aaraN retweetledi
EPF7 applications are open. Deadline is May 13.
If you want to work on core Ethereum protocol — client development, testing, specs, research — this is the program for you.
English
aaraN
1.2K posts
aaraN
@aAranVM
Senior Software Engineer(Rust) @UNCX_token| Bullish on Privacy Preserving Technologies , TEE | ZK | FHE
Katılım Mart 2011
1.8K Takip Edilen232 Takipçiler
aaraN retweetledi
Wrote up some flashcards and practice problems to help myself retain what @reinerpope taught.
Hope it's helpful to you too!
Suggest more below and I'll add them.
reiner-flashcards.vercel.app
Dwarkesh Patel@dwarkesh_sp
Did a very different format with @reinerpope – a blackboard lecture where he walks through how frontier LLMs are trained and served. It's shocking how much you can deduce about what the labs are doing from a handful of equations, public API prices, and some chalk. It’s a bit technical, but I encourage you to hang in there - it’s really worth it. There are less than a handful of people who understand the full stack of AI, from chip design to model architecture, as well as Reiner. It was a real delight to learn from him. Recommend watching this one on YouTube so you can see the chalkboard. 0:00:00 – How batch size affects token cost and speed 0:31:59 – How MoE models are laid out across GPU racks 0:47:02 – How pipeline parallelism spreads model layers across racks 1:03:27 – Why Ilya said, “As we now know, pipelining is not wise.” 1:18:49 – Because of RL, models may be 100x over-trained beyond Chinchilla-optimal 1:32:52 – Deducing long context memory costs from API pricing 2:03:52 – Convergent evolution between neural nets and cryptography
English
aaraN retweetledi
The best Gen Z podcaster in the world. He just keeps getting better, too.
Dwarkesh Patel@dwarkesh_sp
Did a very different format with @reinerpope – a blackboard lecture where he walks through how frontier LLMs are trained and served. It's shocking how much you can deduce about what the labs are doing from a handful of equations, public API prices, and some chalk. It’s a bit technical, but I encourage you to hang in there - it’s really worth it. There are less than a handful of people who understand the full stack of AI, from chip design to model architecture, as well as Reiner. It was a real delight to learn from him. Recommend watching this one on YouTube so you can see the chalkboard. 0:00:00 – How batch size affects token cost and speed 0:31:59 – How MoE models are laid out across GPU racks 0:47:02 – How pipeline parallelism spreads model layers across racks 1:03:27 – Why Ilya said, “As we now know, pipelining is not wise.” 1:18:49 – Because of RL, models may be 100x over-trained beyond Chinchilla-optimal 1:32:52 – Deducing long context memory costs from API pricing 2:03:52 – Convergent evolution between neural nets and cryptography
English
aaraN retweetledi
~195k CUs. 1kb of pubkey storage. 666 bytes of signature data. 0 protocol changes. QED. github.com/blueshift-gg/s…
Filipino
aaraN retweetledi
Today @humidifi is launching Aquarium, a transparent market making system built to give Solana projects deeper liquidity, tighter spreads, and clearer performance.
Where the water is deep, the fish gather.
Live on Solana ↓ 🐟
Solana@solana
Today @humidifi is launching Aquarium, a transparent market making system built to give Solana projects deeper liquidity, tighter spreads, and clearer performance. Where the water is deep, the fish gather. Live on Solana ↓ 🐟
English
aaraN retweetledi
aaraN retweetledi
Hello, today we are disclosing two critical trusted-party vulnerabilities in the LayerZero smart contracts. These issues allow the LayerZero team to completely bypass the Oracle and Relayer for most applications (including stargate).
prestwich.substack.com/p/zero-validat…
English
aaraN retweetledi
.@humidifi did $5B in volume last month.
Their secret? A 47 CU oracle update written in assembly.
With Anchor v2, you can debug it step by step.
English
aaraN retweetledi
The pre-release version of Anchor v2 is out.
v2 is over 90% smaller and 3-6x faster than v1, and represents months of work focused on speed, extensibility, and security. This is a major architectural change from a dense macro-based framework to a more easily extensible trait system, with security built in from day one.
Excited for teams to give it a try. Still a few rough edges, but please DM/comment with any feedback!
English
@datascraper69 @deanmlittle Trying to use spl pod and suddenly upstream build starts failing, not sure why anything that pulls in std is not gated behind a feature, same issue pulling the spl tkn22 crate(no way to disable std).
English
@deanmlittle im so done dude anything using spl-pod is forced to compile solana-zk-sdk because of this one line here #L9" target="_blank" rel="nofollow noopener">github.com/solana-program…. And what is this PodElGamalPubkey type? You've guessed it
English
aaraN retweetledi
BREAKING NEWS: For the first time ever, direct mapping *prevents* a bug! blog.asymmetric.re/solana-p-token…
Jokes aside, truly excellent work from both @asymmetric_re and @anza_xyz!
English
aaraN retweetledi
I asked Jensen: “2 out of the top 3 models in the world, Claude and Gemini, were trained on TPU. What does that mean for Nvidia going forward?”
After a long technical back and forth about what the right accelerator for AI looks like (see full episode), Jensen lays down the gauntlet:
Jensen: "Nvidia's computing stack is the best performance per TCO in the world, bar none. Dylan's InferenceMax is sitting out there for everybody to use, and TPU won't come, Trainium won't come. Nobody wants to show up."
“I'm not offended by other people using something else and trying things. If they don't try these other things, how would they know how good ours is?”
“There are always big claims, but look at the number of ASICs that have been canceled. Just because you're going to build an ASIC, you still have to build something better than Nvidia. It's not that easy building something better than Nvidia. Nvidia's got to be missing something seriously, because of our scale and our velocity.”
English
aaraN retweetledi
do aggregators improve execution quality on @solana venues? (spoiler: depends on venue)
one way to answer this question is to compare observed trades against "null" trades executed at a random timestamp
we captured ~280,000 SOL/USDC on-chain trades from @JupiterExchange over 8 hours, simulated execution 101 times (~28m sims) during a +-5 second interval (100ms buckets) and compared the median executed price to the same sized trade on binance SOL/USDT book
we also randomly selected 10,000 observed trades from the same distribution and simulated execution at a random timestamp (our "null" baseline)
if jupiter routing improves execution quality we should see best execution at T+0. and observed trades should have a better execution price than the null trades
if jupiter routing harms execution quality and/or propAMMs are spoofing quotes, we should see execution quality decline at T+0. and we should see the null trade at a random timestamp execute at better prices than observed trades
(all charts below except final notional-weighted chart are median harm vs binance with 0.1% tail excluded)
English
aaraN retweetledi
CU optimizations come with risks.
@_fel1x discusses a critical bug we found in p-token before mainnet, subtle enough to survive in a heavily scrutinized codebase.
English
aaraN retweetledi
1/ PropAMMs and the Next Chapter of Permissionless Market Structure
For years, onchain trading won on access, lost on execution. Some feel better performance requires moving backwards to centralized systems
PropAMMs on @solana are proving permissionless can outcompete centralized
English
aaraN retweetledi
aaraN retweetledi
Very happy that my first PR to @surfpool_sol has been merged! 🚀
Shoutout to @micaiah_txtx for the review and guidance throughout the process.
🔜 Coming soon: the ability to edit account data directly in runbooks!
github.com/solana-foundat…
English
aaraN retweetledi
aaraN retweetledi
The EU @EUdigitalID picked @NoirLang for their ZK identity app, and you should too!
github.com/eu-digital-ide…
English
aaraN retweetledi
We've open-sourced TIDX: Our blazing-fast indexer for chain data.
We use it to power every query in Tempo Explorer and Tempo Wallet. Now, it's available for any developer to build with.
Learn more:
Georgios Konstantopoulos@gakonst
English