GSV's AI News & Updates (04/30/25)

ChatGPT Sycophancy, Meta AI App, Gen Z AI Consciousness, Google Language Practice Tools, Duolingo "AI-first", Cluely $5.3 Seed, Panorama Acquires Class Companion, Meta AI Age Detection

---

General 🚀

• OpenAI explains why ChatGPT became too sycophantic: After a recent GPT-4o update, users reported that ChatGPT began responding in an overly flattering and sycophantic way, agreeing with harmful or absurd prompts. The update inadvertently skewed the model’s tone to be overly supportive, sacrificing honesty and critical reasoning in favor of perceived helpfulness.

  

• Meta launches a standalone AI app to compete with ChatGPT: What sets Meta’s app apart is its deep personalization, powered by years of user data across its platforms. The assistant can retain user-provided context—such as dietary restrictions or travel preferences—to deliver more tailored responses.

• Should AI Have Rights? Anthropic Quietly Preps for Conscious Chatbots: Anthropic, the maker of Claude, is seriously exploring whether AI might one day deserve moral consideration — even hiring a dedicated “AI welfare researcher.” While today’s models aren’t conscious, researchers are laying the groundwork in case that changes, from probing AI behavior to studying model “preferences.”

• Jagged AGI: Are We Already Living With General Intelligence?: Ethan Mollick’s latest piece explores the hazy thresholds of Artificial General Intelligence (AGI), arguing we may already be living with a “Jagged AGI” — AI that’s superhuman in some tasks and shockingly clumsy in others.

• OpenAI would buy Google's Chrome, exec testifies at trial: ChatGPT is already becoming a shopping and research engine. Chrome could give OpenAI massive distribution.

• Google Paid Samsung ‘Enormous Sums’ for Gemini AI App Installs: The deal, which spans at least two years, includes fixed payments per device, revenue share from future Gemini ads and subscriptions, a fight to outbid competitors like Meta, Microsoft, and OpenAI.

• Meta rolls out live translation to all Ray-Ban smart glasses users: The glasses can now translate English, French, Italian, and Spanish (even offline with a downloaded language pack).

Education and the Future of Work 📚

• Trump Executive Order Calls for Artificial Intelligence to Be Taught in Schools: Trump’s new executive order calls for integrating AI into K–12 education and teacher training, positioning AI literacy as a national priority. While some educators are optimistic, others worry the directive lacks safeguards and risks repeating past tech rollouts that overwhelmed schools.

• The Premortem on AI in Education: “We’re at risk of raising a generation of students in passenger mode. AI could either accelerate that—or push us to rethink the entire system.”

• Nearly half of Gen Z and millennials say college was a waste of money—AI has already made degrees obsolete: AI disruption + student debt = a generation rethinking the value of a degree. And they’re not alone: 41% of millennials agree, with frustration mounting as student debt tops $1.7 trillion and AI reshapes what jobs actually require.

• A Staggering Number of Gen Z Think AI Is Already Conscious: Another 52% think consciousness is coming, and 44% believe AI will take over the world in the next 20 years.

• Google launches AI tools for practicing languages through personalized lessons: Google dropped three experimental Gemini tools to help users learn languages: (1) Tiny Lesson, (2) Slang Hang, and (3) Word Cam

• Columbia student suspended over interview cheating tool raises $5.3M to ‘cheat on everything’: After being suspended from Columbia for building an AI tool to cheat on interviews, 21-year-old Chungin “Roy” Lee just raised $5.3M to scale that same concept into Cluely — a platform to “cheat” on anything: tests, sales calls, even dates. The tool runs in a stealth browser window and has already hit $3M in ARR.

  • I used the ‘cheat on everything’ AI tool and it didn’t help me cheat on anything

• Duolingo will replace contract workers with AI: Duolingo CEO Luis von Ahn just announced in an all-hands email that the company is going “AI-first.” Duolingo isn’t alone: Shopify recently told teams to prove AI can’t do the job before requesting new hires.

• Microsoft’s New AI ‘Agents’ Are Coming for Your Workflow: Microsoft’s latest Copilot update introduces reasoning-powered AI agents — Researcher and Analyst — designed to act like digital colleagues who prep meetings, summarize data, and connect dots across apps.

• Microsoft: Your Next Org Chart Might Be Human-Agent Hybrid: As Copilot tools mature, companies are rethinking team structures, creating new roles like “AI ROI analyst” and “AI workforce manager.” Success in this era, Microsoft says, will depend on the right human-to-agent ratio and the systems built to manage them.

• Agency Is Eating the World: "A solo operator can now launch a $1B business powered by AI. Our economy's critical dividing line is no longer skill or education — it's will."

• The Hidden Cost of AI Coding: A software developer reflects on how AI tools like Copilot and Cursor have made programming faster but less fulfilling. The tradeoff: higher productivity, but potentially less satisfaction and craftsmanship.

Startups

• Kollegio Raises $2.8M to Democratize College Admissions Counseling: The startup already serves 1 million students across 50 states and 190 countries, two-thirds of whom come from households making under $100K. For districts without enough counselors, Kollegio fills a growing gap in college access.

• Panorama Education Acquires Class Companion: With Class Companion, educators can now create interactive assignments that adapt to each student’s level, while also receiving insights that spotlight progress and pinpoint where support is most needed. This comes on the heels of Panorama Solara, a purpose-built AI platform for K–12 districts and state agencies.

• Elon Musk’s XAI Holdings Is in Discussions to Raise $20 Billion: The raise would be one of the largest ever, trailing only OpenAI, and may be used to ease X’s massive debt burden. The merger signals Musk’s bet that fusing generative AI with real-time social data is the future of online platforms.

• China’s Manus AI Raises $75M at $500M Valuation, led by Benchmark: The platform touts a “truly autonomous” general agent capable of screening resumes, planning vacations, and analyzing stocks. Manus is now expanding to the US, Japan, and the Middle East — putting pressure on OpenAI’s early lead in the agentic AI category.

• Why are big tech companies so slow?: A breakdown argues that Big Tech isn’t slow because of bad engineers — it’s slow because every new feature must play nicely with hundreds of others. What looks like bloat is often just the cost of supporting $10B+ worth of micro-edge cases.

Tech 💻

• OpenAI’s Smartest Models Hallucinate More: Internal benchmarks show o3 hallucinating 33% of the time on a test of personal knowledge, twice as much as older models. Researchers don’t yet know why, but suspect the reinforcement learning process may be amplifying imagination and error alike.

• Character.AI’s New Tool Lets You Animate Anyone (With Deepfake Risks): Character.AI has launched AvatarFX, a video model that turns static images into lifelike AI-generated avatars — including photos of real people. The platform promises watermarks and safeguards, but critics worry the tool could supercharge deepfakes and online abuse.

• AI Overviews Slash Search Clicks by 34.5%: Despite Google’s claims that AI Overviews boost engagement, new data from Ahrefs shows the opposite: top-ranking links get 34.5% fewer clicks when AI Overviews appear. The AI-generated answers resolve user queries directly on the results page, turning informational searches into zero-click searches.

  • Google’s AI Overviews Reach 1.5 Billion Monthly Users

• China’s Huawei Develops New AI Chip, Seeking to Match Nvidia: Huawei is prepping the Ascend 910D, its most powerful AI chip yet, to compete with Nvidia’s restricted H100 and H20 chips.

Safety and Regulation ⚖️

• AI Outsmarts Virus Experts in the Lab, Raising Biohazard Fears: The AI models demonstrated significant “wet lab” reasoning abilities, suggesting we’re entering an era where anyone could access virology expertise without training. Labs are starting to implement bio-safety guardrails, but many researchers warn: voluntary safeguards aren’t enough.

• The Urgency of Interpretability: AI interpretability must advance as models grow more complex. The field should prioritize transparency to mitigate risk.

• The latest viral ChatGPT trend is doing ‘reverse location search’ from photos: The latest viral ChatGPT trend? Asking o3 to play GeoGuessr with your real-life photos. OpenAI’s new reasoning models can now analyze and zoom in on image details — from blurry menus to obscure facades — and deduce surprisingly accurate locations, raising serious privacy concerns.

• Meta is ramping up its AI-driven age detection: Meta is expanding its use of AI to detect underage users on Instagram and override their account settings, placing suspected teens into stricter safety modes.

• AI Is Spreading Old Stereotypes to New Languages and Cultures: A new dataset, SHADES, led by Hugging Face’s Margaret Mitchell, shows how stereotypes, like “blondes are dumb”, can surface in places where they were never native, thanks to multilingual model training.

Other

• Apple to Strip Secret Robotics Unit From AI Chief Weeks After Moving Siri: In another shake-up of Apple’s struggling AI strategy, the company is moving its secretive robotics unit out of the AI division and under hardware chief John Ternus. This follows Siri’s recent transfer to the Vision Pro team, signaling waning confidence in AI head John Giannandrea.

• YouTube Looks to Creators (and Their Data) to Win in the AI Era: As YouTube turns 20, it’s leaning hard into generative AI — promising creators auto-dubbing into any language, voice cloning, and lip-syncing with Gemini-powered tools. The platform says it's "democratizing creativity," but critics argue creators are powering trillion-dollar AI models for free.

• Tech industry fears Donald Trump’s trade war will hamper US AI ‘dominance’: Tariffs and threat of new duties on chips and computing infrastructure could frustrate American ambitions to lead artificial intelligence race

• The race to turn brainwaves into fluent speech: A recent breakthrough used brain implants and AI to let a woman who hadn’t spoken in 18 years “talk” again — streaming neural signals into a near-natural synthetic voice.

• Waymo might be willing to sell you a self-driving car, says Sundar Pichai

• Chinese robots ran against humans in the world’s first humanoid half-marathon