Log inSign up
Robert Sasu | dev/acc profile banner
user avatar

Robert Sasu | dev/acc

Multiversᕽ
@SasuRobert
I belong to Jesus. Core developer #MultiversX
Joined August 2014
726
Following
22.7K
Followers
RepliesArticlesMedia
  • user avatar
    Robert Sasu | dev/acc
    Multiversᕽ
    @SasuRobert
    16h
    I have been using LLM as verifier from early 2026. I have been using and integrating this github in the last few months ad well, also making different variants of it with specific verification models. All my agents / workflows are using LLM as verifier. Any single task uses
    user avatar
    Jacky Kwok
    @jackyk02
    Aug 17
    Scaling self-verification with DeepSeek V4 Flash beats Claude Fable 5 on Terminal-Bench 2.1, while being 11x cheaper 💰 As open-source models become more capable, they can now generate large numbers of high-quality candidate solutions and verify their own outputs at very low
    Image
  • user avatar
    Robert Sasu | dev/acc
    Multiversᕽ
    @SasuRobert
    16h
    MultiversX is THE Internet Scale blockchain. The complete sharding model, where interacting with shard X or Y is seamless and the same as interacting with any shards. It is seamless at all levels.
  • user avatar
    Robert Sasu | dev/acc
    Multiversᕽ
    @SasuRobert
    16h
    When the chain works 24/7 , sometimes you get late night calls of issues. We had a few ones in history. That is why we use TLAs, formal verification, non stop audit, agents monitoring the network, auditing the code. 🧑‍💻
    Image
    CodeMultiversX
    Multiversᕽ
    00:00
    CodeMultiversX
    Multiversᕽ
  • user avatar
    Robert Sasu | dev/acc
    Multiversᕽ
    @SasuRobert
    Aug 17
    this is the model to look at. and it also will drive the price of local hardware up, and probably the price of the frontier models down. the open source models are a serious piece of work, even when it comes to their smaller size. this year was simply amazing when it comes to
    user avatar
    Jun Song
    @jun_song
    Aug 16
    My honest takeaways after using Qwen3.8-27b on real world tasks for a few days: Grok-4.6 and Muse Spark-1.2 are definitely benchmaxxed. 27b clearly outperforms both in practice. It even beats Opus-5 on some tasks, especially now that Opus feels severely nerfed and dumber.
  • user avatar
    Robert Sasu | dev/acc
    Multiversᕽ
    @SasuRobert
    Aug 17
    guess what software is most vulnerable in the terms of AI Agents ? All those public institutions and authorities code, which was not done by the best engineers, but by "companies" which made it cheaper. just like in romania or in other states, those official government sites
    user avatar
    Laura Shin
    @laurashin
    Aug 17
    🔓 A breach at France's tax authority exposed personal and tax data for 678,000 people — in the country with the world's most violent crypto attacks.