paul lodder

96 posts

paul lodder

paul lodder

@paul__lodder

i build agentic AI systems and TB-scale weather data infrastructure co-founder https://t.co/RdiA1PGIEt

amsterdam Katılım Ekim 2022
379 Takip Edilen76 Takipçiler
paul lodder
paul lodder@paul__lodder·
is it me or is @uptimerobot surprisingly flaky for a status monitoring service?
English
1
0
0
237
Simon Prammer
Simon Prammer@SimonNom1·
Is there no way to do a Coding setup with Claude Code on Android? There seems to be no legit way to actually use it on Android. Claude App doesn't have CC, CC Web link simply doesn't work. @claudeai
English
1
0
0
71
paul lodder
paul lodder@paul__lodder·
a pattern i use with claude code/codex a lot: have it write a one-off script to diagnose, reproduce, and fix an error/bug and only then apply it into the actual codebase it can iterate in isolation which reduces the cognitive load for the agent
English
1
0
1
92
paul lodder
paul lodder@paul__lodder·
(me coping with my ever-growing dependency on llms to write code for me)
English
0
0
0
40
paul lodder
paul lodder@paul__lodder·
i wonder: when cars were invented, were people worried they'd forget how to ride a horse?
English
1
0
0
49
paul lodder
paul lodder@paul__lodder·
i wonder how many codex features are not implemented out of fear of copycatting claude code too much like not being able to use C-n/C-p to move up/down a list... why do i have to use my arrow keys???
English
0
0
1
133
paul lodder
paul lodder@paul__lodder·
what a time to build
English
0
0
0
53
paul lodder
paul lodder@paul__lodder·
@rickasaurus good point and i think it's context dependent with adapters for instance, removing the right degrees of freedom and defining the protocol that each adapter needs to adhere to feels crucial to scale that to any reasonable number
English
0
0
0
14
Rick
Rick@rickasaurus·
@paul__lodder Show me the math, whats your estimate for cost savings vs effort.
English
1
0
0
56
Rick
Rick@rickasaurus·
It’s dawning on me that the era of building generic systems is over. Code is cheap now, build hyper specific systems quickly. Reuse matters much less.
English
8
0
14
704
paul lodder
paul lodder@paul__lodder·
a $5 Hetzner box + systemctl restart is 90% of what most startups need
English
0
0
0
98
paul lodder
paul lodder@paul__lodder·
cloud is a psy-op making you believe bare metal is rocket science
English
1
0
0
125
paul lodder
paul lodder@paul__lodder·
@DSPyOSS but why would the data be too underspecified for this but not for prompt optimization? cause the problem representation space is so much bigger?
English
0
0
0
19
DSPy
DSPy@DSPyOSS·
@paul__lodder sure but what signal is the optimizer going to use? the data? the data is noisy and underspecified (and small) there's no replacement for programming well, that is, giving the most precise name to your fields
English
2
0
0
42
paul lodder
paul lodder@paul__lodder·
working with @DSPyOSS shifts the focus from prompt engineering to problem representation LLMs are semantic machines. even changing a field name can materially affect results because you're changing the semantic relationship the model sees
English
2
0
1
394
paul lodder
paul lodder@paul__lodder·
@DSPyOSS yeah so it's a bit of a stretch it feels akin to differentiable neural network architectures in the is_spam example, the ultimate output is is_spam, but there are many paths to get there
English
0
0
0
12
paul lodder
paul lodder@paul__lodder·
@DSPyOSS exactly. i guess you'd almost want a meta-optimizer that searches across different problem representations and post-processing configurations to find what optimizes for your end goal
English
1
0
0
33
DSPy
DSPy@DSPyOSS·
@paul__lodder yes, hence "programming---not prompting---language models". it's not so much "even changing a field name can matter". it's *designing field names* is the most pivotal part of the work; it's the bulk of the task definition. that's why even short signatures like `a, b -> c` work.
English
1
0
3
321
paul lodder
paul lodder@paul__lodder·
@DSPyOSS for instance, this pattern has made a big impact for me: instead of is_spam: bool classify as promotional/phishing/personal/automated then return category in [promotional, phishing] gives the model richer semantic structure even though you only need the binary result
English
0
0
0
70
paul lodder
paul lodder@paul__lodder·
@DSPyOSS this tripped me up initially. dspy resembles classical ml approaches, so i assumed labeled data was enough. but problem representation still matters enormously
English
1
0
0
90