Alec
246 posts


why dspy usually wastes your time (and when it doesn't)
the question: "should i use dspy for prompt optimization? it seems like the perfect tool for improving my rag system."
the answer: dspy is great for very specific, well-defined tasks. but for most rag systems, it's a distraction from what actually moves the needle.
here's the reality: dspy works best when you have a clear classification task with measurable accuracy. think 35-class categorization where you can hill-climb on a single metric. but most rag problems aren't like that.
when i'm building a system to extract sales insights from transcripts, i don't have a dataset of "here are all the sales insights." the real work is extracting everything, hand-labeling some examples, and building intuition about what users actually need.
your product isn't just a prompt - it includes how you collect feedback, set expectations in the ui, handle data extraction, and represent chunks in context. if you spend time looking at how the model makes mistakes and what users are asking for, you'll make much more progress improving the product as a whole.
the one place dspy shines: llm-as-judge scenarios. if you have a tonality or factuality evaluation you really care about, it makes sense to label 100 examples yourself and then use prompt optimization tools to create your own judge that aligns with your grades.
but for most rag systems? you're better off manually tweaking prompts while building intuition about your specific use case. that intuition will guide better decisions across your entire system architecture.
English

An end of an era.
When @Whaazz approached me to compete again at the end of 2022 together with @Raiku_Wow and @JuhaniHalme, I never thought we’d find as much success as we did.
I wanted to be the only player to win a major in both NA & EU, and with that goal being well
Echo@EchoGuild
Today, one of the GOATs steps down from competition THANK YOU Chan 🙏 for giving so much to WoW Echo is honored to have been part of your incredible 12-year career🏆 Good luck in your next chapter!
English

@ElliottVenczel @Metaphors__ I remember that gladiatorlossa (the thing that announced spells per voice) was absolutely ridiculed even at the end of MOP. Weakauras are far beyond anything like that (also an old man👴)
English

@Metaphors__ Tournament players practiced without addons, true. However, many ladder players used UIs like this even the original WoTLK. Nothing wrong with people enjoying Cata and using all the tools at their disposal. (Also an old man 👴)
English

To celebrate DOOM's 30th anniversary we're giving you a chance to win a GeForce RTX 4090 + Custom DOOM backplate!
Comment #DOOM30 + #RTX500 below to enter.
Goodluck slayers → nvidia.com/en-us/geforce/…

English

Ding! I'm thrilled to announce that today is my first day as the newest Lead Game Producer on @Warcraft! I truly cannot believe it, but I'm so stoked that I get to keep making this game I love. Let's go!
First order of business, let's talk Turnip Demon Hunters @WarcraftDevs

English

@Jahvilli Love wow and think it has great potential + it's much more complex (and the top players are much more skilled) than people give it credit for. It's hard to follow & casuals/competitors have very different & at times clashing perspectives of the game making it a hard business case
English

WoW Arena is by far the closest thing to majority of the most successful and mainstream sports in the world.
Every class represents a specific playstyle and role to their team, and the skillcap is limitless (when developers are doing their job and consistently making effective changes to the game).
Its very clear that WoW arena has the potential to be the greatest esport ever created and it was very close to achieving that for a time.
Unfortunately Blizzard's lack of effort/ego/ incompetance has been unable to keep the game from growing into what its always meant to be.
Bonus: 1v1s have always been to closest thing to MMA that any e-sport has ever seen (as well as SSB Melee), which is an incredibly fast growing sport that will most likely break the top 5 in terms of popularity soon.
Liquid Eiya@eiyawow
WoW fans are HUNGRY Everyone sitting on the floor just to watch
English

@ferconx @WarcraftDevs None of your arguments adress the underlying issues:
If there is a warlock & the enemy team doesn't have curse dispell, then one of the dps can never crit. Doesn't matter who they hit or what they do, one of them can never crit.
English

@ABCalec @WarcraftDevs I think its ok its not like they can put it on all targets and also how is this fair to train 1 person for 3 mins and not be able to use something to make class feel more safe in Exchange. Its not like classes also cant dispel it plus there are dps who can dispel plus other stuff
English

Would you reconsider how amplify cusre (weakness) works in pvp? Most shuffle lobbies/3scomps do not have curse dispell & people are unaware/uncoordinated.
0% Crit chance is very opressive, considering 20s cd / 20s duration @WarcraftDevs
English

@pizzadawgDK Has anyone done the math to compare Ret (w new talents) to bear form?
English
Alec retweetet


@ElliottVenczel @Dyllbar1 Yes but they were perceived as bad at some point so now noone but them gets to have fun
English














