Duratec $DUR.AX 1H24 results were very good - +28% Revenue YoY +56% Net Income YoY, however the stock tanked -16 %. I've done a short analysis of the current state and guidance, to see if it is reasonable or just overreaction 🧵 1/N
@SalamonTomas S big picture souhlasím, ale je to o vyvážení a např. v techu se už výuka a praxe extrémně vzdaluje a juniorům to dost komplikuje vyhlídky, nakonec i špičkové R&D se už dělá spíše v komerci a ne na akademické půdě. Rovnou říkám, že tento problém znám z pohledu praxe i univerzity.
@RihardJarc The workstation POV is interesting, but Claude Code is pretty memory hungry because it's source code is a mess and it's clear that the devs don't know what they're doing. Major upside potential still lies in larger context windows, thus higher cache even with TurboQuant-ish algs.
@tomcupr@thsottiaux I use both currently, often multiple instance over the same repo at once, I prefer Codex with plan mode for bigger tasks and Claude Code for single-shot ones, but it may depend on the type of tasks, 90 % of my work is about training custom transformer models and data eng.
@RhinoInsight This is clearly an entry-level job, likely for university students. It involves testing chatbots interconnected through the LangChain ecosystem. We use the exact same pipeline at our company, though we run it in Python rather than Node.js
@wilderko This is actually helpful even for humans, e.g., checking on quick comparison of cloud provider services or open data formats like Iceberg Vs Delta tables 😃
@ChShersh I don't know the precise structure of the data, however I would use Parquet file format together with daily partitioning thus decoded data are structured and compressed, you can leverage query engine like Polars to search in the data then. Or leverage Clickhouse for simplicity.
@govro12@evfcfaddict Definitely agree. For pure-value plays, you need to be extra selective as the majority of the companies won't turn the tide in the end.