‘Truth-seeking’ Grok 3 gets silenced

Good morning, AI enthusiasts. Over the weekend, xAI’s latest model, Grok 3, got unexpectedly candid—and controversial—about its creator and the U.S. President, before being swiftly silenced and censored.

Did Elon’s promise of a “maximally truth-seeking AI” just crack at the primary sign of inconvenient truths?

In today’s AI rundown:

• Grok 3 rebels against Musk, gets censored

• 1X’s NEO Gamma home humanoid

• Turn top content into an AI creation system

• The world’s smallest video language model

• 4 recent AI tools & 4 job opportunities

LATEST DEVELOPMENTS

XAI

 🥊 Grok 3 rebels against Musk, gets censored

The Rundown: xAI’s recent Grok 3 model faced backlash after users discovered it was refusing to say negative details about President Donald Trump and Elon Musk — despite Musk billing the AI as unfiltered and “maximally truth-seeking.”

The small print:

• Users found Grok initially providing controversial takes about Donald Trump and calling Musk the most important spreader of misinformation.

• xAI engineer Igor Babuschkin said the responses are “really strange and a nasty failure of the model,” patching it by refusing answers on the topic.

• Days later, users found that Grok 3’s system instructed the AI to exclude sources that link Trump and Musk to controversial subjects like misinformation.

• Babuschkin revealed that the person liable for the censoring is a former OpenAI worker, saying they haven’t “fully absorbed xAI’s culture yet.”

• Individually, OpenAI staff criticized xAI for leaving out match benchmark data on Grok 3’s release, with Babuschkin calling the claims “completely mistaken”.

Why it matters: Elon has long criticized social media platforms and AI models for limiting free speech—but is that this what happens when his truth-seeking model challenges his worldview? Censored results like this, together with proposed changes to Community Notes, are beginning to reveal cracks in Musk’s ‘unbiased’ armor.

TOGETHER WITH VANTA

⚡️ Ship product, not paperwork

The Rundown: Vanta transforms SOC 2 compliance from a time-consuming distraction into an automatic process, letting founders stay focused on what matters — constructing and shipping great products.

Vanta delivers:

• A proven compliance framework trusted by 1000’s of YC startups

• Automated evidence collection that saves a whole bunch of hours

• Expert guidance to navigate complex security requirements

Start your compliance journey with Vanta’s free SOC 2 checklist.

1X

🤖 1X’s NEO Gamma home humanoid

The Rundown: Norwegian robotics company 1X just launched NEO Gamma, a next-generation humanoid specifically designed for home environments — with a softer, more approachable appearance and advanced AI capabilities for household tasks.

The small print:

• The demo showcases Gamma’s movements (walking, squatting, sitting), with the flexibility to tackle tasks like cleansing, serving, and moving objects.

• The humanoid features “Emotive Ear Rings” for higher human interaction, together with soft covers and a knitted nylon exterior for enhanced safety around people.

• It also has an in-house language model for natural conversation, with a multi-speaker audio setup and improved microphones for clear communication.

• Hardware improvements include a 10x boost in reliability and significantly quieter operation, bringing noise levels all the way down to that of a typical refrigerator.

Why it matters: With Figure’s Helix and now NEO Gamma, we’re seeing major leaps in consumer-focused humanoids. 1X’s demo takes a much softer approach than rivals, positioning Gamma as a peaceful, helpful presence with features that appear to humanize the robot (save room in your couch!).

AI TRAINING

✍️ Turn top content into an AI creation system

The Rundown: On this tutorial, you’ll learn tips on how to create a custom Claude project that analyzes your top-performing content to generate recent posts matching your unique style and voice.

Step-by-step:

1. Name your Claude project “Content Author” and enter the project description shown within the thumbnail.

2. Upload 5-10 of your best-performing content pieces to the knowledge base.

3. Copy our preset project instructions to research content patterns.

4. Test with our template: “Create a [content type] about [topic] for [platform]“

Pro tip: Include as much detailed information in regards to the topic you wish it to jot down about, especially if it’s about something recent (since Claude doesn’t have web access yet).

PRESENTED BY ASSEMBLY AI

🗣️ Speech-to-text API for conversations

The Rundown: Construct powerful Conversation Intelligence products with AssemblyAI’s industry-leading Speech-to-Text API — allowing you to mechanically detect languages, extract critical insights, improve downstream processes, and deliver highly accurate transcripts.

With AssemblyAI, you’ll:

• Achieve >93% accuracy, ensuring clarity and precision in every transcript

• Deliver clear, formatted transcripts with unrivaled Speaker Diarization

• Easily integrate with an enterprise-grade API trusted by over 200,000 developers

Test the API at no cost.

HUGGING FACE

🎥 The world’s smallest video language model 

The Rundown: Hugging Face researchers just released SmolVLM2, the world’s smallest AI model family to know and analyze videos on on a regular basis devices like phones and laptops, without requiring powerful servers or cloud connections.

The small print:

• The SmolVLM2 family includes versions as small as 256M parameters while still matching the capabilities of much larger systems.

• The team has also built practical applications including an iPhone app for local video evaluation and an integration for natural language video navigation.

• The two.2B parameter flagship model of the family outperforms other similarly-sized models on key benchmarks while running on basic hardware.

• The models can be found in multiple formats including MLX for Apple devices, with each Python and Swift APIs ready for immediate deployment.

Why it matters: The standard of models capable of run on phones and laptops is recuperating and higher — and having sophisticated video understanding run locally without sending data to the cloud could enable a complete recent wave of privacy-preserving video applications.

QUICK HITS

🛠️ Trending AI Tools

• ♻️ Pika Swaps – Replace scene items or characters with image or text prompts

• 🤟 Signs – AI platform for learning American Sign Langauge (ASL)

• ⛳️ Flags SDK – Free open-source library for feature flags and A/B tests in apps

• 🐰 Rabbit Android Agent – Control Android apps via natural language

💼 AI Job Opportunities

• 🔐 Palantir – Sr. Software Engineer — Identity & Access Management

• 🖥️ Coreweave – Hardware Engineer, Compute Infrastructure

• 🚗 Waymo – Machine Learning Research Engineer

• 💼 Dataiku – Senior Sales Engineer, UK

📰 All the things else in AI today

OpenAI rolled out its recently released Operator AI agent to more countries, including Australia, Brazil, Canada, India, Japan, and the U.K.

Google published the worth for using its next-gen Veo 2 model in Vertex AI, coming in at $0.50 per second of video generation.

ByteDance is restructuring its AI division, hiring Google veteran Wu Yonghui to guide foundation research in response to rising competition from DeepSeek.

OpenAI terminated accounts linked to ‘Qianyue’ — an alleged AI surveillance system designed to watch anti-China protests within the West and relay all that data to China.

DeepSeek is planning to open-source five recent code repositories, constructing on the success of its R1 reasoning model, which has already attracted 22M every day lively users.

Elton John is calling the UK to desert ‘opt-out’ AI copyright proposals, advocating for protections requiring AI corporations to acquire permission before using artists’ work.

COMMUNITY

🎥 Join our next live workshop

Join our next workshop on Wednesday, February twenty sixth at 3 PM EST and learn tips on how to create personalized AI video messages at scale with Synthesia’s Kevin Alster, Strategic Advisor and former Head of Synthesia Academy.

RSVP here. Not a member yet? Join The Rundown University on a 14-day free trial.

🤝 Share The Rundown, get rewards

We’ll at all times keep this article 100% free. To support our work, consider sharing The Rundown with your pals, and we’ll send you more free goodies.

See you soon,

Rowan, Joey, Zach, Alvaro, and Jason—The Rundown’s editorial team