acidbird 🦇🔊🏴‍☠️

degen since 2017, team at @vfat_io | member at D420 & Echo | #degenpower, | always DYOR, NFA

[acidbird 🦇🔊🏴‍☠️ @_acidbird_]

1/ @Project0 @macbrennan_cc

Your statement here:
https://t.co/pFNjMnXAxP
People lost real money.

This was not a smart contract bug.

This was an admin key compromise enabled by serious operational security failures.

[acidbird 🦇🔊🏴‍☠️ @_acidbird_]

@DriftProtocol @project0 didn't share addresses of affected users

[Pearl Research Labs @prlnet]

https://t.co/cvDuRl6iQb

[MegaETH @megaeth]

Every onchain app gives you a new wallet. New balances. More problems.

MOSS is the unified onchain account experience.

One Wallet. Every App. Your Rules.

[acidbird 🦇🔊🏴‍☠️ @_acidbird_]

@mnstr honestly, i am happy af guys, thank you!

[acidbird 🦇🔊🏴‍☠️ @_acidbird_]

@andy8052 happy to help, thx for stimmy! couldn't use the not so great USDM airdrop better honestly 🫡

[acidbird 🦇🔊🏴‍☠️ @_acidbird_]

😅

[Monster @mnstr]

Terminal rewards are live for Monster users! To celebrate we will be giving away free packs randomly all day.

In order to be eligible, quote tweet this post with a pull you did today and make sure to include your username!

Good luck and happy ripping

[Openτensor Foundaτion @opentensor]

Access to intelligence should not depend on a handful of companies or governments.

This is why open, decentralized, permissionless AI matters.

This is why Bittensor matters.

[ether.fi @ether_fi]

1/ Moving money just got a whole lot easier. You can now send fiat in 6 different currencies across 120+ countries.

Pay for your rent, your bills, send money back to your home country and more — all with https://t.co/gbHcksy7eA.

[Max Crypto @MaxCrypto]

Crypto prices since Trump took office:

$BTC: -44%
$ETH: -49%
$XRP: -68%
$SOL: -77%
$DOGE: -79%
$AVAX: -82%
$ADA: -85%
$SUI: -86%
$ENA: -92%
$APT: -93%
$TRUMP: -97.7%
$MELANIA: -99.5%

Thank you, Mr. President.

[The Kobeissi Letter @KobeissiLetter]

It's official.

MicroStrategy, $MSTR, is now facing its biggest unrealized loss in history, at -$10.8 billion.

In other words, after 6 years of buying Bitcoin, the company is now down -17% on its position.

By comparison, the S&P 500 is up +116% over this same timeframe.

Since MicroStrategy sold 32 Bitcoin at $77,135 per coin, their positions has lost -$11.8 billion in value.

This puts MicroStrategy's stock, $MSTR, down -77% since its record high.

Bear market is an understatement.

[ether.fi @ether_fi]

Founders realizing they can get 15% cashback on their Claude bill

[Igor Yuzovitskiy @igoryuzo]

Hackathon, but make it underground.

[ether.fi @ether_fi]

New day, new all-time high in spend volume. Thank you for choosing https://t.co/gbHcksy7eA.

Onwards.

[OFFSHORE PROTOCOL @OffshoreOnMega]

Offshore Protocol season 2 is now live:

Season 1 paid out over $1M USDm in rewards in just 14 days.

S2 introduces new features, more progression, and more action.

Adapt and stay on top of the underground, or end up behind bars.

Build your empire now: https://t.co/og6AMZNkwu

[Omri Weinstein @WeinsteinOmri]

~90% of AI’s energy is devoted to one core, expensive operation: matrix multiplication.

This partly explains why so much research and engineering on LLM inference targets reducing the IO and complexity of MatMuls: O(n³) FLOPs to process only O(n²) data. When n = 30K, the gap is enormous...

A question I find fascinating is whether there exist alternative bilinear operators f(W, X) for “mixing” LLM tokens, that are as “expressive” as MatMul for human-related tasks, but require ≪ n³ FLOPs (and ideally O(n²) IOs) to evaluate.

There have been significant attempts in this direction, starting with Google’s FNet, and continuing with a long line of work on other bilinear operators: tensor algebras, Hadamard products, fast polynomial multiplication / convolutions, Butterfly matrices and other structured operators.

These approaches demonstrated promising results on some benchmarks, but when scaled to SoTA LLMs, they all seem to suffer from substantial accuracy degradation.

Is there something “holy” about MatMul in deep learning after all?

What seems to distinguish MatMul from many other bilinear forms is that it encodes arbitrary change of basis.

This is a qualitative property: A fundamental reason why ML algorithms work is the premise that real-world datasets have underlying structure – a basis/representation in which the data becomes sparse, invariant, clustered, low-dimensional, etc.

For example, changing to Fourier bases such as the DFT reveals frequencies. The DCT compresses images. Wavelets localize scale. Learned projections encode invariant subspaces, task-relevant metrics, and directions of variation. Attention scores depend on the learned embedding of LLM tokens, which are a choice of basis.

Oversimplifying, learning can be viewed as the optimization problem of finding the “right” basis to represent the data. MatMuls are the computational primitive that switches between these bases, i.e. linear projections.

By contrast, other bilinear operators such as elementwise multiplication, or convolution / polynomial multiplication p·q, do not obviously have this same functional interpretation. They impose structure, but they do not represent an arbitrary change of coordinates.

This observation may help clarify what properties would need to be preserved if we are ever to find cheaper bilinear alternatives to MatMuls in deep learning.

I still find the above question intriguing, both in theory and practice…

[DCF GOD @dcfgod]

But the token is all time low

[optimist @optimist]

https://t.co/2MXNHDa5d9

[Nous Research @NousResearch]

Congrats to our team, users, and supporters on the 150,000 GitHub star milestone!

[acidbird 🦇🔊🏴‍☠️ @_acidbird_]

@OffshoreOnMega Swiss for the win gang!

[OFFSHORE PROTOCOL @OffshoreOnMega]

Factions are live.

Pick a crew, stake $DIRTY together, and earn bigger cycle rewards.