Hear why BILL S&E is #1 + earn AirPods 4

BILL is the all-in-one financial platform that simplifies managing bills, invoices, expenses, budgets, and business credit.

As a controller, you need software that works as hard as you do—that’s why BILL Spend & Expense is ranked #1 on G2’s 2025 Best Accounting & Finance Products list. New features include:

• Assistant permissions for seamless delegation

• Auto-freezing cards

• Visa Signature benefit

• API integrations for smarter scaling

Ready to see it in action and claim your AirPods 4?*

Claim Your Airpods

_{*Terms and conditions apply. See offer page for details. The BILL Divvy Card is issued by Cross River Bank, Member FDIC, and is not a deposit product.}

Today:

• Wes Roth, Dylan Debate Superintelligence

• Apple Partners With Anthropic For Xcode

• Anthropic Announces $61.5B Share Buyback

• Google Bypasses AI Training Opt-Out

• Gemini 2.5 Pro Beats Pokémon

Wes Roth gets CONFRONTED by Dylan Curious about AI....

Wes Roth and Dylan Curious recall the moments that convinced them modern AI is learning. OpenAI's reward‑based hide‑and‑seek game showed machines teaching themselves tricks. ChatGPT and GPT‑4 then proved one system can write, reason, and adapt like a new mind. 

They warn that such super intelligence, far smarter than people, could erase jobs, shape laws, or be misused by tyrants. Humans must find purpose beyond work and guard freedom while guiding the technology.

WATCH THE VIDEO ON YOUTUBE

Apple Partners With Anthropic For Xcode

Apple is teaming up with Anthropic to create an AI coding assistant for Xcode using the Claude Sonnet model. The tool can write, edit, and test code via a chat interface and is being tested internally. Apple hasn’t confirmed a public release. This move could help Apple compete in AI, especially after delays with Siri upgrades. It follows broader industry trends, with companies like Microsoft and OpenAI investing heavily in AI-powered coding tools.

Why This Matters

1. Big Tech Momentum Shift: Apple, typically a latecomer in AI, is making strategic moves to catch up by partnering with top players like Anthropic.
   

2. AI-Coding Race Intensifies: The tool highlights growing competition in AI-assisted software development, where Microsoft and OpenAI already lead.
   

3. Claude’s Enterprise Use Expands: Anthropic’s Claude model is now being embedded into real developer workflows, signaling rising trust and adoption of frontier models in coding tasks.

Anthropic Announces $61.5B Share Buyback

Anthropic is buying back shares from employees and ex-employees, valuing the company at $61.5 billion. Eligible workers can sell up to 20% of their equity, capped at $2 million each. This move rewards talent and helps retain top researchers amid fierce AI competition. Though still unprofitable, Anthropic recently raised $3.5 billion and now brings in over $1.4 billion in annualized revenue, underscoring its rapid growth in the AI industry.

Why This Matters

1. Talent Retention in a Competitive Market:
   Anthropic's buyback incentivizes employee loyalty and helps retain elite AI researchers in a talent-constrained industry.
   

2. Signals Confidence and Stability:
   A $61.5B valuation backed by fresh capital and revenue growth shows investor confidence in Anthropic’s trajectory despite its high burn rate.
   

3. Shifts in Startup Culture:
   By allowing ex-employees to participate, Anthropic breaks from norms like those at Scale AI, reflecting evolving strategies in how top AI firms handle compensation and equity liquidity.

Google Bypasses AI Training Opt-Out

Google admitted in court that it can still train its search-related AI tools—like AI Overviews—on web content, even if publishers opt out of AI training. The opt-out controls only apply to DeepMind, Google’s AI lab, not to the teams working on Search. This means web content may still be used to power Google’s products, raising concerns about consent, data use, and publisher rights in the AI era.

Why This Matters

1. Data Consent Loophole:
   It reveals a critical gap in AI training consent, where opting out doesn’t stop all parts of a company from using data.
   

2. Publisher Trust at Risk:
   This undermines trust between content creators and platforms, potentially fueling lawsuits or regulatory pushback.
   

3. Power Concentration Debate:
   Highlights how tech giants can leverage internal structure to maintain dominance in AI development—even when parts of the system are opt-out compliant.

🧠RESEARCH

DeepCritic: Deliberate Critique with Large Language Models

DeepCritic introduces a two-stage method to improve how LLMs critique math solutions step by step. Using long, detailed feedback as training data and refining with reinforcement learning, it builds a critic model that spots errors more accurately and gives better feedback. It outperforms GPT-4o and other models on math critique benchmarks.

T2I-R1: Reinforcing Image Generation with Collaborative Semantic-level and Token-level CoT

T2I-R1 is a new text-to-image model that boosts image generation by adding reasoning at two levels: planning the prompt (semantic-level) and guiding pixel creation (token-level). Using reinforcement learning and a coordinated reward system, it improves results over strong baselines—outperforming the current top model FLUX.1 on key benchmarks by up to 19%.

KeySync: A Robust Approach for Leakage-free Lip Synchronization in High Resolution

KeySync is a new system for lip-syncing video to audio with high accuracy and realism. It tackles common issues like facial expression leakage and occlusions using a smart masking strategy. With strong results on visual quality and a new "LipLeak" metric, KeySync sets a new standard for dubbing and talking head generation.

🛠️TOP TOOLS

Adobe Audio Enhancer - Clean up audio for free with Adobe Podcast AI

Mangoai.co - Create AI Videos in Minutes with AI avatars and voices.

AKOOL – AI Face Swap App - Premium Quality AI Face Swap Tool

Let’s Enhance! - Get clear, high-res images with AI

LeiaPix - The AI platform converting images & videos into 3D

📲SOCIAL MEDIA

🗞️MORE NEWS

• Google’s Gemini 2.5 Pro completed Pokémon Blue with help from a developer using custom tools. Though not a fair benchmark, it shows how advanced AI models are learning to handle games with reasoning, memory, and guidance.

• China is using DeepSeek AI to help design next-generation warplanes. Lead engineer Wang Yongqing says the tech improves research efficiency and inspires new ideas, supporting advanced fighter projects like the J-35 stealth jet.

• Google will soon allow kids under 13 to use its Gemini AI through supervised accounts. While promoted as a homework and storytelling tool, Google warns it can make mistakes and show inappropriate content, urging parents to monitor use and set boundaries.

• A new study finds that Transformer AI models, like GPT and ViT, process information in ways that closely mirror human thinking. By analyzing how predictions change across layers, researchers found strong parallels in tasks involving memory, logic, and perception—suggesting AI may not just copy human answers but echo human reasoning patterns.