amorebernard

[Shining Science @ShiningScience]

Global data reveals Gen Z is the first generation to underperform their parents in attention, memory, and literacy skills.

For decades, global intelligence scores followed an upward trajectory known as the Flynn Effect. However, new neuroscientific research suggests this trend has sharply reversed. According to data from over 80 countries presented by Dr. Jared Cooney Horvath, Generation Z is the first demographic to score lower on key cognitive measures—including IQ, memory, and numeracy—than the Millennials who preceded them. This downturn, which became noticeable around 2010, marks a significant shift in human cognitive development, raising concerns about the long-term impact of modern digital environments on brain function.

Researchers point toward the ubiquitous presence of digital technology as a primary driver for these changes. As screen time increases, it often replaces the deep, interactive learning experiences necessary for building complex neural pathways. Critics emphasize that the shift away from critical thinking toward passive consumption may be fundamentally altering how the younger generation processes information. To combat this decline, experts are urging a return to balanced digital habits and richer, more immersive learning environments that challenge the mind beyond the digital interface.

source: Horvath, J. C. (2026). Digital saturation and the reversal of the Flynn Effect in Generation Z. University of Melbourne Research Reports.

[Demirdjian Twins @demirdjiantwins]

Nano banana + Linah AI + n8n = Ad Factory

This system pumps out TikTok/FB/Insta video ads on autopilot using the latest AI video models.

- No actors.
- No editors.
- No overpriced agencies.

Just endless, scroll-stopping UGC-style ads at scale.

Perfect for e-com brands & growth agencies who need constant creative testing.

Here’s how it works:
→ Drop your product catalog into Airtable
→ n8n pulls product data + hooks
→ Linah AI generates video variations (hooks, demos, product-in-hand)
→ Auto-styles each ad for platform-specific virality
→ Airtable logs everything so you can track winners

24/7 production.
Pennies per video.
You own 100% of the assets.

Want the full template?

Comment “NANO” + like this post, and I’ll DM it to you.
(must be following)

[nxxn @sol_nxxn]

You can basically LIVE on Solana

Phone? - @solanamobile
SIM card? - @helium
Stocks? - @xStocksFi
Gaming? - @playsolana, @_portals_
Wallet? - @solflare
Debit Card? - @KASTcard, @solflare etc.
Savings? - @uselulo
Fitness? - @moonwalkfitness
Health? - @CudisWellness
Real Estate? - @RECCFinance
VPN? - @PingNetwork_io
Travel? - @TravelSwap_xyz

[Renç Korzay @renckorzay]

OpenAI fired this 23-year-old from their Superalignment team.

But he turned his insider knowledge into a $1.5B fund that's outperforming Wall Street by 700% this year.

He says maybe ~200 people in SF understand what's *actually* happening in AI right now.

Here's his thesis: 🧵

[Anna (reality check arc) @basilda_a]

after $trump coin launch, solana's liquidity got sucked away.

if you track wallets you see how crackhead KOLs who larp as millionaires PvP each other on 10-100k mcap toppers everyday. this is state of trenches for the last 6 months.

some not very smart of you even copy trade them and provide with exit liquidity in hopes for quick 10x.

new launches in dry environment kept vamping all solana existing coins, and the ones that survived gonna come back stronger.

i think it's important to stay flexible in crypto and adapt fast to changing market environment.

imho rotators meta is coming to an end. dumb money got liquidated from the market and the ones that stayed with couple of brain cells are actually starting to believe in smth and bagwork again. bcs it's a best strategy for preserving the capital.

so how it was with eth top memes which survived the rinse on eth when solana memes meta came, the same will repeat with solana top meme communities, new generation of memes that were born in these severe market conditions and survived will come back stronger on the next market leg up.

$troll
$usduc
$neet
$mask
$tokabu
$trencher
$chillhouse $house etc.

[Jake Brukhman @jbrukh]

Here’s an accessible breakdown of @PluralisHQ’s incredible paper.

When we train large models on decentralized networks, the idea is to break them down into pieces and have different nodes process the different pieces.

There are a few ways to do this.

One way is low hanging fruit: give each node a copy of the neural network but split the data among them to train in parallel. This is great, but for really large models single machines become the bottleneck as they can’t hold all the nodes and parameters and data.

Enter model parallelism. In this approach, we break the neural networks themselves into pieces on different machines and try to find ways of making the overall training converge.

Model parallelism is hard because a lot of information has to be communicated between nodes. For example, suppose I have two model layers with 4096 neurons. To compute a forward pass, I need to send a 4096-dimensional vector of activations to the next layer. If I want to compute a backward pass, then I need to pass a 4096-dimensional vector of gradients to the previous layer.

Such vectors could be at least 32KB, and there’s a lot of layers and data points, so bandwidth can pile up quickly.

This is the main reason that a lot of people didn’t think decentralized training would work. The bandwidth requirements are so large, on slow internet the model would take forever to synchronize.

In one example I estimated on the back of a napkin, training an 8-layer, 2B parameter model on 4 GPUs could require 1.5 GB of inter-node traffic per step, or 12 GB/s if you want to take one step per second — and steps to train a model could easily go into the millions.

One way scientists have tried to get ahead of these bandwidth requirements is by compressing the data that has to travel between nodes. But all previous research has shown that (lossy) compression schemes don’t work well — as compression goes above 10% or so they tend to destabilize training processes and training fails to converge.

Enter @PluralisHQ’s paper. Pluralis team shows that most of the important information contained in activations and gradients actually lives in a much smaller subspace of the parameter space. During training the model can “learn” what this subspace is and project all the 4096-dimensional vectors into it. What’s amazing is this subset can be as small as 40 dimensions, effectively creating a 100x compression. Moreover, this compression is lossless, so training converges!

Now, instead of having to use 100 GB/s data center fabric, we can train the same model on 80 MB/s home internet.

The implications are huge:

- We can now train huge models on decentralized networks.
- Networks can be geographically distributed and use commodity hardware.
- Models can now be sharded and collectively owned.
- Open source models can be monetized.

[Poonam Soni @CodeByPoonam]

OpenAI, Google, and Anthropic just dropped the guides on:

- Prompt Engineering
- Building effective Agents
- AI in Business
- 601 AI use cases

and so much more...

9 best guides you don’t want to miss:

[Zachary Huang @ZacharyHuang12]

AI Codebase Knowledge Builder
Online Service: https://t.co/nBQpRysBNf
Github: https://t.co/5r4iTsE0kl
Documentation: https://t.co/bXqu61q0Mx
Substack: https://t.co/u2fnnU3EU3

Related YouTube Videos
Development: https://t.co/pYGjlLUQYw
Podcast: https://t.co/WadDxYgMDv

[Sumanth @Sumanth_077]

Microsoft released an AI powered data analysis tool!

Data Formulator is an AI-powered tool for analysts to iteratively create rich visualizations.

It's no-code & 100% open-source

[elvis @omarsar0]

AI Agents vs. Agentic AI

Interesting paper summarizing distinctions between AI Agents and Agentic AI.

It also talks about the key ideas, solutions, and the future.

Here are my notes:

[Dave White @_Dave__White_]

The paper at https://t.co/0domNHDiJW goes into the conceptual framework and implementation pathways in more depth.

We're curious what new applications multiverse finance might unlock.

If you've got ideas, we'd love to hear from you.

7/7

[Mario Nawfal @MarioNawfal]

🚨 UNCANNY VALLEY: LEARNING TO CODE WON’T SAVE YOU

Forget Python. If you don’t know how to think like an algorithm, Danks says the future will leave you behind.

Dr. David Danks:

“We need students to think computationally and algorithmically—but we don’t need them to write Python scripts.”

In the world of AI, it’s not what you can code—it’s how you think that matters.

[TuongVy Le @TuongvyLe12]

1/ Today we’re publishing Crypto and the Evolution of the Capital Markets. Skeptics like to say crypto is a solution in search of a problem, but in reality, it’s an answer to a decades-old one that the traditional securities markets still haven’t fixed: the lack of a more direct, efficient, and trustless system for owning and trading assets. Blockchain and tokenization represent a natural evolution of the capital markets, the same way the move away from paper stock certificates did, half a century ago.

[Nick at ListingTrack @listingtrack]

Two big Bitcoin-holding reverse mergers announced this morning:

Eric Trump's American Bitcoin will reverse merge into Gryphon Digital Mining $GRYP and will trade as $ABTC upon closing. Image from the presentation attached

While, David Bailey's Nakamoto will reverse merge into KindlyMD $KDLY.

[The Coin Republic @TCR_news_]

Here are the top 10 public companies by Bitcoin holdings:

#Microstrategy #MARA #RiotPlatforms #Tesla #CleanSpark #Hut8 #Coinbase #Block #GalaxyDigital #HIVEDigital #BitcoinHolding

[Hassan @nutlope]

My AI side projects this year:

◆ llamacoder​.io: 1.5M users
◆ blinkshot​.io: 1.4M users
◆ llamatutor​.io: 146k users
◆ llamaOCR​.com: 91k users

Here's my full process for building AI apps:

1. Ideation – Get inspired on X, Reddit, Product Hunt, newsletters, YC sites. Keep a list of good ideas & pick one that I'm most excited about.

2. Naming – Come up with a good name for the project – can use tools like domainsGPT to do this.

3. Design – Think through how the app will work and how many screens it will have. Can be a rough sketch, Figma, or tools like v0/bolt/lovable.

4. Building – Make the simplest possible working version – ideally 1 API endpoint working end to end

4. Auth/limits – See if I wanna add auth, rate limit by IP address, add bring your own key, ect depending on the app

5. Prep for launch – Write a nice README, get an OG image for the app, get a nice domain, add analytics

6. Launch – Draft LinkedIn/X posts, make the code open source, & announce it. Keep an eye out for any bugs to fix on socials.

Overall, my philosophy is to try to keep these projects as simple as possible, with a single API call, and focus on building a really nice UI that impresses users. That + making it free & open source are a big part of why I think my apps do well.

Also, by keeping these apps as simple as possible, you spend less time on them and if they flop, you can just move on to the next one. Only double down after you see traction.

[AshutoshShrivastava @ai_for_success]

Best in AI - May 2025

Coding : Gemini 2.5 Pro / Claude 3.7 Sonnet
Multimodal : Google Gemini
Video : Kling 2.0 / Veo-2
Math : o3
Voice : OpenAI AVM / Gemini Live
Writing : Claude 3.7 Sonnet / GPT-4.5
Search : OpenAI Deep Research / Gemini Deep Research
Web Agent : Manus AI / Convergence AI
Image : GPT-4o / Midjourney

Agree ?

[LangChain @LangChain]

🔍 Company Researcher

A multi-agent system that generates real-time company research reports using LangGraph. Specialized nodes analyze business, financial, and market data through an intelligent pipeline.

Check out the demo 📊
https://t.co/8Z8GpsU0DL

[The DeFi Investor 🔎 @TheDeFinvestor]

Crypto Watchlist for the week ahead:

$S - Sonic is expected to make a big announcement on May 15

$KAITO - Kaito will announce a tokenomics upgrade to reward KAITO diamond hands next week

$STX - Stacks will increase the cap for sBTC, its BTC-backed asset for DeFi, on May 15

$BTC - VanEck’s Onchain Economy ETF, the first actively managed crypto ETF, will launch on May 14

$RESOLV - The registration deadline for the Resolv Airdrop is May 16

$DEAI - Zero1 Labs will release its updated technical roadmap for 2025-26 next week

$METIS - Metis's next major network upgrade will go live on May 14

$APT - $65M worth of APT will be unlocked on May 12

If you enjoyed reading this, a like and a repost would be appreciated🫡

[Brett Adcock @adcock_brett]

Another huge week of AI and robotics news.

So, I summarized everything from OpenAI, Google, Meta, Microsoft, FutureHouse, Mistral, Unitree, Stanford, UC Berkeley, Hugging Face, and more.

Here's everything you need to know and how to make sense out of it: