structures. 

“We think there is still many, many years of work to improve these models. We are very eager to collaborate with others and see what the community does with this tool,” one of the researchers said. 

GAIN 1 HOUR EVERY DAY WITH AN AI ASSISTANT

Fyxer AI organizes your inbox, drafts extraordinary emails in your tone of voice, and writes better-than-human meeting notes.

Start your 7-day free trial — no credit card required, and set up in just 30 seconds.

Designed for teams using Gmail, Outlook, Slack, Teams, Google Meet, or Zoom.

The aftermath of the DeepSeek rout 

Hours after Monday’s closing bell, OpenAI Chief Sam Altman finally chimed in on DeepSeek, echoing Nvidia’s statement in conceding that R1 is “an impressive model, particularly around what they're able to deliver for the price.”

Altman assured people that OpenAI “will obviously deliver much better models” — though timelines remain unclear — adding that it’s “legit invigorating to have a new competitor!” 

• “But mostly we are excited to continue to execute on our research roadmap and believe more compute is more important now than ever before to succeed at our mission,” Altman added, an unsurprising take from someone who has just kicked off a $500 billion commitment to building out new compute. 

• The world, according to Altman, is “going to want to use a LOT of AI,” so all that compute will still pay off.

DeepWater’s Gene Munster thinks the market overreacted to DeepSeek, saying that “the demand for the best AI hardware will persist. Hyperscalers, enterprises and sovereign entities are not looking for a cheaper way to achieve AGI. They’re looking for a faster way to get there. Deepseek doesn’t alter that dynamic.” 

The markets recovered a bit Tuesday, following Monday’s massive sell-off. Nvidia finished the day up around 9%, regaining roughly half the ground it lost the day before. Oracle and Broadcom rose 3.6% and 2.6%, respectively; TSMC rose around 5% and Microsoft lifted by about 3%.

The Nasdaq and S&P both ended higher on Tuesday, closing up 2% and .92% respectively.

Morgan Stanley reduced its Nvidia price target to $152 from $166.

The rout was significant enough to grab President Donald Trump’s attention; in a speech Monday, he referred to DeepSeek as a “wake-up call … we need to be laser-focused to win.”  He said it could be a positive development, enabling better AI for less money. 

As I said yesterday, there are still a ton of unknowns related to DeepSeek. We don’t know the full cost of operation — though the massively cheaper API prices indicate that inference isn’t as big of a deal as it is for OpenAI — we don’t know the size of their chip clusters, or what it cost to attain them, or the cost, duration (and training data) relevant to the research and development process.

What does seem clear is that the DeepSeek team has identified a cheaper, more efficient way to train capable language models, something that promises to improve the technical science of generative AI while threatening a rather tenuous business that has thus far been built on a massive circular economy infused with hype designed both to secure investments and to pull consumers in. 

Need a List of Companies to Target?

The PLAY is LeadTalk. It’s KEY that you don’t overpay for list building.

Leadtalk uses AI to analyze your best customers and instantly gives you a list of target accounts based on your ICP.

✔️ No complicated setup

✔️ No credit card required

✔️ Just smarter, faster, account targeting

Never overpay for list building again. Get your first 25 accounts free today.

• In a Monday night speech, Trump said that tariffs on foreign computer chips, semiconductors and pharmaceuticals are coming “in the near future.”

• Cambridge researchers argued in a recent paper that the current rise in generative AI ‘agents’ and chatbots could bring us to the edge of a dangerous future, in which these systems — fed on a diet of intimate psychological data — become leveraged as a tool for covert behavioral influence.

• DeepSeek’s popular AI app is explicitly sending US user data to China (Wired).

• Boeing tries to stem the bleeding after disastrous year (Semafor).

• Filipino tech workers demand protections in AI bill (Rest of World).

• Alibaba’s Qwen team releases AI models that can control PCs and phones (TechCrunch).

• AI companion app Replika faces FTC complaint (Time).

OpenAI launches ChatGPT for the U.S. government 

In its most significant product launch in quite some time, OpenAI on Tuesday unveiled something called ChatGPT Gov, a ChatGPT solution built specifically for use by the U.S. government. 

The details: ChatGPT Gov seems to function relatively similarly to OpenAI’s enterprise plans, with the one major difference being that agencies can deploy ChatGPT in their own Microsoft Azure cloud. 

• This self-hosting of the chatbot, according to OpenAI, “enables agencies to more easily manage their own security, privacy and compliance requirements, such as stringent cybersecurity frameworks,” according to a blog post. 

• OpenAI also said that it believes ChatGPT Gov will “expedite internal authorization of OpenAI’s tools for the handling of non-public sensitive data.” The startup added that ChatGPT Gov will, like its other applications, be subject to OpenAI’s usage policies, which include rules related to legal compliance and rules designed to prevent the use of OpenAI’s systems from causing harm.

Though OpenAI wants ChatGPT Gov to work with sensitive data, it’s not there yet, and it’s unclear when it will be. CPO Kevin Weil told CNBC that ChatGPT Enterprise — which Gov is built on — is currently going through the Federal Risk and Authorization Management Program, and has not yet been cleared for use on non-public information. 

“I know President Trump is also looking at how we can potentially streamline that, because it’s one way of getting more modern software tooling into the government and helping the government run more efficiently,” Weil said.

OpenAI said that, since 2024, more than 90,000 government employees have used ChatGPT. ChatGPT Gov will reportedly be made available within a month.

This push for the government-specific adoption of ChatGPT comes shortly after OpenAI CEO Sam Altman joined Trump to announce a $500 billion, four-year investment in AI data centers. 

OpenAI did not, of course, mention the threat and prevalence of certain unsolved flaws of generative architecture, notably including algorithmic bias and hallucination or confabulation. Issues of harm resulting specifically from algorithmic discrimination have been a prominent — and mounting — point of concern for more than a decade. The problem is multi-fold; since generative AI models are unable to inform their users when they might have hallucinated, or served biased results, non-domain experts (and even domain experts) have a higher chance of placing too much trust in a model. 

In government operations, that’s kind of a big deal. 

It’s unclear both precisely how government agencies might use ChatGPT Gov, as well as how OpenAI might or might not process or otherwise store inputted data. 

OpenAI did not respond to a request for comment regarding any of the above points. 

It’s unclear how much this Gov plan would cost. 

The idea of uncritical adoption seems to be something of a theme lately. Despite OpenAI’s security assurances, I remain uncomfortable with the idea of governments inputting sensitive information into ChatGPT. 

Additionally, I remain uncomfortable with the idea of elements of government operations being automated by ChatGPT, especially without regulations focused on civil rights and ethical use in place.

I worry about over-reliance. 

I don’t imagine that worry will ever go away; OpenAI and other labs seem supremely disinterested in reminding users that generative AI has the capacity to hallucinate and output biased information, an unsurprising reality, since it would likely hurt adoption. 

There’s also an interestingly unsettling element to this, of the potential start of a government becoming reliant on OpenAI, and so subject to OpenAI’s actions, policies and prices, something that feels especially conspicuous given the dethroning threat posed by DeepSeek. 

Whatever it is, this certainly marks a deepening relationship between the government and OpenAI. I don’t know that that’s such a good thing.  

🤔 Your thought process:

Selected Image 2 (Left):

• “Seems to be a clear detailed picture compared to image 1.”

Selected Image 1 (Right):

• “Shrubbery in foreground appears real and emulates nature.”

💭 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 Nvidia:

A whole 33% of you are buying the dip on Nvidia. But 22% think the downward spiral has only just begun.

And 20% think that tech is just so volatile.

Something else:

• “Nvidia is much more than just a chip maker. They have made significant investments in all aspects of the AI industry. So yes I am in, but with a small bet, I am not all in.”

The downward spiral has only just begun:

• “I guess our Tech industry never learn that China is always going to beat the US at making things better and cheaper. And that goes for the Electric Cars too.”