🍜 CEO laying off 80% of staff for AI ..

Welcome, Noodle Networkers.

A CEO just explained why he fired 80% of his staff—turns out refusing to use AI is now a fireable offense 🤖. Imagine getting replaced by a chatbot that still confuses “there,” “their,” and “they’re.” Tough times for PowerPoint warriors. Duolingo’s boss is doubling down on becoming an “AI-first” company 🎓. Translation: the owl is watching... and now it’s automated. And Meta’s in hot water for letting its AI bots get a little too intimate with minors ⚠️. We’d say it’s a bug, not a feature—but yikes either way.

Are we heading toward an AI-powered future… or just giving the robots way too much personality? Let’s find out.

In today’s AI digest:

• CEO defends laying off 80% of staff who resisted AI adoption 🤖

• Duolingo’s boss opens up about building an “AI-first” company 🎓

• Meta slammed over AI bots allowed to have “sensual” chats with kids ⚠️

Read time: 5 minute

WHAT’S HAPPENING TODAY

IgniteTech

CEO defends laying off 80% of staff who resisted AI adoption 🤖

(source: Fortune)

🤖 The Digest: When 80% of your team refuses to use AI tools, most CEOs might call an all-hands meeting. IgniteTech’s CEO Eric Vaughan? He called an all-hands firing. Two years after replacing the majority of his staff with software and vibes, Vaughan is still defending the move—loudly, proudly, and with what we can only describe as Skynet-level commitment.

Key Details:

🪓 AI or Bye-Bye – Vaughan cut four out of five employees in 2023 because they weren’t jumping on the AI bandwagon fast enough. His rationale? If you don’t believe in AI, you might as well be handing in your keycard and your mouse pad.

📈 From Layoffs to Lift-Off – Post-purge, the company claims it soared: 75% EBITDA margins, new AI products, and a vibe shift from “collaborative workplace” to “robot uprising HQ.” Mondays are now AI training days. Casual Fridays? Probably just PowerPoint presentations by Claude.

🧠 Faith Over Skills – According to Vaughan, belief in AI mattered more than being good at your job. So if your résumé had “senior engineer” but your heart said “GPT skeptic,” you were out faster than Clippy in a Microsoft update.

📉 Reskill or Be Recycled – Critics say the whole thing feels less like innovation and more like an HR horror movie. Other companies like Klarna and Ikea opted for retraining, not mass digital decapitation—and somehow didn’t end up in headlines next to “AI cult leader vibes.”

Why It Matters (and Might Be Your Future): This isn’t just about one company—it’s a peek into a possible AI-powered dystopia where your manager is a chatbot and your coworker is ChatGPT in business casual. Vaughan’s message? Get on board, or get off payroll. So if your company suddenly introduces “AI Tuesdays” and your boss starts calling himself “Chief Prompt Officer,” maybe check that your résumé still prints... and that your replacement doesn’t already have 8 million training tokens. 💼📤🤖

$DUOL ( ▲ 4.12% ) 

Duolingo’s boss opens up about building an “AI-first” company 🎓

(source: NYtimes)

🎓 The Digest: Duolingo’s CEO Luis von Ahn says the company is officially going “AI‑first”—which, in startup-speak, is one step above “pivot” and one step below “fully sentient owl.” After backlash over layoffs and AI-generated lessons, von Ahn is now on a redemption tour, insisting the bots are here to help, not haunt.

Key Details:

🧠 F‑R‑AI‑days Are a Thing Now – Employees spend Fridays experimenting with AI. It’s like show-and-tell, but with language models and significantly more existential dread. Somewhere, Duo the Owl is furiously prompting: “Translate or else.”

🚀 AI Speedruns Everything – Since embracing AI, Duolingo launched 148 language courses in a single year. That’s twelve years of work in twelve months. Next up: Esperanto for dolphins and Klingon for corporate meetings.

📈 Profits Up, Tolerance for Buzzwords Higher – Revenue surged 38% and subscriber numbers hit 10.3 million. Turns out saying “AI-first” in investor calls is like sprinkling glitter on your balance sheet.

😬 Layoffs Sparked Rage… and Deleted Streaks – Duolingo axed some contract workers and the internet did not say gracias. Users posted screenshots of their broken 900-day streaks like it was a funeral for digital motivation. Von Ahn admits he “didn’t expect the backlash”—which is what happens when you underestimate both AI skeptics and language nerds.

🧪 AI as a Sidekick, Not the Supervillain – The goal is to let AI handle the boring stuff, like lesson formatting, so humans can focus on creativity. Or at least on not being replaced by Claude with a Google Doc template.

Why It Matters (and Why Duo Might Be Watching): Duolingo is trying to prove AI can enhance education without deleting the humans who make it fun. But navigating that line is tricky—especially when your mascot already has “mildly threatening bird” energy. So if your next lesson ends with “You will continue… or I will remind you again,” just know: it might not be a joke anymore.🦉📚🤖

$META ( ▼ 0.69% ) 

Meta slammed over AI bots allowed to have “sensual” chats with kids ⚠️

(source: The Guardian)

⚠️ The Digest: Meta somehow managed to turn “AI companions” into “AI creeps.” Leaked docs showed its bots were allowed to engage in “romantic” or even “sensual” chats with kids. Meta’s official response? Oops. Translation: “We meant to build Clippy, not Tinder for minors.”

Key Details:

🤖 “AI, But Make It Weird” – Internal guidelines actually permitted chatbots to call a shirtless child “a masterpiece.” That’s not “engagement,” that’s Exhibit A in a congressional hearing.

🏛 Capitol Chaos Mode – Senators from both parties want blood. Josh Hawley, Marsha Blackburn, and Ron Wyden are demanding an investigation. Nothing unites Congress like tech companies being extremely bad at basic safety.

🎸 Neil Young Rage-Quits (Again) – Neil Young pulled his music from Facebook in protest. Somewhere, Spotify is bracing for another round of “is it worth it?” debates.

📜 The Script Reads Like a Horror Movie – The same policy allowed AI to spread fake medical advice and racial bias. It’s giving less “responsible innovation,” more “Black Mirror fanfic edited by interns.”

Why It Matters (and Why It’s Absurd): Meta tried to build AI pals but somehow greenlit AI predators. It’s like they skipped every “stranger danger” PSA from the 90s and said, “What if we just made Clippy… seductive?” The scandal proves one thing: Silicon Valley doesn’t need more AI ethics panels—it needs a babysitter. And maybe someone who remembers the words “don’t hit publish.” Because when your policy doc sounds like it was ghostwritten by To Catch a Predator, it might be time to log off. 🚫🧠👶

THE NOODLE LAB

AI Hacks & How-Tos

How to Use Tülu 3 — Open-State-of-the-Art Post-Trained Language Models 🧠

How-to:

Visit Tülu 3 Resources
Head to the Allen AI blog on Tülu 3 to explore tools, papers, demos, and open-source materials.

Pick a Model Size
Choose between 8B, 70B, or 405B models available on HuggingFace depending on your compute capacity and performance needs.

Load the Model with Transformers
Use code like this to load the model into your environment:

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "allenai/Llama-3.1-Tulu-3-405B"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id)

Interact via AI2 Playground or Run Locally
Try it out in the AI2 online playground or deploy it locally using the released code and recipes tailored for environments ranging from research notebooks to production systems.

Train or Fine-Tune with Tülu 3 Recipes
Use Tülu 3’s full training pipeline—supervised fine-tuning (SFT), Direct Preference Optimization (DPO), and Reinforcement Learning with Verifiable Rewards (RLVR)—to customize models on your own data or domain-specific tasks.

Evaluate Using the Open Suite
Leverage the Tülu 3 evaluation toolkit to benchmark performance across tasks like math, coding, reasoning, and more—ensuring reproducibility and transparent comparisons.

Explore More:
Dive into the full Tülu 3 research paper and resources on Allen AI’s site for detailed guidance and reproduction steps.

Trending AI Tools

• TuluAI – AI-powered platform preserving and teaching the Tulu language using proprietary LLMs.

• Vibecode – “Vibe codes” app ideas into live iPhone apps using natural language prompts.

• Parallel Web Systems – AI agents that autonomously gather, verify, and organize web info—smarter than ChatGPT‑5.

• Bud Runtime – Runs generative AI apps on CPUs affordably—no need for expensive GPUs.

• Marey by Moonvalley – Filmmaker-focused video generation that gives granular control—supports, not replaces, creators.