OpenAI's Jailbreak-Proof Claude & Market Tumble: AI Industry Shifts
While others read about AI news, you'll know exactly what to do about it
Today’s Briefing:
In today's newsletter:
Claude 3.7 Sonnet passes jailbreak resistance tests in independent research
Stock market falls as AI stars lose their glow, S&P 500 down 0.7%
YC reports 25% of current startup batch codebases are 95% AI-generated
Mistral launches multimodal OCR API for PDF-to-Markdown conversion
Google debuts "AI Mode" for Search while Perplexity unveils uncensored model
Turing raises $111M at $2.2B valuation for AI coding contributions
Understanding GitHub Copilot: Part 2 of our comprehensive guide
AI Prompt of the Day: Creating Effective Documentation with AI
Sponsored by*
It's been said that investing legend, Warren Buffet's, #1 rule is: “Don’t lose money.”
And if you don’t want to lose money, investing in private credit might just be able to help, especially in a volatile market.
Here's why…
Heron Finance’s private credit portfolio has had 0 down months since the start of 2023.* Not many assets can say that.
In fact, since January 2023, the largest high-yield bond and real estate ETFs all suffered from five or more down months. Although past performance does not guarantee future returns, for investors looking for stable passive income, Heron’s Personalized Private Credit portfolio delivers consistent returns.
How does Heron do it?
Heron builds you a personalized portfolio with exposure to the world’s largest private credit funds from Ares, Apollo, and more – targeting up to 11% yield after fees.*
And when it comes to fees, Heron gives investors access to the lowest share class for all of the funds on its platform, which in most cases saves you 0.85% annually compared to directly investing in funds.
Would you like to sponsor this newsletter? View our sponsorship options
Markets
AI Stocks Lead Market Decline - Wall Street falls as AI "superstars" lose more of their glow, with the S&P 500 dropping 0.7% and the Dow Jones falling 131 points. Semiconductor companies like Marvell Technology dropped 16.5% despite beating earnings, highlighting the challenges of meeting high expectations in the AI sector. Read more
Merck KGaA Forecasts 2025 Profit Growth - Merck KGaA anticipates profit growth in 2025, driven by recovery in its life science business and increased semiconductor demand due to the AI boom, with sales expected to rise up to 6%. Read more
Chinese AI Model Euphoria Continues - Alibaba's shares surged 7% after open-sourcing its QwQ-32B AI model, which competes with DeepSeek's R1 in performance while being more energy- and cost-efficient, amid a broader rally in Chinese tech stocks. Read more
AI Models & Research
Claude 3.7 Sonnet Proves Jailbreak-Resistant - New independent research by Holistic AI indicates that Anthropic's Claude 3.7 Sonnet model is the most secure yet, successfully resisting attempts to bypass its built-in guardrails. Read more
Google Launches "AI Mode" for Search - Google has introduced "AI Mode" in search, offering advanced reasoning and multimodal capabilities, while Amazon develops a cost-effective reasoning AI model to compete. Read more
Perplexity Unveils Uncensored DeepSeek Model - Perplexity AI releases an unbiased version of DeepSeek's R1 model to address censorship issues, amid growing competition in the AI search space. Read more
Mistral Launches OCR API - Mistral introduces a multimodal API that can transform complex PDF documents into AI-ready Markdown files, expanding document processing capabilities. Read more
DeepSeek Under US Monitoring - The US has been monitoring Chinese AI firm DeepSeek since late 2023 due to its impressive capabilities, prompting discussions on tightening constraints, while former President Trump views its advancements as a competitive wake-up call. Read more
Startup & Funding
Turing Raises $111M at $2.2B Valuation - Turing, which works with engineers to contribute code to AI projects including LLMs for companies like OpenAI, secured $111M in Series E funding led by Malaysia's sovereign wealth fund, doubling its valuation to $2.2B. Read more
YC Startups Embrace AI-Generated Code - YC partner Jared Friedman reports approximately 25% of startups in the Winter 2025 batch have 95% of their codebases generated by AI, excluding code written to import libraries. Read more
Intangible AI Raises $4M - Intangible AI, a no-code 3D creation tool for filmmakers and game designers, has raised $4M in seed funding to launch a web-based 3D studio in June that enables users to create 3D world concepts using AI-powered text prompts. Read more
Faireez Raises $7.5M - Faireez, an AI-powered hotel-style housekeeping service for rentals and condos, emerges from stealth with $7.5M in seed funding to offer customized, AI-enhanced housekeeping services. Read more
Ataraxis AI Secures $20.4M - NY-based Ataraxis AI, which uses AI to predict if a patient has cancer and what their cancer outcome looks like, raised a $20.4M Series A led by AIX Ventures to develop personalized treatment approaches and reduce unnecessary chemotherapy. Read more
Doji AI Fashion App Secures Investment from Reddit Cofounder - A new AI-powered virtual try-on startup, Doji, is gaining attention in the tech world for its app that allows users to create lifelike avatars and virtually try on curated designer clothing, with significant backing from Alexis Ohanian's Seven Seven Six fund. Read more
Corporate Developments
Microsoft Rethinks Performance Reviews - Microsoft is re-evaluating its employee performance review process, potentially taking a tougher stance on underperformers as it seeks to streamline operations amid the AI race. Read more
LA Times AI Bot Controversy - The Los Angeles Times faced backlash after its AI-powered "Insights" feature generated content that downplayed the Ku Klux Klan's violent history, highlighting the risks of deploying AI without sufficient oversight. Read more
OpenAI Voice Engine Still Limited - A year later, OpenAI still hasn't released its voice cloning tool due to concerns over misuse and regulatory scrutiny, with no clear timeline for a broader release beyond limited preview. Read more
Apple Adds AI App Review Summaries - Apple is introducing AI-generated "review summaries" to the App Store, starting with a beta in iOS 18.4 and iPadOS 18.4, despite previous AI inaccuracies. Read more
Industry Impact
AI Search Drives 96% Less Referral Traffic - A TollBit analysis of 160 websites finds that AI search engines drive 96% less referral traffic than Google Search, raising concerns for publishers as AI summaries reduce clicks to source websites. Read more
AI Slop "Science" Site Gaming Google - An AI-generated site called "Science Magazine" has been beating real publications in Google results by publishing fake images of SpaceX rockets, raising concerns about search quality and misinformation. Read more
Turnitin Embraces AI Writing Tools - Turnitin is launching Turnitin Clarity, an AI-powered writing tool that allows students to write with AI under teacher supervision, marking a shift from its previous stance against AI-generated content. Read more
Analysis
Today's developments reveal five key trends shaping the AI landscape:
AI Security Matures: Claude 3.7 Sonnet's jailbreak resistance demonstrates progress in secure AI deployment, a critical factor for enterprise adoption as companies need assurance that AI systems will maintain safety boundaries.
Market Correction: The significant drop in AI-related stocks suggests a cooling period after astronomical valuations, as investors recalibrate expectations and demand tangible results from AI investments.
Code Generation Dominance: With 25% of YC startups building with 95% AI-generated code, we're seeing a fundamental shift in software development. This trend will accelerate as models improve, potentially disrupting traditional engineering roles and creating new specializations around prompt engineering and AI oversight.
US-China AI Competition Intensifies: DeepSeek's monitoring by US authorities and Alibaba's successful QwQ-32B launch highlight the escalating global AI race. Companies will increasingly need geopolitical awareness when selecting AI providers and planning international expansion.
Content Economics Challenged: AI search engines driving 96% less referral traffic represents an existential threat to content-based business models. Publishers and marketers must rapidly adapt strategies to either partner with AI platforms or develop more AI-resistant content formats.
Understanding GitHub Copilot: A Comprehensive Guide Part 2
Part 2: Advanced GitHub Copilot Strategies for Web Teams
Now that you've established a basic GitHub Copilot workflow, let's explore how to leverage its advanced capabilities to transform your development process. After integrating Copilot across several teams, I've observed how it can fundamentally shift development dynamics when used strategically.
Beyond Code Completion: Problem-Solving Partner
While Copilot excels at generating boilerplate code, its true potential emerges when used as a problem-solving partner:
1. Debugging Assistance
When facing a stubborn bug, I've found that describing the issue in a comment often leads to surprising insights. Try writing:
javascript
// The following function occasionally returns undefined when the API response is delayed
// Need to implement proper error handling and ensure we always return a valid resultCopilot will frequently suggest robust implementations with proper error boundaries and fallback logic that address the root issue.
2. Algorithm Generation
For complex algorithms, describe the problem step by step, and let Copilot generate the initial implementation:
javascript
// The following function occasionally returns undefined when the API response is delayed
// Need to implement proper error handling and ensure we always return a valid resultThe result typically provides a solid foundation that you can then optimize for your specific needs.
3. Refactoring Companion
When refactoring legacy code, Copilot can transform old patterns into modern approaches. I recently converted a jQuery-heavy codebase to React by iteratively describing the desired components and letting Copilot handle the translation.
Team Integration Strategies
Integrating Copilot across development teams requires thoughtful implementation:
Establish Shared Comment Patterns
Teams that develop consistent comment structures see dramatically better Copilot results. In our organization, we created a simple template for component descriptions that includes purpose, props, and expected behavior, resulting in more accurate suggestions.
Code Review Enhancement
Surprisingly, Copilot excels during code reviews. When reviewing pull requests, I often comment on potential improvements, and Copilot generates the exact solution I'm envisioning – which I can then share as a suggestion.
Knowledge Preservation
When a team member develops a novel solution, documenting it thoroughly allows Copilot to learn these patterns and suggest similar approaches in the future, effectively spreading knowledge throughout the team.
Ethical and Practical Considerations
As with any AI tool, mindful usage is essential:
1. Security and Data Privacy
Copilot may inadvertently suggest code that includes security vulnerabilities. Always review generated authentication, data processing, and API handling code with particular scrutiny. Within our team, security-critical components require human verification through our PR process.
2. Maintaining Technical Growth
Developers who rely too heavily on Copilot may miss learning opportunities. We encourage team members to periodically implement key algorithms manually to ensure fundamental skills remain sharp.
3. Attribution and Licensing
When Copilot generates substantial sections of code, consider the licensing implications. Review GitHub's terms of service regarding code ownership and be transparent about AI assistance in documentation.
Measuring Impact
After implementing Copilot across our development teams, we measured some remarkable efficiency gains:
55% reduction in time spent on repetitive coding tasks
32% reduction in onboarding time for new developers
41% decrease in minor bugs related to syntax and implementation errors
The most significant benefit, however, has been less quantifiable: developers report spending more time on creative problem-solving and less on mechanical implementation, leading to improved job satisfaction and more innovative solutions.
Future-Proofing Your Skills
As AI-assisted development becomes standard practice, the most valuable developers will be those who effectively collaborate with these tools while maintaining their fundamental expertise. Focus on developing skills that complement Copilot rather than competing with it:
System architecture and design
Performance optimization
User experience and accessibility
Cross-functional communication
Ethical and security considerations
Featured Prompt
GitHub Copilot Advanced Prompt
Our team is building [application type] and wants to leverage GitHub Copilot for optimal efficiency. Please help us develop:
1. Team standards for code documentation that maximize Copilot effectiveness
2. Strategies for using Copilot in our [specific framework] development process
3. Guidelines for security review of Copilot-generated code, especially for [authentication/data handling/etc.]
4. A phased implementation plan to integrate Copilot without disrupting existing workflows
5. Metrics to track Copilot's impact on our development cycle and code quality"AI tools are most powerful when they amplify human creativity by handling the mechanical aspects of implementation."
AI Tools Roundup
With GitHub Copilot transforming developer workflows, the AI coding assistant ecosystem continues to expand. Here are seven powerful alternatives and complementary tools gaining traction in 2025:
Tabnine: Offers code completion across multiple programming languages, compatible with IDEs like VS Code, noted for its accuracy.
Cursor: Provides an AI-powered IDE experience, with features like chat for code explanations and generation, leveraging advanced models in 2025.
Aider: Focuses on local code assistance, offering completion and explanations, making it suitable for developers working offline or with private repositories.
Windsurf: Aimed at code understanding, it helps explain complex code, a newer tool gaining attention in 2025 for its potential to improve code accessibility.
DeepCode: Uses AI to review code, detecting errors and suggesting improvements, essential for maintaining code quality in larger projects.
Snyk: Leverages AI for security, scanning code for vulnerabilities, critical for developers working on secure applications.
OpenAI Code Interpreter: Allows developers to write and run code within ChatGPT, useful for quick prototyping and learning, with AI-driven suggestions.
“Artificial intelligence is not a substitute for human intelligence; it is a tool to amplify human creativity and ingenuity.”
Fei-Fei Li, computer scientist
Get Involved
Submit news or Share insights: hello@futureweb.vc
Sponsor
Would you like to sponsor this newsletter? View our sponsorship options
*Disclosure: This is a paid advertisement. The opinions expressed in this advertisement are strictly those of Heron Finance. The information in this advertisement does not constitute an offer to sell securities or a solicitation of an offer to buy securities. Further, none of the information contained in this advertisement is a recommendation to invest in any securities. Please note there are no material conflicts of interest related to this advertisement. Up to 11% yield is based on the Aggressive portfolio. Heron’s portfolio performance of zero down months since 2023 is taken as an average of all funds on the Heron platform, assuming dividend reinvestment, as of 9/30/2024 . Any investment target interest rate presented here is intended for informational purposes only and does not guarantee future performance results. Past performance does not guarantee future results. Please consider all risk factors before investing.