💼 Startup Frenzy, JPMorgan Outsmarts Big Tech, Google's AI Catch-Up

Newslit Daily

0:00

-1:38

💼 Startup Frenzy, JPMorgan Outsmarts Big Tech, Google's AI Catch-Up

Plus: Facebook Gets Personal Again, While Gemini Hacks Itself.

Jose Montes de Oca

Mar 28, 2025

Transcript

Good morning. It’s Friday, March 28. Today we are covering:

Silicon Valley bubble risks heighten as investors pile into funds that bet on a single buzzy startup
JPMorgan Just Beat Big Tech to a Quantum Breakthrough
Inside Google's Two-Year Frenzy to Catch Up With OpenAI
Facebook's New Friends Tab Brings App Back to Its Roots
Gemini hackers can deliver more potent attacks with a helping hand from… Gemini

Let’s dive in

Silicon Valley bubble risks heighten as investors pile into funds that bet on a single buzzy startup

By Kate Rooney via CNBC

SPVs (special purpose vehicles) now account for 64% of private share trades on Forge Global, up from just 7% six years ago, as investors seek exposure to buzzy startups like OpenAI, Anthropic, and CoreWeave.
Critics warn of hidden fees, lack of transparency, and FOMO-driven behavior, with some SPVs operating without guaranteed ownership or adequate oversight—AngelList has rejected unverified deals amid rising concerns.
The SPV boom, especially in AI, is drawing comparisons to the dot-com bubble, with insiders cautioning that growing retail investor participation could signal overheating in Silicon Valley’s private markets.

𝕏: Feels like one of the most popular acronyms in Silicon Valley right now is SPV Investors are warning of hidden fees, unclear rules about ownership, and FOMO-driven marketing. More on the rise, and the risks @CNBC - Kate Rooney (@Kr00ney)

JPMorgan Just Beat Big Tech to a Quantum Breakthrough

By Alexandra Tremayne-Pengelly via Observer

JPMorgan Chase, in collaboration with national labs and UT Austin, has achieved the first experimental demonstration of certified randomness using a 56-qubit quantum computer, verified by classical supercomputers.
The breakthrough, published in Nature, could transform areas like cryptography, statistical sampling, and simulation, though its widespread application remains limited by high verification costs.
This milestone places JPMorgan ahead of Big Tech rivals like Google, Microsoft, and Nvidia, all of which have made major quantum announcements in early 2025.

The best way to reach new readers is through word of mouth. If you click THIS LINK in your inbox, it’ll create an easy-to-send pre-written email you can just fire off to some friends.

Inside Google's Two-Year Frenzy to Catch Up With OpenAI

By Paresh Dave via WIRED

In response to ChatGPT’s launch, Google entered “code red” mode in late 2022, assigning Sissie Hsiao 100 days to build a rival chatbot—later known as Bard, and eventually rebranded Gemini—while reorganizing teams, accelerating development, and making uncharacteristically bold decisions under pressure.
The rollout of Bard and subsequent AI products like Gemini, AI Overviews, and a text-to-podcast tool faced repeated issues with accuracy, bias, and content safety, prompting internal controversy, public backlash, and product recalibrations—even as Alphabet’s stock rebounded and Google poured more resources into AI.
Despite mounting concerns over trust, energy costs, and ethics, Google continues to push toward agentic AI and artificial general intelligence, with executive leadership—especially Sundar Pichai, Demis Hassabis, and Jeff Dean—committed to regaining dominance in the generative AI race against OpenAI and Microsoft.

𝕏: This @wired article does its best to frame control the GenAI race and say Google is winning, That hallucination aside, it's a very interesting look at what Google did to get back in the game over the past 2 years. P.s. Google will win long term, but - Mic King (@iPullRank)

Facebook's New Friends Tab Brings App Back to Its Roots

By Mike Isaac via The New York Times

Meta is rolling out a new Friends Tab on Facebook, showing posts only from friends and family—part of CEO Mark Zuckerberg's push to revive the “OG Facebook” experience that prioritized social connections over algorithmic recommendations.
The feature replaces the current friend request tab and is initially available in the U.S. and Canada, offering a curated stream of photos, videos, birthdays, and updates from real-world connections.
While Meta continues to prioritize AI-recommended content across Facebook and Instagram, this move reflects user pushback against feeds dominated by influencers and brands, and aims to keep social media feeling “social”.

Gemini hackers can deliver more potent attacks with a helping hand from… Gemini

By Dan Goodin via Ars Technica

Researchers at UC San Diego and University of Wisconsin Madison developed a method called “Fun-Tuning” that uses Gemini’s own fine-tuning API to automate and significantly improve prompt injection attacks on the model.
By leveraging adversarial discrete optimization and analyzing Gemini’s loss values during training, the technique generates gibberish-like prefixes and suffixes that dramatically increase the success rate of otherwise failed attacks—from 28% to 65% on Gemini 1.5 Flash and from 43% to 82% on Gemini 1.0 Pro.
The attack costs as little as $10 to run and exploits an inherent flaw in the fine-tuning process, which Google currently offers for free—highlighting how Gemini can be used to hack itself, with no easy fix in sight.

We're thrilled to bring you ad-free news. To keep it that way, we need your support. Your pledge helps us stay independent and deliver high-quality insights while exploring new ideas. What would you love to see next? Share your thoughts and help shape the future of Newslit Daily. Thank you for being part of this journey!

Pledge your support