Kons Anagnostopoulos

Beauty and technology will save the world.

[Kons Anagnostopoulos @CaesarKons]

The Gentle Singularity https://t.co/ShYsbyZ3oK

[PJ Ace @PJaccetturo]

This is crazy. Everything you see is prompted and trained on hundreds of hours of Tom & Jerry footage.

We've been talking with a few Japanese studios that want to do this for their older IPs. You just need training data to create new episodes. 🤯

[Paul Graham @paulg]

I found this hard to believe, but it's really true. The movement against plastic straws began with a mistaken statistic in a 9 year old boy's school project that went viral. How can plastic straws still be banned anywhere?

[Wall St Engine @wallstengine]

SemiAnalysis published an analysis on DeepSeek, addressing recent claims about its cost and performance. $NVDA

The report states that the widely circulated $6M training cost for DeepSeek V3 is incorrect, as it only accounts for GPU pre-training expenses and excludes R&D, infrastructure, and other critical costs. According to their findings, DeepSeek’s total server CapEx is around $1.3B, with a significant portion allocated to maintaining and operating its GPU clusters.

The report also states that DeepSeek has access to roughly 50,000 Hopper GPUs, but clarifies that this does not mean 50,000 H100s, as some have suggested. Instead, it’s a mix of H800s, H100s, and the China-specific H20s, which NVIDIA has been producing in response to U.S. export restrictions. SemiAnalysis points out that DeepSeek operates its own datacenters and has a more streamlined structure compared to larger AI labs.

On performance, the report notes that R1 matches OpenAI’s o1 in reasoning tasks but is not the clear leader across all metrics. It also highlights that while DeepSeek has gained attention for its pricing and efficiency, Google’s Gemini Flash 2.0 is similarly capable and even cheaper when accessed through API.

A key innovation cited is Multi-Head Latent Attention (MLA), which significantly reduces inference costs by cutting KV cache usage by 93.3%. The report suggests that any improvements DeepSeek makes will likely be adopted by Western AI labs almost immediately.

SemiAnalysis also mentions that costs could fall another 5x by the end of the year, and that DeepSeek’s structure allows it to move quickly compared to larger, more bureaucratic AI labs. However, it notes that scaling up in the face of tightening U.S. export controls remains a challenge.

[Kons Anagnostopoulos @CaesarKons]

Extraordinary to say the least!

[Paul Graham @paulg]

This is an extraordinary video.

[Kons Anagnostopoulos @CaesarKons]

Impressive developments!

#GenAI

[Rowan Cheung @rowancheung]

It's been a huge week for AI, and it's only Tuesday.

Major developments from OpenAI's Advanced Voice, Friend, Meta SAM 2, Speechmatics, Perplexity, Midjourney, Runway, Leonardo, and NVIDIA.

Here's everything you need to know:

[Elon Musk @elonmusk]

Yeah, @Jack is right

[Kons Anagnostopoulos @CaesarKons]

@0x_ale Can't wait for more details Ale, ping me!

[George Mack @george__mack]

Underrated skill in 2024: Strategic Ignorance

In a world of infinite information, strategic ignorance has never been more valuable.

Every minute, 500 hours of video are uploaded to YouTube

Every day, 500 million tweets are uploaded to X

Every year, 67 million people die

The truth: The 24 hours allotted to you each day isn't even capable of consuming 0.0000001% of the world's events.

Fun hack:

When people mock you for not having an opinion on the new current thing: Start asking them about the current situation in Djibouti, Jezekstan and Eswatini.

You will not only reveal they aren't up to date on the current situation: They didn't even know those countries existed.

Note: Djibouti and Eswatini are real countries. Jezekstan is a country I made up. It's a fun way of testing if they will pretend to know about a made up country.

This test shows they don't actually care about ignorance -- it's just a proxy for shame and control about the new current thing.

When you look at the total world events that day -- we are all ignorant.

The good news is that there's different settings available in the ignorance video game: Low agency ignorance and High agency Ignorance.

1. Low Agency Ignorance

Low agency ignorance passively downloads the loudest voice on the news or For You algorithms.

It's an infinite abyss that never ends. "I don't know" is a sign of weakness.

Low agency ignorance may appear selfless, but in the pursuit of trying to keep on top of everything, it keeps on top of nothing.

2. High Agency Ignorance

High agency ignorance is strategic.

It is aware that attention is the scarcest asset in the age of information abundance.

High agency ignorance proactively seeks information from high value sources. It has firewalls against passive consumption of the world's worst events or social media drama.

"I don't know" is a sign of strength. High agency ignorance may appear selfish, but in the pursuit of strategic focus, it can actually have impact on something.

-----

Choose your character:

We are all in the ignorance gutter, but some are strategically looking at the stars.

Some are unstrategically getting angry about the worst event that happened today -- which they won't even remember one year from now.

-----

PS. If you enjoyed this, it's taken from my newsletter. Get all my best ideas in your inbox when you sign up. Link below.

[Kons Anagnostopoulos @CaesarKons]

@MadVincent666 @zaratustracrypt @MadLads @armaniferrante 🦾🦿

[Ethan Mollick @emollick]

Junior lawyers & legal outsourcing (LPOs) compared to AI in a New Zealand study: “Advanced models match or exceed human accuracy in determining legal issues. In speed, LLMs complete reviews in mere seconds… offering a staggering 99.97 percent reduction” https://t.co/Ok1NiA3ieh

[Kons Anagnostopoulos @CaesarKons]

@litocoen Anyone can share an invite code? Thanks

[Thanasis Vergoulis @vergoulis]

Since many of you are congratulating me for my inclusion in Clarivate's list for Highly Cited Researchers (https://t.co/Iu99slBK3c), after thanking u for your kind words, let me grab the opportunity to explain in a thread random reasons why this is *not* a big deal... (1/8)

[vitalik.eth @VitalikButerin]

Another maybe-controversial take of mine is that I don't think we should be enthusiastically pursuing large institutional capital at full speed. I'm actually kinda happy a lot of the ETFs are getting delayed. The ecosystem needs time to mature before we get even more attention.

[Antonios Karampatzos @karampatzos]

Για την υπόθεση των υποκλοπών

[Thanasis Koukakis @nasoskook]

Ο @VasLabropoulos αποκαλύπτει στα @ta_nea Σαββατοκύριακο πως «ενδείξεις για υποκλοπές ή ύποπτες παρακολουθήσεις υπήρξαν εντός του κυβερνητικού κόμματος, για υπουργό που φαίνεται να μην είχε και τόσο καλές σχέσεις με άλλους συνεργάτες του Πρωθυπουργού»

[Andy Baio @waxpancake]

The Stable Diffusion Discord has a channel devoted to spectacular failures and it’s full of gems

[Kons Anagnostopoulos @CaesarKons]

🤣🤣🤣

[Tom Warren @tomwarren]

How it started: How it's going:

[Balaji @balajis]

The Thiel Fellowship was attacked. Denounced. Vilified.

Yet it somehow found the founders of Ethereum, Figma, and more.

It proved its point. Some don’t need college.

Now, the fellowship’s cofounder tells you just how they proved the establishment wrong.
https://t.co/FNF0BAHMxO

[Kons Anagnostopoulos @CaesarKons]

@paulg Even more accurate on the art domain, where the true masterpieces stand the test of time.