Doing some research on marketing with AI tools.
The ideal state is an e2e workflow (social post, creative, landing page, whatever) with a human check at the very end.
But, so far, every single demo I've seen is just trash. Either I haven't come across a truly elite skill file, or marketing still requires a heavy handed human touch.
@ianlapham@benaratame Teammate from one of my jobs was working late one night. Said she’ll probably be up until like 9-10pm finishing up work. I told her if she’s staying up working, I’m staying up too.
Using AI to learn is a double edged sword.
On one hand, I’ve been able to accelerate learning by asking very precise questions.
On the other, I’ve dulled the “free thinking” that links orthogonal information, learned from reading primary sources, to my research question that grounds it in broader context.
@mikelauofficial@higgsfield_ai I’ll key an eye out on your page when it drops. DM me if you remember! Been watching you from the sidelines for a minute.
@defipoodle@SuccinctLabs@yukonresearch@zksecurityXYZ Yes, most proving systems are in the process of transitioning to PQ. Succinct is also working on it (our work is being used by Ethereum LeanVM). Starkware has been there for a while. I think ZKsync just recently upgraded their prover.
@Gem3a@AnthropicAI@SuccinctLabs Great recap! The AI industry has a lot to learn from cryptographers working on ZK in our industry. There's a lot of overlap.
@thunderking_ola Yes, depending on the density of technical information. More technically dense (cryptography) I’ll hand write 60%+. For integrations/eng I’ll lean more on LLMs. But requires thoughtful skill files to get it closer to where I want it.
B player reporting for duty 👋😂
First off, Georgios is a G, he’s done a lot for Ethereum, and while this is a response to his tweet, it is more about showing another way to view things.
This is a long one, so strap in:
I agree that inspiring the next generation of serious engineers matters and so does accountability, and maybe we should take a few lessons from Apple on presenting the work/making things more visible. BUT my opinion as someone in Ethereum, or in sharper vernacular, “a paid shill,” doesn’t matter here. If we’re really going to talk about this, let’s just go through the programming for Devcon and let everybody else decide.
Let’s start with a call to formally verify things, which would include the awesome thing Georgios released this week, Solar.
@dhsorens’ session explored using LLMs and formal verification to write the EVM directly in RISC-V assembly.
https://t.co/EGqderXXxf
Mooly presented agents building Aave V4 spokes against formal specifications with the Certora Prover checking if the implementation satisfies specifications. This is a serious answer to the question of what Ethereum engineering looks like as AI gets better at writing code.
https://t.co/vlcnQPA7Qm
There is also the next generation of developers working on key stacks.
Arsh and Parth are showing off their fellowship work on Pureth in Reth.
https://t.co/pQ6T7nPGsj
Ayush is presenting work bringing Geth’s FilterMaps approach to Reth, to address the growing cost of log queries.
https://t.co/HwPcmlIUCp
There’s also tons for people who want Ethereum to run faster and ship more reliably.
@nero_eth walked through the engineering behind L1 scaling and the coordination needed to turn research into changes faster than ever before, ready for the AI age.
https://t.co/9wyRQjOD9M
Mark is tackling the storage problem that comes with much higher throughput.
https://t.co/ZneEiR0LCK
On accountability, some of the most interesting sessions are about making technical claims easier to check.
Seulkee’s independent executable EVM specification in Lean comes with public source and a case study describing discrepancies found and reported upstream.
https://t.co/qXhouRJu0m
@fradamt and Cody’s session on verifying zkEVMs explicitly discusses what credible correctness claims require and where residual risk remains.
Towards Formally Verified zkEVMs: https://t.co/QE2Tg8xSUV
And the AI work extends into the research process itself.
Rose is presenting an approach to letting agents propose ZK optimisations while checking candidates, benchmarking them and feeding the results back into the next iteration.
Autoresearch for ZK Optimization:
https://t.co/1euSaofOqU
@ncsgy was presenting the search for signatures suited to Ethereum’s quantum security needs, including an open competition using AI driven search.
https://t.co/EzX8wvPsT3
There’s practical work for people trying to make this usable too!
A workshop on verifying blockchain responses and simulating transactions in the browser.
https://t.co/ovlkvGoql4
A production engineering session covering accounts!
https://t.co/HxoTvYBz2s
And to keep the opening programme separate from the rest of the conference, it included:
Post Quantum Transition, Justin and Kev
https://t.co/FmnwzHh3HN
Reducing the Trust Surface, Derek and Mooly
https://t.co/83iu7IJT1s
End to End Privacy, Nico and Luca
https://t.co/xgHHI0qPiQ
None of these links obliges anyone to enjoy a keynote ofc but there are hard problems here that could occupy the minds of exceptional engineers (and their agents!)
I do agree that we should make our ambition much more visible, and i hope to do my part.
Ethereum.