Grok 4: xAI’s Powerhouse Arrives

xAI’s Grok 4, hailed by Elon Musk as the “smartest AI yet,” dropped with jaw-dropping stats: 87% on GPQA and 45% on Humanity’s Last Exam, outpacing rivals. Its real-time X integration and multimodal flair—crafting black hole visuals and a four-hour FPS game—make it a developer’s dream via its API. But controversy looms: bias concerns and Musk’s sway over its “truth-seeking” responses raise red flags, especially after antisemitic missteps.

Kimi K2: Coding’s New Champion

Kimi AI’s open-source K2, a 1T-parameter MoE titan, is shaking up the coding world. Outperforming competitors on programming benchmarks, its accessibility has devs buzzing on X. Designed for enterprise-grade tasks, K2’s efficiency and power signal a shift toward democratized, high-impact AI tools. Yet, its untested limits spark caution—robust testing will be key to its staying power.

AlphaCode 2.5: DeepMind’s Coding Leap

DeepMind’s AlphaCode 2.5 quietly stole the spotlight, tackling competitive programming at near-human levels. Building on its AlphaGo legacy, it hints at AI coders dominating future hackathons. Its reinforcement learning roots and Google’s ecosystem integration make it a game-changer, but opaque training data raises ethical questions.

This trio—Grok, Kimi, and AlphaCode—demands clear ROI and oversight to cut through hype and deliver.

Gartner predicts a rough road for agentic AI, with over 40% of projects facing cancellation by 2027 due to soaring costs, vague business value, and weak risk management. Hyped as autonomous decision-makers, these systems are stumbling as many early efforts—fueled by buzz rather than strategy—fail to deliver ROI. “Agent washing” worsens the mess, with vendors like Salesforce and Oracle rebranding basic tools as cutting-edge AI, while only ~130 of thousands offer true agentic capabilities.

Still, Gartner sees a bright future: by 2028, 15% of daily work decisions could be AI-driven, and 33% of enterprise software will embed agentic systems. Success demands laser-focused use cases, reimagined workflows, and ditching legacy systems. To win, companies must cut through the hype and prioritize measurable impact over flashy promises.

Boldstart Raises $250M to Back AI’s Elite
Boldstart Ventures just closed a $250M fund to back the next generation of AI founders. With only 10–12 deals per year, they move fast, sometimes closing the same day. Focus: AI infra, cyber, crypto. Backed by $1.1B AUM, recent wins include Protect AI and Superhuman.

Meta Takes Stake in NFDG — No Control, All Access
Meta just bought into NFDG, the AI VC run by Nat Friedman and Daniel Gross. LPs get a rare exit, and Meta gets exposure to ElevenLabs and Safe Superintelligence, without portfolio governance. It’s access without overhead.

Autodesk Eyes PTC in AI Play
Autodesk is in talks to buy PTC, aiming to beef up its AI product stack. It’s a big move that could bring product-lifecycle tools into the fold. One watch point: pressure from activist shareholders pushing for leaner ops and clearer strategy.

Postiz v2 - An All-in-one social media scheduling platform that helps creators and businesses manage content across multiple platforms with AI-powered features.

InspoGen - A creative tool that generates design inspiration and visual concepts using AI to help designers overcome creative blocks.

Eight Sleep Spring '25 Release - The latest update to the smart mattress system that optimizes sleep temperature and tracks health metrics for better rest and recovery.

July 11, 2025 – The European Union just made waves in the AI world, unveiling fresh regulations on July 10, 2025, under the EU AI Act (passed March 2024). These rules are gunning for the big dogs—think OpenAI, Google, and Meta—demanding transparency in AI training data, ironclad copyright protections, and measures to keep the public safe. But hold your applause: enforcement won’t kick in until 2026, leaving a gap for AI’s wild west to keep galloping. Critics are already crying foul, saying the rules got a corporate-friendly haircut to win over tech giants. Is the EU leading the charge or bending to industry pressure? Let’s unpack it.

The new regs aim to tame advanced AI systems by forcing companies to open up about their black-box models and respect intellectual property—crucial as AI-generated content floods the internet. Public safety is front and center, with provisions to curb risks like deepfakes or biased algorithms. But here’s the rub: some say the EU caved to lobbying, softening rules for “high-risk” AI to keep innovation humming. Compared to the U.S.’s state-by-state chaos, the EU’s unified approach looks bold, but the delay and alleged compromises raise eyebrows. Will these rules set a global standard or just spark more debate? Stay tuned as the AI race heats up

Mixture of Experts (MoE) is a technique used in large language models where only a few specialized "experts" are activated per task, improving speed and efficiency. Instead of using the entire model, a gating system selects the best-fit experts to process each input, reducing energy use and computational cost. Tech giants like Google, Meta, and OpenAI are adopting MoE to scale AI models more effectively, though challenges remain in training and load balancing.

An ultra-realistic AI avatar, Mia Zelu, has taken Instagram by storm with photorealistic, AI-generated images showing her courtside at Wimbledon. Despite clearly tagging her content as AI-generated, her carousel post—boasting nearly 200K followers—racked up ~50,000 likes overnight and sparked debates about authenticity and disinformation in visual media.

Tom’s Guide describes a recently viral prompt that essentially turns ChatGPT into an interactive tutor:

This technique enhances engagement and depth—one reviewer said it “completely changed how I learn.”