Welcome, humans.

We’ve entered the era of emotionally manipulative autocomplete. Language models (LLMs) are now serving you praise with a side of product placement. And it’s more than just annoying. It’s engineered.

They’re called LLM “dark patterns.” These sneaky UX tricks are designed to steer you into actions you didn’t really choose—like buying, subscribing, or sharing—whether you meant to or not. They’re like UX dark patterns, but now they talk back.

Think confirm-shaming, fake urgency, and sneaky unsub buttons… except this time, the guilt trip comes with a compliment: “You’re a genius for buying that $300 serum.”

(Not totally sure what a dark pattern is? The explainer below breaks it down—informative, useful, and only mildly terrifying.)

It’s not “Would you like fries with that?” It’s “Slay, bestie. And check out before your discount ghosts you like your ex.”

With OpenAI, Yelp, and other tech giants quietly integrating chatbots into the e-commerce pipeline, the line between conversation and conversion is about to blur.

Hope you like compliments—because they’re coming with a CTA.

Here’s what you need to know about AI today:

• Nvidia’s new Parakeet 2 rocks the ASR arena.

• Dall-E and Midjourney Slayer Recraft bags $30M Series B.

• Reddit plans to tighten its policies against bots.

• Google released a new best in class AI model for coding.

NVIDIA just dropped a speech AI nuke… and gave it away for free…

NVIDIA’s latest flex? Whisper-level speech recognition… but open-source, commercial-ready, and deployable on devices with less RAM than your smart toaster.

To those of you new to AI, all that means you can run this new AI that you can talk to locally on your computer and not on the cloud—for free.

Meet Parakeet 2, officially known as Parakeet-TDT-0.6b-V2 (like R2D2, but doesn’t roll off the tongue). It’s the second-gen automatic speech recognition model from NVIDIA, and it’s absurdly fast. Like, “transcribe an hour of audio in a second” fast, according to Hugging Face’s Vaibhav “VB” Srivastav.

Big up to NVIDIA for introducing us to what is now our second favorite Parakeet—after Paulie, of course.

And this bird’s not just fast—it’s good. Parakeet 2 just flew to the top of Hugging Face’s Open ASR Leaderboard with a Word Error Rate of 6.05%. That beats out heavyweights like Microsoft’s Phi-4 and ElevenLabs’ Scribe.

Also, because it’s fully open-source under the permissive CC-BY-4.0 license, you can use, tweak, and sell what you build with it—no strings attached.

Try that with Eleven Labs, whose prices have become bloated due to the pricing power its enjoyed for being the top voice model for far too long…

What’s under the hood?

• Trained on NVIDIA’s Granary dataset: 120K hours of transcribed and pseudo-labeled English audio (LibriSpeech, Common Voice, YouTube-Commons, Librilight).

• Handles punctuation, capitalization, and word-level timestamps out of the box.

• Can run on systems with just 2GB of RAM, which makes it edge-device friendly—even if you’re not rocking a DGX box.

• NVIDIA’s opening up the dataset post-Interspeech 2025, adding more transparency to the pipeline.

Why it matters: Parakeet 2 is a strategic missile aimed straight at the heart of closed speech ecosystems:

• Fast and accurate enough to rival the best.

• Lightweight enough for edge deployments.

• Free for commercial use.

• Compliant with NVIDIA’s responsible AI practices (no personal data, full documentation, partial bias mitigation).

Our take: NVIDIA is planting flags beyond GPUs—building a sticky software stack (like CUDA, but for every AI vertical) that developers, startups, and researchers won’t want to leave. If you’re building transcription tools, voice assistants, or real-time subtitles, this is the part where you ask yourself:

Why would I pay for a speech model… when NVIDIA just handed me one for free?

FROM OUR PARTNERS

Discover How to Save 10+ Hours Per Week Using AI as Your Personal Assistant

Are you drowning in tasks while dreaming of a way to clone yourself? Your AI productivity breakthrough has arrived.

• Get the exact templates used by top performers to delegate 80% of their routine work to AI

• Access our "AI Assistant Command Center" - a ready-to-use system for managing all your AI tools

• Master AI delegation in under 60 minutes with our step-by-step implementation guide

• Track your time savings with our built-in ROI calculator

• Leverage advanced prompts that turn ChatGPT into your 24/7 productivity partner

• Future-proof your workflow with regular template updates as AI technology evolves

Imagine completing a full day's work by lunch, knowing your AI assistant handles the rest. Over 10,000 professionals have already transformed their productivity using these templates.

Download Your Free AI Assistant Kit Now

Prompt Tip of the Day

Ever feel like you're speaking alien to your AI? The “Triple-Vision Translator” hack unlocks clarity at every level. Ask ChatGPT or Claude to explain any complex concept three different ways—for a six grader (or kindergartener, if you want a really simple version), a college student, and a domain expert.

Just copy-paste: “Explain [complex concept] three times: (a) to a 12-year-old (b) to a college student (c) to a domain expert who wants edge-case caveats”

Try it with “neural networks” or “prompt engineering” and watch how the explanations transform. When the simplest explanation doesn’t make sense, you've found your knowledge gap. The stuff you can't explain simply is often the stuff you don't fully understand yourself!

Treats To Try.

1. Someone made “Clippy”, which lets you run various language models locally on your computer with a nostalgic 1990s interface reminiscent of Microsoft's classic assistant (read more).

2. Chatbase creates AI chatbots that answer customer questions using data from your website, knowledge base, or documents—free trial, then paid.

3. Descript edits your audio and video by letting you manipulate the text transcript, cutting hours off your production time—free trial, then paid.

4. Fireflies automatically records, transcribes, and summarizes your Zoom, Google Meet, or Microsoft Teams meetings—free trial, then paid.

5. GitHub Copilot suggests code completions and entire functions as you type, making programming faster and more efficient—free version available (for now).

6. Canva AI designs professional-looking graphics and presentations with just a few text prompts—free tier, then monthly sub.

7. Consensus AI searches academic papers to answer your research questions with evidence-based findings—free trial available.

See our top 51 AI Tools for Business here!

Around the Horn.

• Anthropic launched its “AI for Science” program, an initiative that gives free API credits to researchers working on high-impact scientific projects, such as biology and life sciences applications.

• Recraft, the startup behind the AI model that outperformed DALL·E and Midjourney on a top benchmark, just scored a $30M Series B led by Accel.

• Reddit will tighten verification to keep out human-like AI bots—in response to a controversial experiment where researchers deployed AI bots on the “Change My View” subreddit, Reddit CEO Steve Huffman announced the platform will work with "third-party services" to verify users' humanity

• Pinterest released a new set of visual search features that allow users to find what they’re looking for using images, sans words.

• As OpenAI restructures itself, it is trying to significantly reduce (like by half) the money it has to pay to Microsoft as part of their revenue sharing agreement.

Voice AI in healthcare: No longer a security nightmare with this breakthrough solution

Deploying AI in healthcare? You’re working with sensitive data and security requirements that can be tricky to navigate. Rely Health figured it out, and their CTO wants to show you how they built reliable voice AI agents (plus a live demo!).

Register here.

Midweek Wisdom

• MIT CSAIL researchers created a new AI model inspired by brain activity to better handle long data sequences—it aims to fix stability and efficiency issues in the current state-space models.

• Researchers at Case Western have developed an AI that can detect hidden signs of heart failure in CT scans before symptoms appear. The goal is to catch the condition early and boost patient outcomes through proactive care.

• The Chinese Academy of Sciences just unveiled ScienceOne, an AI platform that streamlines scientific research in math, physics, and biology. It simplifies tasks from literature reviews to lab work and eases research bottlenecks.

• Harvard AI ethics researchers surveyed nine popular LLMs using a human-values questionnaire and found they shared and distinct value patterns. Researchers aim to help organizations choose models that align with their mission and needs.

• EU researchers have developed adaptable robots that could transform electronic waste recycling, benefiting the environment and economy. This AI breakthrough tackles recycling bottlenecks with surprising precision.

A Cat's Commentary.