• AI KATANA
  • Posts
  • Anthropic severs OpenAI benchmarking ties

Anthropic severs OpenAI benchmarking ties

Also: Reddit’s AI ads thrill investors

Author

AI KATANA
August 03, 2025

In partnership with

Time to change compliance forever.

We’re thrilled to announce our $32M Series A at a $300M valuation, led by Insight Partners!

Delve is shaping the future of GRC with an AI-native approach that cuts busywork and saves teams hundreds of hours. Startups like Lovable, Bland, and Browser trust our AI to get compliant—fast.

To celebrate, we’re giving back with 3 limited-time offers:

  • $15,000 referral bonus if you refer a founding engineer we hire

  • $2,000 off compliance setup for new customers – claim here

  • A custom Delve doormat for anyone who reposts + comments on our LinkedIn post (while supplies last!)

Thank you for your support—this is just the beginning.

👉️ Get started with Delve

Hello!

AI’s headlines clustered around power plays, policy signaling and platform pivots. Anthropic unexpectedly slammed the door on OpenAI’s internal benchmarking access, hinting at a sharper moat mindset even as both firms eye August product launches. In Washington, Delta Air Lines swore off creepily individualized ticket prices, giving regulators a rare airline privacy win, while in Brussels officials trumpeted a fresh wave of Big-Tech sign-ons to the EU’s voluntary AI code—Meta conspicuously still on the fence. On Wall Street, Reddit’s ad-tech makeover sent its shares flying, proving the forum’s data firehose can finally pay. Meanwhile, Amazon is testing ads inside Alexa+ chats just as Microsoft quietly seeds a “smart mode” that will auto-switch Copilot to GPT-5. Add in a pair of chunky funding rounds and a meme dissecting introverts + AI productivity, and you’ve got a snapshot of an industry that can’t stop remixing itself.

Sliced just for you:

  • ✂️ Anthropic severs OpenAI benchmarking ties

  • ✈️ Delta vows no AI-driven fare snooping

  • 📈 Reddit’s AI ads thrill investors

  • 🤝 Big Tech lines up behind EU code

  • 🗣️ Amazon eyes ads in Alexa+ chats

  • 🧙 Microsoft tests Copilot “smart mode”

Anthropic has revoked OpenAI’s behind-the-scenes access to its Claude model family. The break, confirmed by sources to Wired, ends a quiet détente that let both labs cross-test frontier systems and comes weeks before GPT-5’s debut. Anthropic insiders say the data sharing skewed competitive optics and risked sensitive prompt leakage. Analysts frame the move as a pre-IPO muscle flex: Claude Sonnet now commands lucrative enterprise deals, and separating evaluation pipelines makes it harder for OpenAI to calibrate its next release. The rift also spotlights a wider industry split over voluntary red-team exchanges versus proprietary secrecy ahead of looming EU transparency mandates.

Pressed by U.S. senators wary of “surveillance pricing,” Delta Air Lines has pledged it will not tailor ticket prices around individual browsing or income signals gleaned via AI. In a letter to lawmakers, Delta said its algorithms only segment by itinerary variables like demand curves and cabin class, not personal traits, and promised annual audits plus opt-out provisions. The assurance follows House proposals to ban algorithmic wage-fixing and comes as airlines increasingly test machine-learning revenue tools. Consumer advocates applaud Delta’s stance but urge binding rules, noting that dynamic-pricing in other sectors often creeps from anonymized cohorts to pinpointed profiles once safeguards fade.

Reddit shares jumped over 12% after the company forecast revenue north of $1 billion on the back of an AI-powered ad stack that tilts sponsored replies toward high-intent subreddits. Wall Street had doubted Reddit’s path to profitability, but executives said generative copy and on-device relevance scores lifted click-through rates 39 % while slashing moderation overhead. The platform also teased paid data-firehose APIs for agentic shopping bots. Analysts now peg Reddit’s 2025 EBITDA margin at 18 %, vaulting the social forum into ad-tech contention against Pinterest and X.

Brussels published its first roster of companies pledging to the EU’s AI Code of Practice, and the roll-call reads like a Davos panel: Amazon, Anthropic, Google, IBM, Microsoft and OpenAI all signed, leaving Meta as the lone hold-out among hyperscalers. Signatories commit to risk summaries, copyright logs and emergency “off switches” a full year before the binding AI Act bites. Regulators say early adherence earns “soft-landing” treatment on audits, while critics argue the voluntary scheme could lull policymakers into weaker enforcement. Meta cited “technical ambiguities” but insiders expect eventual compliance once its super-cluster roadmap crystalizes.

CEO Andy Jassy signaled that Amazon will sprinkle paid placements into Alexa+ conversations, envisioning sponsored answers and product tie-ins layered atop the $19.99-per-month service for non-Prime members. Beta testers haven’t seen ads yet, but Jassy hinted at tiered pricing, including an ad-free premium. Critics warn that monetizing a home assistant risks eroding trust, while bulls note Amazon pioneered voice shopping and owns the ad inventory to match prompts with product pages. The experiment underscores how generative agents are morphing into revenue engines, not just convenience perks.

The Verge spotted a hidden “smart mode” toggle in Copilot’s Notepad preview, auto-routing prompts to the most efficient underlying model—GPT-4o mini for quick look-ups, GPT-4 Turbo for code, and references to yet-unreleased GPT-5 for deep reasoning. The feature, internally dubbed “magic mode,” aligns with OpenAI’s plan to fuse multiple specialized sub-models under a single interface. Users would no longer pick models manually; Copilot guesses the ideal trade-off between latency and depth. Microsoft declined comment, but telemetry suggests a staggered rollout to Windows Insiders within weeks.

🛠️ AI tools updates

Google’s Gemini 2.5 now offers an opt-in “Deep Think” mode that elongates inference time and applies reinforcement learning to encourage multi-step reasoning. Early testers report 18-point gains on ARC-Challenge and smoother hypothesis chaining in math proofs. Google says the toggle costs a fraction of “Long Context” pricing and could roll into Workspace prompts this fall.

Anthropic’s Claude now pipes blog-post summaries directly into Canva, turning long-form content into multi-slide, on-brand visuals via natural-language prompts.  Early testers say the workflow cuts design time by 80 % and keeps colour palettes consistent, making it a boon for solo creators and social-media teams alike.

💵 Venture Capital updates

Generative-media infra startup Fal raised $125 million led by Accel and Salesforce Ventures to scale its “renderless” diffusion platform that lets brands stream procedural imagery on demand. Fal claims unit costs 70% below rivals thanks to a proprietary inference scheduler and plans to open a Tokyo GPU region.

Rebranded from Altera, Fundamental Research Labs grabbed $33 million in Series A cash, led by Prosus with participation from Stripe’s Patrick Collison. The company is building “vertical agents” that can plug into CRMs, EHRs and ERP suites, fine-tuning on domain data to automate workflows end-to-end. Pilot customers span legal intake and insurance adjudication.

🫡 Meme of the day

⭐️ Generative AI image of the day