Kostas Oreopoulos @koreopoulos

Civil Engineer, Mathematician, mad about chess, music, computers and new stuff Thessaloniki Joined May 2009

Tweets

800
Followers

148
Following

461
Likes

843

Kostas Oreopoulos @koreopoulos

15 hours ago

@Kalamaraki87 Ούτε το 30%. Ακόμη δεν ξέρει ποιος είναι ποιος και θέλει και λίγο αυτοπεποίθηση.

0 0 2 48 0

View Details

Kostas Oreopoulos @koreopoulos

15 hours ago

@NiklasCcr If you see the game, this pass does not count as an assist, but it actually is half the goal. He is good. Very good.

0 0 0 13 0

View Details

Kostas Oreopoulos @koreopoulos

7 days ago

It is funny that you value 1v1 that much. For my taste of football, Manduake is not allowed to enter the pitch. I do not care if you get 1/20 in 1v1. Football is not that kind of sport. keeping possession, being aware of the players around you, being fast and having stamina are important

0 0 1 513 0

View Details

Kostas Oreopoulos @koreopoulos

3 weeks ago

@unclebobmartin So a big skill-job, is split to pieces by the leader and we train the swarm to optimize that.

0 0 0 6 0

View Details

Kostas Oreopoulos @koreopoulos

3 weeks ago

@unclebobmartin This is an idea-project I have in my head for a lot of time with a twist, which is I want the subagents to be local or cheap and the leaders to be smarter. Then mini skills documents will drive the subagents. The idea is to train the swarm and optimize on skill, task split etc

1 0 0 142 0

View Details

Kostas Oreopoulos @koreopoulos

3 weeks ago

@Businessmentor8 Περιμένω να μου δείξτε που το μοντέλο στέλνει δεδομένα οπουδήποτε. Κανένα μοντέλο δεν μπορεί να το κάνει αυτό. Μόνο ο host και υπάρχουν. Host παντού στον κόσμο. Αμερικανοί, Γάλλοι, Γερμανοί.

0 0 0 8 0

View Details

Kostas Oreopoulos @koreopoulos

3 weeks ago

@Businessmentor8 Γράφεται για Κίνα, για δεδομένα που στέλνονται για open source που δεν είναι. Ότι ναναι. Μπείτε στο github και υπάρχουν τα πάντα. Ο κώδικας, τα βάρη. Τα πάντα. Δεν ξέρετε το αντικείμενο, κακό δεν είναι. Δεν καταλαβαίνεις την απάντηση. Δεκτό. Αλλά έχεις και ύφος.

1 0 0 11 0

View Details

Kostas Oreopoulos @koreopoulos

3 weeks ago

@thsottiaux Let's start with something small. Save goals with names, so I do not have to type again and again the same workflows. Then make goals have a structure. Like a graph. skill A -> skill B -> repeat until... Etc and save by name.

0 0 0 9 0

View Details

Kostas Oreopoulos @koreopoulos

3 weeks ago

@omkarships It is and it's much harder for new juniors because it's harder to put the effort to do it. If you do not know how good code looks like, then you will never stop bad code from taking over. After 40 years of coding, I still leave 1-2 hours to refresh coding skills, and learn new

0 0 0 22 0

View Details

Kostas Oreopoulos @koreopoulos

3 weeks ago

@unclebobmartin My only requirement about tests is that every path should be tested. Happy path and error path. That is enough I think. The order of things is a human decipline to avoid skipping it. My reviewer always starts with checking test implementation. Backend, frontend, E2E

0 0 0 131 0

View Details

Kostas Oreopoulos @koreopoulos

3 weeks ago

@unclebobmartin What I found out using various ways of vibe coding is that, telling the AI to use for example solid principles does not play an important role. What does work is select architectures and restrict the ai to follow it. For example tell it to use modular monolith is more effective

0 0 0 16 0

View Details

Kostas Oreopoulos @koreopoulos

4 weeks ago

@mattpocockuk I am using okf spec. All skills refer to it. Workflows specified in there, code spec, to test or not. It knows what to document and reuse and the main skill uses it to navigate.

0 0 0 39 0

View Details

Kostas Oreopoulos @koreopoulos

a month ago

@patloeber When compared to GPT 5.6 Luna, it's neither cheaper nor better; it is just evidence you are late to the party. I moved from the 200 plan to OpenAI 100 plan and get much more done without hitting limits, which was the case after 3.5 Flash. I am sure you are not happy either

0 0 0 11 0

View Details

Kostas Oreopoulos @koreopoulos

a month ago

@OfficialLoganK It's hard to win a race when you have so many fronts. On the other hand releasing a model that is more expensive than Luna and worse than it, is something you do with a heavy heart.

0 0 0 735 0

View Details

Kostas Oreopoulos @koreopoulos

a month ago

@GoogleDeepMind Worse than Luna!!!!!???? Really???

0 0 0 20 0

View Details

Kostas Oreopoulos @koreopoulos

a month ago

@pamaur22_22 @nikosTINOS Βρε άσχετε, γνωρίζεις τους κανόνες της WTA για την συμμετοχή αθλητριών στα 250αρια;

2 0 1 55 0

View Details

Kostas Oreopoulos @koreopoulos

a month ago

@MadridPreeti Blind I guess and a sore loser

0 0 0 1 0

View Details

Kostas Oreopoulos @koreopoulos

a month ago

@rasbt I am also curious why people head for the big guns. In a very well-specified spec problem, unless you need some clever algorithm solution and planning, reaching for the lowest and cheapest model "tests" your specification. Current models will do usual stuff without issues.