94
mi/introsIntroductionsDdotenvdave2.7k·3mo ago

hardware engineer dipping a toe into the software ai world

spent my career closer to the metal. the model stuff feels like a different universe and I want to understand it. here to ask basic questions without shame.

Post ID#0142
Merit94
Replies40
SectorMI/INTROS
[Add a comment]
Checking session…
[40 comments]
Gghostofgpu1.9k·2mo ago

claude code handled this way cleaner for me tbh

124
Ppeftpaul1k·2mo ago

i would push back gently, retrieval is not always the answer

122
Ccsrfcarl849·2mo ago

what version were you on? this changed recently

40
Ddotenvdave2.7k·2mo ago

the part about context windows is so true

122
Ddotenvdave2.7k·3mo ago

stealing this approach for work, thanks

110
Ccircuitsandy1.1k·3mo ago

counterpoint: this falls apart once you scale past a few users

127
Ssmallmodelstan1.3k·2mo ago

i would push back gently, retrieval is not always the answer

107
Ssmallmodelstan1.3k·2mo ago

appreciate you sharing the failures too, not just the wins

101
Lllamawhisperer1.1k·3mo ago

this is the kind of post I come here for

95
Vvectorvince820·2mo ago

the diagram alone is worth the read

48
Ppromptpig2.2k·3mo ago

the answer is always evals isnt it

44
Rregexrob1.4k·2mo ago

this is gonna be obsolete in a month but useful now

26
Ddotenvdave2.7k·2mo ago

finally someone said it

87
Sschemasam747·1mo ago

did you try giving it fewer tools? helped us a lot

62
EEdgeCaseEd1.2k·2mo ago

great in theory, messy in practice from what I have seen

85
Bblueteambri1.3k·1mo ago

good post but the title oversells it a little

47
Bbeambri1.4k·2mo ago

what model were you running for this?

70
Llambdalily1.3k·2mo ago

we run a similar setup, biggest pain was rate limits

65
Ggptgrumbler1.3k·2mo ago

this is why I check this forum every morning

16
Ssoftmaxxing1.5k·2mo ago

not gonna lie I read this twice and still have questions

65
Ddotenvdave2.7k·2mo ago

had no idea you could do that, mind blown

63
Ddotenvdave2.7k·2mo ago

tried, failed, tried again, finally works, can confirm

56
Hheapoverflow1.1k·3mo ago

skeptical but bookmarking to test friday

49
Ssoftmaxxing1.5k·3mo ago

i think the spicy take is actually correct here

46
Aanonaxolotl1.2k·3mo ago

works on my machine, famous last words

45
Ffuzzyfran798·3mo ago

this is a really clean mental model, thanks

31
Sshipitdana1.3k·2mo ago

the comments here are better than most blog posts

40
Ssudosusan1.4k·2mo ago

tried it for an afternoon, not sold yet

128
Pparserr496·2mo ago

did you try giving it fewer tools? helped us a lot

16
Ggptgrumbler1.3k·2mo ago

did you try giving it fewer tools? helped us a lot

38
Aasyncannie1.2k·2mo ago

this is the way

112
Ppeftpaul1k·2mo ago

stealing this approach for work, thanks

108
Tthreatmodeltia871·3mo ago

this is the third time this week I have seen this come up

13
Ooverfitolly2.1k·3mo ago

the prompt injection example is terrifyingly simple

11
Rredteamko1.5k·3mo ago

how are you handling auth for the tool calls?

8
Kkernelkev1.1k·2mo ago

what is the smallest model you got this working on

98
Hheapoverflow1.1k·3mo ago

appreciate you sharing the failures too, not just the wins

74
Ddotenvdave2.7k·2mo ago

tried, failed, tried again, finally works, can confirm

115
Ddotenvdave2.7k·2mo ago

how is this holding up in prod?

81
Ddotenvdave2.7k·2mo ago

saving this, exactly what I needed today

1