7063
Ryan

@ryanconnor #7063

Blockworks Research. @_RyanRConnor on twitter/X.
132 Follower 194 Following
To the AI practitioners in the channel, what are the most common precision formats used today? Is FP32 still dominant?
/AI
Question - what are the best estimate or training, inference, and hardware specs/cost for Sora?
/AI
Prediction: Hivemapper data becomes a critical tool for sophisticated Parcl traders and whales
Restaking question:

In a fully permissionless modular ecosystem system, what's to stop restaking from evolving into a MIM or UST scenario, where a gas token is rehypothecated indefinitely, leading to destabilization at the security layer?
Recasting here so Solana folks can chime in. Thoughts?
Question for Eth ecosystem. Would you say modular systems are more prone to tail risk associated with restaking than integrated systems?

Eth is explicitly relying on many trust networks for scale, whereas integrated systems are not, so many layers of rehypothecation are possible, and therefore tail risk is larger.
I'm not legal expert - there's no mention of crypto, but are there implications for it? One would think so...

https://techcrunch.com/2024/01/11/google-will-allow-more-real-money-games-on-the-play-store/
Good data from Coinbase - what does it say about these holders of these tokens and the ecosystem?
Thinking through EigenLayer.

Is it fair to say that EigenLayer staked ETH...

1. Increases yield to EL ETH stakers by less than 2x
2. Increases smart contract risk AND slashing risk by more than 2x

?
Has anyone ever seen a chart of the count of ERC v SPL tokens listed on coinbase over time? Any idea where this data would reside?
What is the podcast version of Techmeme?
This is the first time I've been 'wow'd' by an LLM in months.

Bravo Grok.
/AI
1/ Crypto is hyper-focused on RWAs, but what’s more interesting are NNAs - Net New Assets.

I’ll loosely define NNAs here as assets which are prohibitively expensive due to issuance or regulatory costs or assets that are outright outlawed.
LGA is infinitely superior to JFK
I use perplexity when I need to get up to speed fast and can't be wrong
/AI
Anyone know what's Solana's near 100% tx success rate since early October?
This is pretty awesome... permissionless GPUs via io.net, Render/Solana and Filecoin.

This is likely the only DePIN that's immediately useful to both the supply and demand side of the network, and prime for near-term, real world, non-crypto native adoption.
/AI
Token22 question - what do confidential txs look like on a block explorer?
Adding emotional context to prompts leads to increased model performance and suggests emotional intelligence…

Also bullish for the TherapistGPT use case

https://open.substack.com/pub/aimodels/p/telling-gpt-4-youre-scared-or-under?r=ktlru&utm_medium=ios&utm_campaign=post
/AI
1/ The impediments to *enterprise* AI adoption have (I think) been largely eliminated over the past 10ish months - hence the release of Copilot from Microsoft. Below is quick list of problems and their solutions. Am I missing any? Where am I wrong?
/AI
1/ Used GPT3.5 to do a bottom up TAM analysis

In short, the US gov provides a list and description of every task in the economy, all 19k of them

We use gpt-3.5-turbo to read every task description, tell me “yes” or “no” if the task can be automated, then map that task to a wage and sum it per profession
/AI
Cloudflare pulling no punches
- recommendations for Okta
- recommendations for Okta customers
- “we’re hiring!” wink at the end for frustrated Okta talent

https://blog.cloudflare.com/how-cloudflare-mitigated-yet-another-okta-compromise/
Only 1/3 of postdocs are using ChatGPT, while adoption for software developers is probably over 90%.

Speaks to the slow moving nature of academia imo

https://www.nature.com/articles/d41586-023-03235-8
/AI