GPT-5.2 Derives New Theoretical Physics Formula [2026]

by

3 Key Takeaways from GPT-5.2’s Derivation of New Formulas in Theoretical Physics

  • Proved that gluon scattering amplitudes, considered zero for decades, are non-zero under specific conditions.
  • GPT-5.2 conjectured the formulas and completed the proof after 12 hours of reasoning.
  • Joint pre-print published with researchers from IAS, Harvard, and Cambridge.

Textbooks Were Wrong — Gluon Amplitudes Are Not Zero

OpenAI has derived a new result in theoretical physics using GPT-5.2. The pre-print is titled “Single-minus gluon tree amplitudes are nonzero.”[OpenAI]

Gluons are particles that transmit the strong nuclear force. Existing textbooks stated that the tree-level amplitude must be 0 when only one gluon has negative helicity. This research shows that this conclusion does not hold in the half-collinear region.[Kevin Weil]

GPT-5.2’s 12-Hour Reasoning Process

GPT-5.2 Pro first conjectured the formulas by simplifying complex equations. Subsequently, a scaffolding version completed the formal proof after approximately 12 hours of reasoning.[OpenAI]

The formula was verified to satisfy the Berends-Giele recursion relations. Researchers from the Institute for Advanced Study at Princeton, Vanderbilt University, Cambridge University, and Harvard University participated in the research.[Investing.com]

Expanding from Gluons to Gravitons

This amplitude has already been extended to gravitons, and further generalizations are in progress. This is significant as an example of AI directly deriving new results in physics.

However, it is still in the pre-print stage, so peer review remains. Hope this is helpful.

Frequently Asked Questions (FAQ)

Q: What is a single-minus gluon tree amplitude?

A: It is the tree-level scattering amplitude of an arrangement where only one of several gluons has negative helicity. Tree-level refers to calculating only the most basic interactions without quantum corrections. For decades, this value was considered to be 0, but it has now been revealed that it is non-zero in the half-collinear region.

Q: Does GPT-5.2 understand physics?

A: Whether it understands like a human is debatable. In this case, GPT-5.2 discovered patterns in the equations, conjectured formulas, and generated a proof through 12 hours of reasoning. Physicists verified the results and published them as a pre-print. Its usefulness as a mathematical tool is clear.

Q: What is the real-world impact of this discovery?

A: It presents a new perspective on the theory of scattering amplitudes in particle physics. It shows that textbook assumptions do not hold under certain conditions, and extensions from gluons to gravitons are also in progress. In the long term, it can contribute to refining the theoretical predictions of particle collision experiments.

If you found this helpful, please subscribe to AI Digester.

References

3 Reasons Why Safety Talent is Leaving OpenAI and xAI [2026]

by

OpenAI·xAI Safety Talent Exodus: 3 Key Takeaways

  • xAI: 6 out of 12 co-founders have resigned
  • OpenAI: Mission Alignment team disbanded, policy executive fired
  • AI industry trust shaken as safety personnel depart

Half of xAI’s Founding Team Vanished

Half of Elon Musk’s xAI co-founding team has left. Six out of 12 members have resigned, and over 11 engineers declared their departure last week alone.[TechCrunch]

Recently, Jimmy Ba, who reported directly to Musk, and Tony Wu, in charge of multimodality, left the company. Musk stated it was a “reorganization to increase execution speed,” but many departures were voluntary.[Fortune]

OpenAI Disbands Safety Team, Fires Dissenters

OpenAI disbanded its Mission Alignment team, a team of 6-7 people responsible for communicating the company’s mission internally and externally. The team leader’s title was changed to ‘Chief Futurist,’ while the rest were scattered to other departments.[TechCrunch]

More controversially, Ryan Beiermeister, VP of Policy, was fired. He was reportedly terminated on discrimination charges after opposing ChatGPT’s ‘adult mode.’ He has vehemently denied the allegations.[TechCrunch]

AI Industry Prioritizes Speed Over Safety

At Anthropic, Mrinank Sharma, head of the safety research team, also resigned, stating that “the world is in danger.” It’s unusual for safety personnel to leave all three companies simultaneously.[CNN]

The pattern is consistent. As product release speeds increase, safety checks are sidelined, and those who object leave. The departure of safety personnel poses a significant long-term risk.

Frequently Asked Questions (FAQ)

Q: Why did xAI’s co-founders leave?

A: Officially, it’s a reorganization. However, pressure for product development, regulatory issues, and the Grok deepfake controversy are all contributing factors. It’s a mix of voluntary resignations and restructuring, with Musk explaining it’s to improve execution speed.

Q: What is OpenAI’s Adult Mode?

A: It’s a feature that allows adult content on ChatGPT. CEO of the app division, Fidji Simo, announced its planned release in Q1 2026. The policy VP’s dismissal after opposing it has sparked controversy, with critics arguing that profit is being prioritized over safety.

Q: Why is the departure of AI safety talent a problem?

A: Safety researchers and policy officers monitor AI to ensure it operates ethically. Their departure weakens internal checks and balances. As we approach AGI, the absence of safety personnel can lead to unpredictable risks.

If you found this helpful, please subscribe to AI Digester.

References

Claude Code v2.1.42 Update: 3 Changes [2026]

by

Claude Code v2.1.42, 3 Bugs Fixed

  • Fixed resume command session title bug
  • Fixed incorrect display of Opus 4.6 release notification
  • Improved image-related error messages

What Anthropic Fixed

Anthropic released Claude Code v2.1.42 on February 13th. This patch focuses on bug fixes without any new features.[GitHub]

The first issue was with the /resume command. When reopening a previous session, the interrupt message was incorrectly displayed as the session title. The second issue was that the Opus 4.6 release notification was also appearing for Bedrock, Vertex, and Foundry users.[GitHub]

Error Messages Are Now More User-Friendly

The third improvement is to the image dimension limit error message. If an error occurs when using multiple images, it now guides users to use the /compact command to reduce the context.[GitHub]

Claude Code is a CLI tool that allows you to use Claude from the terminal. It handles code editing, Git operations, and more using natural language. These small patches being released quickly are a sign of active maintenance.

Claude Code in the CLI AI Tool Competition

GitHub Copilot CLI, Cursor, Aider, and other terminal-based AI coding tools are competing. Claude Code’s strength is that it can directly utilize Anthropic’s latest models. Although it’s a small patch, it shows a direction of finely tuning the user experience.

Frequently Asked Questions (FAQ)

Q: Are there any new features in Claude Code v2.1.42?

A: No. This release focuses solely on bug fixes. It fixed three issues: the resume command session title bug, the incorrect display of the Opus 4.6 notification, and the image error message. New features are expected to be added in the next minor or major release.

Q: Can I use Opus 4.6 on Bedrock or Vertex?

A: Whether Opus 4.6 is currently supported on Bedrock, Vertex, and Foundry depends on the schedule of each cloud provider. This patch only fixed the issue where the notification was incorrectly displayed on those platforms. You should check the documentation for each platform for actual availability.

Q: How do I install Claude Code for the first time?

A: Install it globally with the command: npm install -g @anthropic-ai/claude-code. Node.js 18 or higher is required. After installation, you can run the claude command in the terminal. An Anthropic API key or a Claude Pro or higher subscription is required.

If you found this article helpful, please subscribe to AI Digester.

References

Half of xAI’s Co-founders Have Left — How Will Musk Respond? [2026]

by

Half of xAI’s Co-Founders Resign — Key Takeaways

  • 6 out of 12 co-founders have resigned
  • Musk has reorganized xAI into 4 divisions
  • This exodus comes ahead of a potential IPO after the SpaceX merger

Half the Founding Team Gone in Just One Year

Six out of xAI’s 12 co-founders have left. Jimmy Ba and Tony Wu announced their departures on consecutive days in the first week of February.[TechCrunch] Previously, Kyle Kosic went to OpenAI, and Szegedy and Babuschkin also departed.[CNBC]

Transition to a 4-Division System

Musk has divided xAI into Grok (chatbot), Coding, Imagine (video), and Macrohard (AI agent).[CNBC] He expressed confidence that Grok Code will be top-tier within 2-3 months.

Merger, IPO, and Internal Conflicts

xAI was merged into SpaceX and valued at $250 billion.[Fortune] The departure of key personnel ahead of a June IPO is a concern.

The CSAM controversy surrounding Grok, leading to regulatory investigations, is also a factor.[TechCrunch] While Musk called it an “evolution,” the fact that some of the departing members are starting a new company together suggests it’s more than just a simple replacement.

Frequently Asked Questions (FAQ)

Q: Who left xAI?

A: Jimmy Ba, Tony Wu, Kyle Kosic, Christian Szegedy, Igor Babuschkin, and Greg Yang. That’s half of the 12 founders. Reasons vary from new ventures to joining competitors and health issues, but the timing overlaps.

Q: What does Macrohard do?

A: It’s xAI’s AI agent division, responsible for enterprise automation, led by Toby Pohlen. The name seems to be a nod to Microsoft.

Q: What impact does the SpaceX merger have on xAI?

A: SpaceX’s funding could accelerate AI development. However, IPO pressure could prioritize commercialization, and there’s potential for culture clashes.

If you found this helpful, please subscribe to AI Digester.

References

AI CRM Monaco [2026] by Brex CRO Alum

by

Monaco, an AI CRM Challenging Salesforce, Emerges

  • Sam Blond, formerly CRO of Brex, unveils AI-native CRM ‘Monaco’
  • Stripe’s Collison brothers, YC’s Garry Tan, and others participate in investment
  • Strategy to replace legacy CRMs with an AI-native alternative

Sam Blond, Returns to Entrepreneurship After VC Stint

Sam Blond joined Brex as its 20th employee and, as CRO, grew revenue to hundreds of millions of dollars[TechCrunch]. He joined Founders Fund but left after just 18 months[TechCrunch]. Monaco, which he had been secretly developing, was unveiled on February 11th.

AI-Native Design Targeting Salesforce’s Legacy

Monaco is a CRM designed from the ground up with AI. While Salesforce adds AI to its existing system, Monaco has AI at its core. Salesforce’s FY2026 Q3 revenue grew 8.6% year-over-year, showing its strength, but its legacy complexity has been a long-standing complaint[CNBC].

Stripe co-founders, the Collison brothers, and YC CEO, Garry Tan, participated in the investment[TechCrunch].

AI CRM Market Competition Intensifies

Monaco isn’t alone. Clarify ($15 million) and Aurasell ($30 million) are also building AI-native CRMs[GeekWire]. However, Blond’s strength lies in his experience at Brex, where he generated 80% of revenue through outbound sales. He can design a CRM from a practical, field-oriented perspective, not just a technological one.

Frequently Asked Questions (FAQ)

Q: What is Monaco?

A: It’s an AI-native, all-in-one CRM founded by Sam Blond. Instead of adding AI to an existing CRM, it’s designed with AI at its core. It aims to be a Salesforce alternative and has received investment from the founders of Stripe and the CEO of YC.

Q: Who is Sam Blond?

A: He’s an early member and former CRO of Brex. After growing revenue to hundreds of millions of dollars, he became a partner at Founders Fund before founding Monaco. He’s a well-known figure in the Silicon Valley sales scene.

Q: How is an AI-native CRM different from a traditional CRM?

A: Traditional CRMs require a lot of manual input, and AI is added later. AI-native CRMs are designed with AI from the start, with AI handling data collection and analysis, reducing manual work and automatically providing insights.

If you found this helpful, please subscribe to AI Digester.

References

Monaco Challenges Salesforce with AI CRM [2026]

by

A VC from Founders Fund Launches an AI CRM

  • Sam Blond unveils Monaco, an AI-native CRM startup
  • Stripe founders, the Collison brothers, and YC CEO Garry Tan invested
  • It’s structured as a direct challenge to the Salesforce-centric CRM market

From Brex CRO to VC, and Back to Entrepreneurship

Monaco’s founder, Sam Blond, has a unique background. He worked as CRO at fintech company Brex before joining Founders Fund as a partner in 2023.[TechCrunch] However, he left the VC firm after only 18 months, deciding that full-time investing wasn’t the right fit for him.

Monaco is the result of about a year of stealth mode development. It’s an all-in-one CRM system designed with AI as a core architectural element, not just an add-on feature.[TechCrunch]

Why AI-Native CRMs Threaten Salesforce

Salesforce is the undisputed leader in the CRM market. But AI-native startups are rapidly gaining ground. Attio has raised over $100 million, and Reevo has secured $80 million.[Upstarts Media]

Traditional CRMs layer AI on top of an old database. AI-native CRMs, on the other hand, are built from the ground up with AI handling data organization and pipeline management. This structural difference is more significant than it seems.

What the Investor Lineup Tells Us

Monaco’s list of investors is noteworthy. Stripe co-founders, the Collison brothers, and YC CEO Garry Tan participated.[TechCrunch] Key figures in payment infrastructure and the startup ecosystem are betting on sales software.

The launch of enterprise AI agents by OpenAI and Anthropic, which is shaking up the stock prices of existing SaaS companies, is also a factor.[Fortune] This signals that the entire enterprise software market could be undergoing a major reshuffle.

Frequently Asked Questions (FAQ)

Q: What is Monaco?

A: Monaco is an AI-native CRM startup founded by Sam Blond, former partner at Founders Fund and former CRO at Brex. It aims to replace traditional CRMs like Salesforce with an all-in-one sales management platform designed with AI as its core architecture. The founders of Stripe and the CEO of YC have invested in the company.

Q: How is an AI-native CRM different from a traditional CRM?

A: Traditional CRMs add AI on top of an outdated database structure. AI-native CRMs are designed from the start to have AI automatically handle data organization, pipeline management, and sales action suggestions. This structural difference reduces manual input and increases sales efficiency.

Q: Is Salesforce’s market position really under threat?

A: It’s unlikely to be significantly shaken in the short term. However, Attio, Reevo, Day AI, and others are raising tens to hundreds of millions of dollars, and the launch of enterprise AI agents by OpenAI and Anthropic is putting pressure on the entire existing SaaS market to restructure.

If you found this helpful, please subscribe to AI Digester.

References

ChatGPT Gets Ads — From Adobe to Target [2026]

by

ChatGPT Ad Era: 3 Key Takeaways

  • OpenAI introduced ads to ChatGPT starting February 9th.
  • Major brands like Adobe, Audible, Ford, and Target are participating.
  • Ads are only shown to Free and Go plan users; paid subscribers are exempt.

OpenAI Finally Monetizes with Ads

OpenAI has started placing ads in ChatGPT. Sponsored ad testing began on February 9th for free and Go users in the US.[OpenAI]

Ads are displayed at the bottom of AI responses with a “Sponsored” label. Conversation content is not shared with advertisers.[TechCrunch]

Participating Brands and Agencies

Adobe, Audible, Ford, Mazda, and Target are participating in the pilot. Agencies Omnicom, WPP, and Dentsu have joined as well.[Adweek]

The minimum participation cost is $200,000 (approximately ₩280 million). The strategy is to validate the advertising value through small-scale testing.[Adweek]

No Ads for Paid Subscribers

Ads are only shown to adult Free and Go plan users in the US. Pro, Business, and Enterprise subscribers are not targeted.[OpenAI Help]

Contextually relevant ads based on keywords will appear in shopping-related conversations.[Target]

Will AI Chatbot Ads Become the Industry Standard?

Google has already included ads in AI Overview, and Perplexity is also testing sponsored questions. The atmosphere suggests that AI chatbot advertising is becoming an industry standard.

There are concerns that advertising could degrade the reliability of AI responses. The key is balancing user experience and revenue.

Frequently Asked Questions (FAQ)

Q: Are ChatGPT ads visible to all users?

A: No. They are only shown to Free and Go plan users in the US. Pro, Business, Enterprise, and Education subscribers will not see ads. OpenAI has stated that it will maintain an ad-free environment for paid subscribers.

Q: What is the cost to participate in advertising?

A: The minimum is $200,000 (approximately ₩280 million). Small-scale testing is underway to validate valuable advertising formats for users. Currently, over 30 clients are participating.

Q: Can advertisers see the conversation content?

A: No. OpenAI does not share conversation content with advertisers. Ads are displayed based on keywords, and personal data is not sold.

If you found this article helpful, please subscribe to AI Digester.

References

TechCrunch Disrupt 2026 Exhibition: 3 Things to Know

by

TechCrunch Disrupt 2026 Exhibition Participation: 3 Key Takeaways

  • October 13-15, held at Moscone West in San Francisco
  • Expected attendance of over 10,000 investors, founders, and tech leaders
  • Exhibition tables provide opportunities for pipeline building and deal closing

Where is Disrupt being held this year?

TechCrunch Disrupt 2026 will be held at Moscone West in San Francisco from October 13-15.[TechCrunch] Over 10,000 investors, founders, and tech leaders will gather.

You can directly introduce your company’s products with an exhibition table sponsorship.[TechCrunch] Pipeline building and deal closing are key.

Meaningful Exhibition Opportunity for Startups

It’s not just a booth. You can demonstrate your product in front of decision-makers. If you’re an AI startup, it can lead directly to investment meetings.

Exhibition tables are limited in number.[TechCrunch] If you’re interested, it’s a good idea to apply early.

Points to Consider for AI Startups

If you are a Korean AI startup considering entering the US market, it is worth considering. You can expect a different effect than online pitching.

You need to consider exhibition costs and travel expenses. It’s realistic to calculate the ROI and then decide.

Frequently Asked Questions (FAQ)

Q: When and where is TechCrunch Disrupt 2026 held?

A: It will be held at Moscone West in San Francisco from October 13-15, 2026. Over 10,000 tech industry professionals are expected to attend for three days. It’s a place to meet investors and founders.

Q: What is the cost to participate with an exhibition table?

A: You can check the specific prices on the official TechCrunch website. Prices vary depending on the sponsorship package. The quantity is limited, so it’s a good idea to inquire early as it may close early.

Q: What are the benefits for an AI startup to participate in Disrupt?

A: You can demonstrate your product directly in front of investors. There is a possibility of leading to investment meetings on-site, and you can also expect media exposure. There are also opportunities to discover partners through networking.

If this article was helpful, please subscribe to AI Digester.

Reference Materials

Meta’s $27 Billion Data Center: 3 Problems Revealed by Icefall [2026]

by

Meta Hyperion Data Center: 3 Key Issues

  • Winter storm blacks out 1 million homes, data centers activate 35GW backup
  • Louisiana campus expands by 3,650 acres, $27 billion investment
  • Concerns about residents bearing the cost of power plants

Power Struggle Emerges Amidst the Ice Storm

In January 2026, Winter Storm Fern slammed the southeastern United States, leaving over 1 million homes without power.[Fortune] The Department of Energy ordered data centers to activate their backup generators, a whopping 35 gigawatts worth.[TechXplore]

4 Times the Size of Central Park, $27 Billion

Meta is building a massive data center in Richland Parish, Louisiana. It spans 3,650 acres and represents a $27 billion investment.[Fortune] It will consume three times the annual electricity of New Orleans.[CBS News]

Controversy Over Shifting the Burden to Residents

The cost to build three new gas power plants for the data center is $3 billion. A key point of contention is Meta’s contract, which allows them to withdraw after just four years.[Earthjustice] If they leave, the cost of the 30-year power plants falls on 1.1 million subscribers. Construction site traffic accidents have also increased by 600%.[Louisiana Illuminator]

Transparent Contracts are the Solution

Google has implemented a demand response agreement, postponing non-urgent computations during peak load. Research also suggests that flexible power usage could secure 100 gigawatts without new power plants. Transparent contracts for cost sharing and emergency power management are urgently needed.

Frequently Asked Questions (FAQ)

Q: What is the size of Meta Hyperion?

A: It’s located in Richland Parish, Louisiana, and covers 3,650 acres – almost 4 times the size of Central Park. The investment is $27 billion, with 500 permanent employees and 5,000 construction workers.

Q: Could electricity rates increase?

A: Highly likely. The cost of the 3 power plants is $3 billion, and if Meta withdraws, the remaining cost will be passed on to 1.1 million subscribers. Earthjustice has requested an investigation.

Q: What about data center power during storms?

A: The Department of Energy ordered the activation of backup generators. Emergency generators can produce 35 gigawatts. Expanding demand response agreements is an alternative.

If you found this helpful, please subscribe to AI Digester.

References

Meta’s $27 Billion Data Center: 3 Problems Revealed by the Cold Wave

by

Meta’s $27 Billion Data Center: 3 Problems the Cold Snap Exposed

  • Louisiana cold snap leaves 130,000 homes without power, raising electricity concerns
  • Meta data center projected to consume 3 times the power of New Orleans
  • Residents express concerns about electricity rate hikes and power grid instability

The Data Center Power Controversy Triggered by the Cold Snap

Meta is building a $27 billion AI data center in northern Louisiana.[The Verge] The January cold snap left 130,000 homes without power. The data center isn’t directly related as it’s still under construction. However, residents are worried about what will happen when three times the power of New Orleans is added to an already fragile power grid.

$650 Million Savings: Really?

Entergy is building three new gas power plants. Meta claims it will cover the construction costs for 15 years and provide $650 million in savings.[The Verge] However, Logan Burke, director of AAE, points out that transmission line costs are excluded. Increased gas demand could also drive up prices during cold snaps.

Residents and Experts Demand Re-analysis

Donna Collins, a fifth-generation resident, says she “can’t imagine the changes that are coming.”[The Verge] The Union of Concerned Scientists has called for a re-analysis, arguing that Entergy’s power grid stability analysis is inadequate.[UCS] This aligns with the growing backlash against data centers across the United States.

Frequently Asked Questions (FAQ)

Q: What is the scale of Meta’s Louisiana data center?

A: It’s Meta’s largest project, costing $27 billion. It will consume three times the power of New Orleans and is scheduled for completion in 2030.

Q: How will the data center affect residents’ electricity bills?

A: Meta has promised $650 million in savings over 15 years. However, consumer groups believe that omitted transmission line costs and increased gas demand could lead to higher rates.

Q: Is the recent cold snap directly related to the data center?

A: Not yet. It’s still under construction and didn’t contribute to the recent power outages. However, the concern is that adding such a large demand to a fragile power grid could exacerbate problems during future disasters.

If you found this helpful, please subscribe to AI Digester.

References