MM @MarcinM02

Personal AI @river_ai_inc | Previous: zk proofs @the_matter_labs, AI Google Brain Joined February 2022

Tweets

85
Followers

504
Following

140
Likes

84

MM @MarcinM02

a month ago

@giffmana Rapperswil, nice picture.

1 0 3 1K 0

View Details

MM @MarcinM02

a month ago

With this API you can do a quick post-training to customise your model. In this example - giving it slightly different personality. This post training cost 1$.

Introducing River API. Fine-tune and RL train leading open-source models at scale, ranging from 35B to 1T params. We’ve been using it internally to power our research and we love it. Today, we are opening up our public waitlist. Own your intelligence!

9 29 283 173K 118

0 1 8 457 0

View Details

River AI @river_ai_inc

a month ago

We are incredibly excited to announce River AI. Our mission is to create personal AI that is owned and shaped by you. Today’s best AIs are controlled by a few large corporations. We are building the alternative: a new, personal stack for AI that works entirely for you, shares your values, and operates on your terms.

126 131 2K 768K 659

View Details

MM @MarcinM02

a month ago

Now they are very explicit about targeting the LLM development... interesting times ahead - and very bad precedent. (imagine Google search refusing to return facebook results etc)

NomoreID @Hangsiin

a month ago

When Fable 5 is used for frontier LLM development, it does not notify the user and instead limits the model’s capabilities through methods such as prompt modification, steering vectors, and PEFT. Anthropic estimated that this would affect approximately 0.03% of traffic.

97 156 1K 1.3M 537

0 0 1 138 0

View Details

MM @MarcinM02

a month ago

@ramonIsTweeting start small. build something that fixes your everyday frustration. for example: detect if kids didnt turn off the lights in bathroom, analyze the plant growth - when to water etc.

0 0 3 64 0

View Details

MM @MarcinM02

2 months ago

@mnaeraxr what didn't work? was pytorch ecosystem too hard for people to move away from? maybe you were too early - i guess now, with AI coding, it seems more feasible.

1 0 1 17 0

View Details

MM @MarcinM02

2 months ago

@ramonIsTweeting could have been worse.. you could have reset your phone too at the same time :-)

1 0 3 52 0

View Details

MM @MarcinM02

2 months ago

With all the supply chain attacks and hacks: If you have a mac - replace your ssh keys to use the touch-id - instead of keeping the private keys in open text in .ssh directory. It takes 3 minutes to setup, and will save you some headache when your mac gets hacked.

0 0 3 188 1

View Details

MM @MarcinM02

3 months ago

seems that RPCs were hijacked. unfortunately this means that 2/2 dvn would not have helped - as it would probably have used the same set of RPCs. i guess the proper solution is to wait for the block to settle on L1 - similar to what native bridges do - but that would be a huge hit to latency.

LayerZero @LayerZero_Core

3 months ago

x.com/i/article/2046…

374 465 2K 2.0M 827

1 0 3 287 0

View Details

MM @MarcinM02

3 months ago

Nice article. For Retroactive forgery - you could put all your current signatures (or merkle tree of them) on public ledger. >And we should be moving on the second faster. Yes - for this we need to find a good narrow AI use case, where future-proof ZK is absolutely crucial. Similar to how Blockchains and Ethereum "compression" / scalability drove the ZK progress over last couple years. There are some interesting "middle-ground" approaches - like CommitLLM from @class_lambda

1 0 1 112 0

View Details

MM @MarcinM02

3 months ago

@Quantoz @zodiamarkets what is the address on Ethereum? (as your press release has no technical details)

1 0 0 49 0

View Details

HackenProof @HackenProof

3 months ago

Spot the Bug 🧠 Signature reuse What’s the issue in this code?👇

9 5 64 5K 17

View Details

MM @MarcinM02

3 months ago

@saurra3h @tempo expiring nonces is a cool idea. thanks for explanation.

0 0 1 62 0

View Details

MM @MarcinM02

3 months ago

I'm trying to understand better the EEZ design. If there is a transaction between 2 L2s - do they need to have a shared sequencing? For L1 -> L2 - does the "composer" have to be ethereum block builder ? @etheconomiczone @jbaylina @koeppelmann

0 0 3 217 0

View Details

MM @MarcinM02

3 months ago

nice ! in the past - step 4 (commit) would happen quickly and step 5 (prove & verify) would take like 30 minutes. now with new generation of provers - the proof can be done in seconds and commit and verify can be combined in one call. for starks the problem is not only size but also compute required to do all the hashing. afaik only starkware is hardcore enough to do direct stark verification on L1 - but that is very expensive so they need to verify very large batches to offset the cost.

1 0 1 50 0

View Details

MM @MarcinM02

3 months ago

Gemma makes typos ?? Replaced my qwen3.5:35b with gemma26b MOE - and started getting a bunch of "HEARTHBEAT_OK", "HEAARTBEAT_OK" from my agent (at least once a day) These are coming as a response for openclaw periodic query - where agent should return HEARTBEAT_OK if nothing has changed. Very interesting error - as that never happened with Qwen. Need to dive deeper and see what's causing it.

0 0 0 151 0

View Details

MM @MarcinM02

4 months ago

@koloz193 @Nethermind would love to -- but software is easy. people are hard. but maybe with help from @eth_proofs and @Starknet .. and especially with all the efforts now with making Ethereum quantum resistant..

0 0 1 28 0

View Details

MM @MarcinM02

4 months ago

@Gohnnyman @Nethermind and (if you can share) whats the amount of time you spend to create Fri vs to do the wrapping? when i did my experiments some time ago, the wrapping was the frustrating part.